_Final_Submission_.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "**Final Submission**",
      "provenance": [],
      "collapsed_sections": [
        "PlNV4QdBJ8zn",
        "m4Ile6DJW53u",
        "Ul0PkiRlXB33",
        "7z-XFaJeHAkO",
        "SYYJYMUt5sCE",
        "bg68AMvDA1mi",
        "6T_bhYg8zKmP",
        "AaWUyMnNzQOI",
        "P3WG3IstB2Zp",
        "01LHDnvsrySt"
      ],
      "toc_visible": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "widgets": {
      "application/vnd.jupyter.widget-state+json": {
        "2b371eb8229c4a6cbd1d5997e872a498": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_7bfe1110aad24de996256edd240e2b2d",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_626a6cac52b149d9888cae64ba29bff6",
              "IPY_MODEL_b1f109e85e624f1a89dd1c5626d36d20"
            ]
          }
        },
        "7bfe1110aad24de996256edd240e2b2d": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "626a6cac52b149d9888cae64ba29bff6": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_badf9b4e568f4791977893f80e253135",
            "_dom_classes": [],
            "description": "Downloading: 100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 231508,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 231508,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_5ea88c352a404d82bbce7a6224430e3c"
          }
        },
        "b1f109e85e624f1a89dd1c5626d36d20": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_a3daa289ede04498a16a2d369b6cf4f3",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 232k/232k [00:01&lt;00:00, 154kB/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_05967b9481594b93901efdb0ff5ad809"
          }
        },
        "badf9b4e568f4791977893f80e253135": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "5ea88c352a404d82bbce7a6224430e3c": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "a3daa289ede04498a16a2d369b6cf4f3": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "05967b9481594b93901efdb0ff5ad809": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "7d3c5337e8834ef783858acdf85fe56a": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_bb5e7f0bbab44e87bd62517e37bb47ae",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_373688c770da4ec1ab63bc8f80733a7c",
              "IPY_MODEL_c2867adc2e254b62960d91ef9bc4b36f"
            ]
          }
        },
        "bb5e7f0bbab44e87bd62517e37bb47ae": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "373688c770da4ec1ab63bc8f80733a7c": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_0daab2bffc9b42b38de836a13986231a",
            "_dom_classes": [],
            "description": "Downloading: 100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 28,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 28,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_b32bbcc602964a59826b73ad3f6cb8a0"
          }
        },
        "c2867adc2e254b62960d91ef9bc4b36f": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_0423c6395eaa4d6aa9a4cc6328b4e571",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 28.0/28.0 [00:01&lt;00:00, 26.3B/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_38b53480950048d28c5ed3640e84d455"
          }
        },
        "0daab2bffc9b42b38de836a13986231a": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "b32bbcc602964a59826b73ad3f6cb8a0": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "0423c6395eaa4d6aa9a4cc6328b4e571": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "38b53480950048d28c5ed3640e84d455": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "19cfc37cea5c4238822a64cd141371c0": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_28c4636aff0747a2a93230d87f3db14c",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_064f00e64d334c1b845425f3cac6fc52",
              "IPY_MODEL_108b456a34b141a787dc0f705d380594"
            ]
          }
        },
        "28c4636aff0747a2a93230d87f3db14c": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "064f00e64d334c1b845425f3cac6fc52": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_600ce8dc01f7467f9f7843c3a23a6d5d",
            "_dom_classes": [],
            "description": "Downloading: 100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 466062,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 466062,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_7c4ac4eb30fb41388a234ddbf9a9a160"
          }
        },
        "108b456a34b141a787dc0f705d380594": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_d417b3c5e34b43e19f8316813324e620",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 466k/466k [00:00&lt;00:00, 1.00MB/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_ddba104147f947009106ad3158e4d459"
          }
        },
        "600ce8dc01f7467f9f7843c3a23a6d5d": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "7c4ac4eb30fb41388a234ddbf9a9a160": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "d417b3c5e34b43e19f8316813324e620": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "ddba104147f947009106ad3158e4d459": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "4eae936a95dd4715b15c7a739cda0954": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_079cf310324e4c3aa6df3d26d50343c3",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_00ca1d6fa93e472d836354c91d962442",
              "IPY_MODEL_ed7cf1d2e2e74502b92d97e58885e6d5"
            ]
          }
        },
        "079cf310324e4c3aa6df3d26d50343c3": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "00ca1d6fa93e472d836354c91d962442": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_32ebf0bef3d349a2a32c85169b228278",
            "_dom_classes": [],
            "description": "Downloading: 100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 570,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 570,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_133bf8f03f054b9cb77fc845d731250b"
          }
        },
        "ed7cf1d2e2e74502b92d97e58885e6d5": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_ba14ebc0ea864fe29d50e8c85d213e2e",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 570/570 [00:14&lt;00:00, 39.1B/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_4b8c05cf9508421ba7ad89afbb208a91"
          }
        },
        "32ebf0bef3d349a2a32c85169b228278": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "133bf8f03f054b9cb77fc845d731250b": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "ba14ebc0ea864fe29d50e8c85d213e2e": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "4b8c05cf9508421ba7ad89afbb208a91": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "866512ec005e4bc9b028d3c8a78d9178": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_82e5ee9998a34ad3b52f4a46bd01cd62",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_f6f8b0bcf5e249438df182fd45c90234",
              "IPY_MODEL_020732fd790c43c4925f0e63635564d6"
            ]
          }
        },
        "82e5ee9998a34ad3b52f4a46bd01cd62": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "f6f8b0bcf5e249438df182fd45c90234": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_f4037784f4e94c1c80212271619d8658",
            "_dom_classes": [],
            "description": "Downloading: 100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 440473133,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 440473133,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_377461a288454f1eafa8a826a9ecc30f"
          }
        },
        "020732fd790c43c4925f0e63635564d6": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_0fb14f0864634ba6bc0173d0201014ad",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 440M/440M [00:13&lt;00:00, 32.3MB/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_5ce1cbbd966a40aea625bf01b687513d"
          }
        },
        "f4037784f4e94c1c80212271619d8658": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "377461a288454f1eafa8a826a9ecc30f": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "0fb14f0864634ba6bc0173d0201014ad": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "5ce1cbbd966a40aea625bf01b687513d": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        }
      }
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PlNV4QdBJ8zn"
      },
      "source": [
        "# 0. Setup"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "2rn16kHKU6jt",
        "outputId": "33f41bb2-a229-4e1a-ff16-e703bc7a6203"
      },
      "source": [
        "!pip install transformers"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting transformers\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/d8/b2/57495b5309f09fa501866e225c84532d1fd89536ea62406b2181933fb418/transformers-4.5.1-py3-none-any.whl (2.1MB)\n",
            "\u001b[K     |████████████████████████████████| 2.1MB 16.1MB/s \n",
            "\u001b[?25hCollecting tokenizers<0.11,>=0.10.1\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/ae/04/5b870f26a858552025a62f1649c20d29d2672c02ff3c3fb4c688ca46467a/tokenizers-0.10.2-cp37-cp37m-manylinux2010_x86_64.whl (3.3MB)\n",
            "\u001b[K     |████████████████████████████████| 3.3MB 56.4MB/s \n",
            "\u001b[?25hRequirement already satisfied: requests in /usr/local/lib/python3.7/dist-packages (from transformers) (2.23.0)\n",
            "Requirement already satisfied: filelock in /usr/local/lib/python3.7/dist-packages (from transformers) (3.0.12)\n",
            "Requirement already satisfied: tqdm>=4.27 in /usr/local/lib/python3.7/dist-packages (from transformers) (4.41.1)\n",
            "Requirement already satisfied: importlib-metadata; python_version < \"3.8\" in /usr/local/lib/python3.7/dist-packages (from transformers) (3.10.1)\n",
            "Requirement already satisfied: regex!=2019.12.17 in /usr/local/lib/python3.7/dist-packages (from transformers) (2019.12.20)\n",
            "Collecting sacremoses\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/75/ee/67241dc87f266093c533a2d4d3d69438e57d7a90abb216fa076e7d475d4a/sacremoses-0.0.45-py3-none-any.whl (895kB)\n",
            "\u001b[K     |████████████████████████████████| 901kB 53.8MB/s \n",
            "\u001b[?25hRequirement already satisfied: packaging in /usr/local/lib/python3.7/dist-packages (from transformers) (20.9)\n",
            "Requirement already satisfied: numpy>=1.17 in /usr/local/lib/python3.7/dist-packages (from transformers) (1.19.5)\n",
            "Requirement already satisfied: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in /usr/local/lib/python3.7/dist-packages (from requests->transformers) (1.24.3)\n",
            "Requirement already satisfied: idna<3,>=2.5 in /usr/local/lib/python3.7/dist-packages (from requests->transformers) (2.10)\n",
            "Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.7/dist-packages (from requests->transformers) (2020.12.5)\n",
            "Requirement already satisfied: chardet<4,>=3.0.2 in /usr/local/lib/python3.7/dist-packages (from requests->transformers) (3.0.4)\n",
            "Requirement already satisfied: zipp>=0.5 in /usr/local/lib/python3.7/dist-packages (from importlib-metadata; python_version < \"3.8\"->transformers) (3.4.1)\n",
            "Requirement already satisfied: typing-extensions>=3.6.4; python_version < \"3.8\" in /usr/local/lib/python3.7/dist-packages (from importlib-metadata; python_version < \"3.8\"->transformers) (3.7.4.3)\n",
            "Requirement already satisfied: six in /usr/local/lib/python3.7/dist-packages (from sacremoses->transformers) (1.15.0)\n",
            "Requirement already satisfied: click in /usr/local/lib/python3.7/dist-packages (from sacremoses->transformers) (7.1.2)\n",
            "Requirement already satisfied: joblib in /usr/local/lib/python3.7/dist-packages (from sacremoses->transformers) (1.0.1)\n",
            "Requirement already satisfied: pyparsing>=2.0.2 in /usr/local/lib/python3.7/dist-packages (from packaging->transformers) (2.4.7)\n",
            "Installing collected packages: tokenizers, sacremoses, transformers\n",
            "Successfully installed sacremoses-0.0.45 tokenizers-0.10.2 transformers-4.5.1\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TKKctamET7kV"
      },
      "source": [
        "import pandas as pd\n",
        "import numpy as np\n",
        "import torch\n",
        "import torch.nn as nn\n",
        "from torch.autograd import Variable\n",
        "import torch.nn.functional as functional\n",
        "import matplotlib.pyplot as plt\n",
        "from transformers import BertForSequenceClassification, AdamW, BertConfig\n",
        "import gc\n",
        "from transformers import BertModel\n",
        "from sklearn.metrics import roc_auc_score,f1_score\n",
        "import time\n",
        "import datetime"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "m4Ile6DJW53u"
      },
      "source": [
        "# 1. Load the dataset, set up the GPU, install the transformer"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "TvtocKlhUgp0",
        "outputId": "4ad612ed-bec6-4af0-db46-360432c537f4"
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Mounted at /content/drive\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "g4FdB1zNVOl_",
        "outputId": "4917a657-bfd4-45e3-de8e-c64a43306192"
      },
      "source": [
        "torch.manual_seed(0)\n",
        "use_cuda = torch.cuda.is_available()\n",
        "device = torch.device('cuda' if use_cuda else 'cpu')\n",
        "if use_cuda:\n",
        "    torch.cuda.manual_seed(0)\n",
        "    \n",
        "print(\"Using GPU: {}\".format(use_cuda))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Using GPU: True\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Z5qBfBTIUhKi"
      },
      "source": [
        "train = pd.read_csv('/content/drive/Shareddrives/CS682 Project/Constraint_Train.csv')\n",
        "val = pd.read_csv('/content/drive/Shareddrives/CS682 Project/Constraint_Val.csv')\n",
        "train[\"label\"] = train[\"label\"].map({\"real\": 1, \"fake\": 0})\n",
        "val[\"label\"] = val[\"label\"].map({\"real\": 1, \"fake\": 0})"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 399
        },
        "id": "QheKql0rU4KW",
        "outputId": "207b7763-2fe6-4255-af3c-e5ed1ac7e723"
      },
      "source": [
        "train"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>id</th>\n",
              "      <th>tweet</th>\n",
              "      <th>label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>1</td>\n",
              "      <td>The CDC currently reports 99031 deaths. In gen...</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>2</td>\n",
              "      <td>States reported 1121 deaths a small rise from ...</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>3</td>\n",
              "      <td>Politically Correct Woman (Almost) Uses Pandem...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>4</td>\n",
              "      <td>#IndiaFightsCorona: We have 1524 #COVID testin...</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>5</td>\n",
              "      <td>Populous states can generate large case counts...</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6415</th>\n",
              "      <td>6416</td>\n",
              "      <td>A tiger tested positive for COVID-19 please st...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6416</th>\n",
              "      <td>6417</td>\n",
              "      <td>???Autopsies prove that COVID-19 is??� a blood...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6417</th>\n",
              "      <td>6418</td>\n",
              "      <td>_A post claims a COVID-19 vaccine has already ...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6418</th>\n",
              "      <td>6419</td>\n",
              "      <td>Aamir Khan Donate 250 Cr. In PM Relief Cares Fund</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6419</th>\n",
              "      <td>6420</td>\n",
              "      <td>It has been 93 days since the last case of COV...</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>6420 rows × 3 columns</p>\n",
              "</div>"
            ],
            "text/plain": [
              "        id                                              tweet  label\n",
              "0        1  The CDC currently reports 99031 deaths. In gen...      1\n",
              "1        2  States reported 1121 deaths a small rise from ...      1\n",
              "2        3  Politically Correct Woman (Almost) Uses Pandem...      0\n",
              "3        4  #IndiaFightsCorona: We have 1524 #COVID testin...      1\n",
              "4        5  Populous states can generate large case counts...      1\n",
              "...    ...                                                ...    ...\n",
              "6415  6416  A tiger tested positive for COVID-19 please st...      0\n",
              "6416  6417  ???Autopsies prove that COVID-19 is??� a blood...      0\n",
              "6417  6418  _A post claims a COVID-19 vaccine has already ...      0\n",
              "6418  6419  Aamir Khan Donate 250 Cr. In PM Relief Cares Fund      0\n",
              "6419  6420  It has been 93 days since the last case of COV...      1\n",
              "\n",
              "[6420 rows x 3 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 399
        },
        "id": "TOQxiuEzVmLJ",
        "outputId": "543bfc29-8e1f-40f4-f612-33fd38b13382"
      },
      "source": [
        "val"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>id</th>\n",
              "      <th>tweet</th>\n",
              "      <th>label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>1</td>\n",
              "      <td>Chinese converting to Islam after realising th...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>2</td>\n",
              "      <td>11 out of 13 people (from the Diamond Princess...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>3</td>\n",
              "      <td>COVID-19 Is Caused By A Bacterium, Not Virus A...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>4</td>\n",
              "      <td>Mike Pence in RNC speech praises Donald Trump’...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>5</td>\n",
              "      <td>6/10 Sky's @EdConwaySky explains the latest #C...</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2135</th>\n",
              "      <td>2136</td>\n",
              "      <td>Donald Trump wrongly claimed that New Zealand ...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2136</th>\n",
              "      <td>2137</td>\n",
              "      <td>Current understanding is #COVID19 spreads most...</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2137</th>\n",
              "      <td>2138</td>\n",
              "      <td>Nothing screams “I am sat around doing fuck al...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2138</th>\n",
              "      <td>2139</td>\n",
              "      <td>Birx says COVID-19 outbreak not under control ...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2139</th>\n",
              "      <td>2140</td>\n",
              "      <td>Another 4422 new coronavirus cases have been c...</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>2140 rows × 3 columns</p>\n",
              "</div>"
            ],
            "text/plain": [
              "        id                                              tweet  label\n",
              "0        1  Chinese converting to Islam after realising th...      0\n",
              "1        2  11 out of 13 people (from the Diamond Princess...      0\n",
              "2        3  COVID-19 Is Caused By A Bacterium, Not Virus A...      0\n",
              "3        4  Mike Pence in RNC speech praises Donald Trump’...      0\n",
              "4        5  6/10 Sky's @EdConwaySky explains the latest #C...      1\n",
              "...    ...                                                ...    ...\n",
              "2135  2136  Donald Trump wrongly claimed that New Zealand ...      0\n",
              "2136  2137  Current understanding is #COVID19 spreads most...      1\n",
              "2137  2138  Nothing screams “I am sat around doing fuck al...      0\n",
              "2138  2139  Birx says COVID-19 outbreak not under control ...      0\n",
              "2139  2140  Another 4422 new coronavirus cases have been c...      1\n",
              "\n",
              "[2140 rows x 3 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 8
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 399
        },
        "id": "Ji3I9nRYFEhA",
        "outputId": "fb8ecab4-125b-4c72-b115-ec54d0930565"
      },
      "source": [
        "data = pd.concat([train, val], axis=0, ignore_index=True).drop([\"id\"], axis=1)\n",
        "data"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>tweet</th>\n",
              "      <th>label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>The CDC currently reports 99031 deaths. In gen...</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>States reported 1121 deaths a small rise from ...</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>Politically Correct Woman (Almost) Uses Pandem...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>#IndiaFightsCorona: We have 1524 #COVID testin...</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>Populous states can generate large case counts...</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>8555</th>\n",
              "      <td>Donald Trump wrongly claimed that New Zealand ...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>8556</th>\n",
              "      <td>Current understanding is #COVID19 spreads most...</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>8557</th>\n",
              "      <td>Nothing screams “I am sat around doing fuck al...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>8558</th>\n",
              "      <td>Birx says COVID-19 outbreak not under control ...</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>8559</th>\n",
              "      <td>Another 4422 new coronavirus cases have been c...</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>8560 rows × 2 columns</p>\n",
              "</div>"
            ],
            "text/plain": [
              "                                                  tweet  label\n",
              "0     The CDC currently reports 99031 deaths. In gen...      1\n",
              "1     States reported 1121 deaths a small rise from ...      1\n",
              "2     Politically Correct Woman (Almost) Uses Pandem...      0\n",
              "3     #IndiaFightsCorona: We have 1524 #COVID testin...      1\n",
              "4     Populous states can generate large case counts...      1\n",
              "...                                                 ...    ...\n",
              "8555  Donald Trump wrongly claimed that New Zealand ...      0\n",
              "8556  Current understanding is #COVID19 spreads most...      1\n",
              "8557  Nothing screams “I am sat around doing fuck al...      0\n",
              "8558  Birx says COVID-19 outbreak not under control ...      0\n",
              "8559  Another 4422 new coronavirus cases have been c...      1\n",
              "\n",
              "[8560 rows x 2 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ul0PkiRlXB33"
      },
      "source": [
        "# 2. Begin Tokenize the data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "referenced_widgets": [
            "2b371eb8229c4a6cbd1d5997e872a498",
            "7bfe1110aad24de996256edd240e2b2d",
            "626a6cac52b149d9888cae64ba29bff6",
            "b1f109e85e624f1a89dd1c5626d36d20",
            "badf9b4e568f4791977893f80e253135",
            "5ea88c352a404d82bbce7a6224430e3c",
            "a3daa289ede04498a16a2d369b6cf4f3",
            "05967b9481594b93901efdb0ff5ad809",
            "7d3c5337e8834ef783858acdf85fe56a",
            "bb5e7f0bbab44e87bd62517e37bb47ae",
            "373688c770da4ec1ab63bc8f80733a7c",
            "c2867adc2e254b62960d91ef9bc4b36f",
            "0daab2bffc9b42b38de836a13986231a",
            "b32bbcc602964a59826b73ad3f6cb8a0",
            "0423c6395eaa4d6aa9a4cc6328b4e571",
            "38b53480950048d28c5ed3640e84d455",
            "19cfc37cea5c4238822a64cd141371c0",
            "28c4636aff0747a2a93230d87f3db14c",
            "064f00e64d334c1b845425f3cac6fc52",
            "108b456a34b141a787dc0f705d380594",
            "600ce8dc01f7467f9f7843c3a23a6d5d",
            "7c4ac4eb30fb41388a234ddbf9a9a160",
            "d417b3c5e34b43e19f8316813324e620",
            "ddba104147f947009106ad3158e4d459"
          ]
        },
        "id": "_GKlXaeWVxht",
        "outputId": "2dc67684-0513-41ce-b5b7-834bd3e9faec"
      },
      "source": [
        "from transformers import BertTokenizer\n",
        "print('Loading BERT tokenizer...')\n",
        "tokenizer = BertTokenizer.from_pretrained('bert-base-uncased', do_lower_case=True)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Loading BERT tokenizer...\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "2b371eb8229c4a6cbd1d5997e872a498",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, description='Downloading', max=231508.0, style=ProgressStyle(descripti…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "7d3c5337e8834ef783858acdf85fe56a",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, description='Downloading', max=28.0, style=ProgressStyle(description_w…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "19cfc37cea5c4238822a64cd141371c0",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, description='Downloading', max=466062.0, style=ProgressStyle(descripti…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6DaToN5PWUag"
      },
      "source": [
        "tweets = data.tweet.values\n",
        "labels = data.label.values"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "1I6Dj3Ik_q1V",
        "outputId": "188dc471-0ebf-4bc1-dfc7-37e5d26895c9"
      },
      "source": [
        "tweets"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array(['The CDC currently reports 99031 deaths. In general the discrepancies in death counts between different sources are small and explicable. The death toll stands at roughly 100000 people today.',\n",
              "       'States reported 1121 deaths a small rise from last Tuesday. Southern states reported 640 of those deaths. https://t.co/YASGRTT4ux',\n",
              "       'Politically Correct Woman (Almost) Uses Pandemic as Excuse Not to Reuse Plastic Bag https://t.co/thF8GuNFPe #coronavirus #nashville',\n",
              "       ...,\n",
              "       'Nothing screams “I am sat around doing fuck all during lockdown” quite like confident assumption that other people are sat around doing fuck all during lockdown https://t.co/Hs3rZAAl1h',\n",
              "       'Birx says COVID-19 outbreak not under control because ‘people are on the move’ https://t.co/YFqi6f7Rvj',\n",
              "       \"Another 4422 new coronavirus cases have been confirmed in the UK - the highest daily number since 8 May. It's up from 4322 new cases reported on Friday and the overall total nationwide now stands at 385936. Read the latest here: https://t.co/1TYdoRZ9Yr https://t.co/96F7njc7sM\"],\n",
              "      dtype=object)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 12
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "_r3VwxkewLVk",
        "outputId": "018bdbc2-bb42-41e3-a77c-095b7305e15f"
      },
      "source": [
        "import re\n",
        "from string import punctuation\n",
        "def preprocess(data):\n",
        "    #remove url and hashtag\n",
        "    for i in range(data.shape[0]):\n",
        "        text=data[i].lower()\n",
        "        text1=''.join([word+\" \" for word in text.split()])\n",
        "        data[i]=text1\n",
        "    giant_url_regex = ('http[s]?://(?:[a-zA-Z]|[0-9]|[$-_@.&+]|'\n",
        "        '[!*\\(\\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+')\n",
        "    mention_regex = '@[\\w\\-]+'\n",
        "    hashtag_regex = '#[\\w\\-]+'\n",
        "    space_pattern = '\\s+'\n",
        "\n",
        "    for i in range(data.shape[0]):\n",
        "        text_string = data[i]\n",
        "        parsed_text = re.sub(hashtag_regex, '', text_string)\n",
        "        parsed_text = re.sub(giant_url_regex, '', parsed_text)\n",
        "        parsed_text = re.sub(mention_regex, '', parsed_text) \n",
        "        #remove punctuation\n",
        "        parsed_text = re.sub(r\"[{}]+\".format(punctuation), '', parsed_text) \n",
        "        parsed_text = re.sub(space_pattern, ' ', parsed_text)\n",
        "        data[i] = parsed_text\n",
        "    return data\n",
        "tweets = preprocess(tweets)\n",
        "print(tweets)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "['the cdc currently reports 99031 deaths in general the discrepancies in death counts between different sources are small and explicable the death toll stands at roughly 100000 people today '\n",
            " 'states reported 1121 deaths a small rise from last tuesday southern states reported 640 of those deaths '\n",
            " 'politically correct woman almost uses pandemic as excuse not to reuse plastic bag '\n",
            " ...\n",
            " 'nothing screams “i am sat around doing fuck all during lockdown” quite like confident assumption that other people are sat around doing fuck all during lockdown '\n",
            " 'birx says covid19 outbreak not under control because ‘people are on the move’ '\n",
            " 'another 4422 new coronavirus cases have been confirmed in the uk the highest daily number since 8 may its up from 4322 new cases reported on friday and the overall total nationwide now stands at 385936 read the latest here ']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "U2oBkEwCWlSJ",
        "outputId": "09c9b246-57b6-4c3d-97a2-ae69fbcf3bf6"
      },
      "source": [
        "# Print the original sentence.\n",
        "print(' Original: ', tweets[0])\n",
        "\n",
        "# Print the sentence split into tokens.\n",
        "print('Tokenized: ', tokenizer.tokenize(tweets[0]))\n",
        "\n",
        "# Print the sentence mapped to token ids.\n",
        "print('Token IDs: ', tokenizer.convert_tokens_to_ids(tokenizer.tokenize(tweets[0])))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            " Original:  the cdc currently reports 99031 deaths in general the discrepancies in death counts between different sources are small and explicable the death toll stands at roughly 100000 people today \n",
            "Tokenized:  ['the', 'cdc', 'currently', 'reports', '99', '##0', '##31', 'deaths', 'in', 'general', 'the', 'disc', '##re', '##pan', '##cies', 'in', 'death', 'counts', 'between', 'different', 'sources', 'are', 'small', 'and', 'ex', '##pl', '##ica', '##ble', 'the', 'death', 'toll', 'stands', 'at', 'roughly', '1000', '##00', 'people', 'today']\n",
            "Token IDs:  [1996, 26629, 2747, 4311, 5585, 2692, 21486, 6677, 1999, 2236, 1996, 5860, 2890, 9739, 9243, 1999, 2331, 9294, 2090, 2367, 4216, 2024, 2235, 1998, 4654, 24759, 5555, 3468, 1996, 2331, 9565, 4832, 2012, 5560, 6694, 8889, 2111, 2651]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "GJ42kHldWoCN",
        "outputId": "d6287420-c299-40ee-b61b-e83d4ae5877b"
      },
      "source": [
        "max_len = 0\n",
        "ind = [100,200,300,400,500,512]\n",
        "for i in ind:\n",
        "  count = 0\n",
        "  for tweet in tweets:\n",
        "      max_len = max(max_len, len(tweet))\n",
        "      if len(tweet)>i:\n",
        "        count+=1\n",
        "  print(\"Count of sentence length over {} is: \".format(i), count)\n",
        "print('Max sentence length: ', max_len)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Count of sentence length over 100 is:  5667\n",
            "Count of sentence length over 200 is:  2327\n",
            "Count of sentence length over 300 is:  59\n",
            "Count of sentence length over 400 is:  20\n",
            "Count of sentence length over 500 is:  13\n",
            "Count of sentence length over 512 is:  10\n",
            "Max sentence length:  8672\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fR-xM4fuXR6F",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "5bf6de3b-82f7-4462-a514-8ef866f49e00"
      },
      "source": [
        "input_ids = []\n",
        "attention_masks = []\n",
        "for tweet in tweets:\n",
        "    encoded_dict = tokenizer.encode_plus(\n",
        "                        tweet,                      # Sentence to encode.\n",
        "                        add_special_tokens = True, # Add '[CLS]' and '[SEP]'\n",
        "                        max_length = 512,           # Pad & truncate all sentences.\n",
        "                        pad_to_max_length = True,\n",
        "                        return_attention_mask = True,   # Construct attn. masks.\n",
        "                        return_tensors = 'pt',     # Return pytorch tensors.\n",
        "                   )\n",
        "    \n",
        "    # Add the encoded sentence to the list.    \n",
        "    input_ids.append(encoded_dict['input_ids'])\n",
        "    \n",
        "    # And its attention mask (simply differentiates padding from non-padding).\n",
        "    attention_masks.append(encoded_dict['attention_mask'])\n",
        "# Convert the lists into tensors.\n",
        "input_ids = torch.cat(input_ids, dim=0)\n",
        "attention_masks = torch.cat(attention_masks, dim=0)\n",
        "labels = torch.tensor(labels)\n",
        "\n",
        "# Print sentence 0, now as a list of IDs.\n",
        "print('Original: ', tweets[0])\n",
        "print('Token IDs:', input_ids[0])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Truncation was not explicitly activated but `max_length` is provided a specific value, please use `truncation=True` to explicitly truncate examples to max length. Defaulting to 'longest_first' truncation strategy. If you encode pairs of sequences (GLUE-style) with the tokenizer you can select this strategy more precisely by providing a specific strategy to `truncation`.\n",
            "/usr/local/lib/python3.7/dist-packages/transformers/tokenization_utils_base.py:2079: FutureWarning: The `pad_to_max_length` argument is deprecated and will be removed in a future version, use `padding=True` or `padding='longest'` to pad to the longest sequence in the batch, or use `padding='max_length'` to pad to a max length. In this case, you can give a specific length with `max_length` (e.g. `max_length=45`) or leave max_length to None to pad to the maximal input size of the model (e.g. 512 for Bert).\n",
            "  FutureWarning,\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Original:  the cdc currently reports 99031 deaths in general the discrepancies in death counts between different sources are small and explicable the death toll stands at roughly 100000 people today \n",
            "Token IDs: tensor([  101,  1996, 26629,  2747,  4311,  5585,  2692, 21486,  6677,  1999,\n",
            "         2236,  1996,  5860,  2890,  9739,  9243,  1999,  2331,  9294,  2090,\n",
            "         2367,  4216,  2024,  2235,  1998,  4654, 24759,  5555,  3468,  1996,\n",
            "         2331,  9565,  4832,  2012,  5560,  6694,  8889,  2111,  2651,   102,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0])\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7z-XFaJeHAkO"
      },
      "source": [
        "#3 Train and test Procedure"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NYg27RYqs0mX"
      },
      "source": [
        "Approach #1"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "tN27Pm6tA1GQ",
        "outputId": "0088e7f4-b919-4736-b809-a2c1adaa360f"
      },
      "source": [
        "from torch.utils.data import TensorDataset, random_split\n",
        "\n",
        "# Combine the training inputs into a TensorDataset.\n",
        "dataset = TensorDataset(input_ids, attention_masks, labels)\n",
        "\n",
        "# Create a 90-10 train-validation split.\n",
        "train_size = int(0.9 * len(dataset))\n",
        "val_size = len(dataset) - train_size\n",
        "\n",
        "train_dataset, val_dataset = random_split(dataset, [train_size, val_size],generator=torch.Generator().manual_seed(42))\n",
        "\n",
        "print('{:>5,} training samples'.format(train_size))\n",
        "print('{:>5,} validation samples'.format(val_size))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "7,704 training samples\n",
            "  856 validation samples\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fpdrQqGLGrzW"
      },
      "source": [
        "from torch.utils.data import DataLoader, RandomSampler, SequentialSampler\n",
        "\n",
        "batch_size = 16\n",
        "\n",
        "train_dataloader = DataLoader(\n",
        "            train_dataset,  \n",
        "            shuffle = True,\n",
        "            batch_size = batch_size \n",
        "        )\n",
        "\n",
        "validation_dataloader = DataLoader(\n",
        "            val_dataset,\n",
        "            shuffle = False,\n",
        "            batch_size = batch_size \n",
        "        )"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oTpHMNz1nX7g"
      },
      "source": [
        "def format_time(elapsed):\n",
        "    elapsed_rounded = int(round((elapsed)))\n",
        "    return str(datetime.timedelta(seconds=elapsed_rounded))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "E8EEOXKpEL8I"
      },
      "source": [
        "# 3.1 BERT Fine Tune\n",
        "best acc: 0.9579439252336449"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000,
          "referenced_widgets": [
            "4eae936a95dd4715b15c7a739cda0954",
            "079cf310324e4c3aa6df3d26d50343c3",
            "00ca1d6fa93e472d836354c91d962442",
            "ed7cf1d2e2e74502b92d97e58885e6d5",
            "32ebf0bef3d349a2a32c85169b228278",
            "133bf8f03f054b9cb77fc845d731250b",
            "ba14ebc0ea864fe29d50e8c85d213e2e",
            "4b8c05cf9508421ba7ad89afbb208a91",
            "866512ec005e4bc9b028d3c8a78d9178",
            "82e5ee9998a34ad3b52f4a46bd01cd62",
            "f6f8b0bcf5e249438df182fd45c90234",
            "020732fd790c43c4925f0e63635564d6",
            "f4037784f4e94c1c80212271619d8658",
            "377461a288454f1eafa8a826a9ecc30f",
            "0fb14f0864634ba6bc0173d0201014ad",
            "5ce1cbbd966a40aea625bf01b687513d"
          ]
        },
        "id": "HQiIl1oWOVg0",
        "outputId": "4d2694f6-e67d-4bc1-83bb-ee8df456fa1a"
      },
      "source": [
        "from transformers import BertForSequenceClassification, AdamW, BertConfig\n",
        "model = BertForSequenceClassification.from_pretrained(\n",
        "    \"bert-base-uncased\",\n",
        "    num_labels = 2, \n",
        "    output_attentions = False, \n",
        "    output_hidden_states = False,\n",
        ")\n",
        "model.to(device)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "4eae936a95dd4715b15c7a739cda0954",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, description='Downloading', max=570.0, style=ProgressStyle(description_…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "866512ec005e4bc9b028d3c8a78d9178",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, description='Downloading', max=440473133.0, style=ProgressStyle(descri…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "Some weights of the model checkpoint at bert-base-uncased were not used when initializing BertForSequenceClassification: ['cls.predictions.bias', 'cls.predictions.transform.dense.weight', 'cls.predictions.transform.dense.bias', 'cls.predictions.decoder.weight', 'cls.seq_relationship.weight', 'cls.seq_relationship.bias', 'cls.predictions.transform.LayerNorm.weight', 'cls.predictions.transform.LayerNorm.bias']\n",
            "- This IS expected if you are initializing BertForSequenceClassification from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n",
            "- This IS NOT expected if you are initializing BertForSequenceClassification from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n",
            "Some weights of BertForSequenceClassification were not initialized from the model checkpoint at bert-base-uncased and are newly initialized: ['classifier.weight', 'classifier.bias']\n",
            "You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "BertForSequenceClassification(\n",
              "  (bert): BertModel(\n",
              "    (embeddings): BertEmbeddings(\n",
              "      (word_embeddings): Embedding(30522, 768, padding_idx=0)\n",
              "      (position_embeddings): Embedding(512, 768)\n",
              "      (token_type_embeddings): Embedding(2, 768)\n",
              "      (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "      (dropout): Dropout(p=0.1, inplace=False)\n",
              "    )\n",
              "    (encoder): BertEncoder(\n",
              "      (layer): ModuleList(\n",
              "        (0): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (1): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (2): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (3): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (4): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (5): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (6): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (7): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (8): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (9): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (10): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (11): BertLayer(\n",
              "          (attention): BertAttention(\n",
              "            (self): BertSelfAttention(\n",
              "              (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "            (output): BertSelfOutput(\n",
              "              (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "              (dropout): Dropout(p=0.1, inplace=False)\n",
              "            )\n",
              "          )\n",
              "          (intermediate): BertIntermediate(\n",
              "            (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "          )\n",
              "          (output): BertOutput(\n",
              "            (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "      )\n",
              "    )\n",
              "    (pooler): BertPooler(\n",
              "      (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "      (activation): Tanh()\n",
              "    )\n",
              "  )\n",
              "  (dropout): Dropout(p=0.1, inplace=False)\n",
              "  (classifier): Linear(in_features=768, out_features=2, bias=True)\n",
              ")"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 19
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FGuHHk81Ommo"
      },
      "source": [
        "optimizer = AdamW(model.parameters(),\n",
        "                  lr = 5e-5, \n",
        "                  eps = 1e-8 \n",
        "                )\n",
        "epochs = 4\n",
        "criterion = nn.CrossEntropyLoss()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "6BWCuZolQpp_",
        "outputId": "936f6ce1-c946-49ff-9f80-bf4a5e9825b5"
      },
      "source": [
        "import random\n",
        "import numpy as np\n",
        "\n",
        "seed_val = 42\n",
        "\n",
        "random.seed(seed_val)\n",
        "torch.manual_seed(seed_val)\n",
        "torch.cuda.manual_seed_all(seed_val)\n",
        "\n",
        "training_stats = []\n",
        "total_t0 = time.time()\n",
        "best_accuracy = 0\n",
        "for epoch_i in range(0, epochs):\n",
        "    #Training\n",
        "    print(\"\")\n",
        "    print('Epoch {:} / {:}'.format(epoch_i + 1, epochs))\n",
        "    print('Training...')\n",
        "\n",
        "    t0 = time.time()\n",
        "    total_train_loss = 0\n",
        "    total_train_accuracy = 0\n",
        "    model.train()\n",
        "    for step, batch in enumerate(train_dataloader):\n",
        "\n",
        "        input_ids = batch[0].to(device)\n",
        "        input_mask = batch[1].to(device)\n",
        "        labels = batch[2].to(device)\n",
        "\n",
        "        model.zero_grad()        \n",
        "        out = model(input_ids, token_type_ids=None, attention_mask=input_mask, labels=labels)\n",
        "        loss = out[0]\n",
        "        logits = out[1]\n",
        "  \n",
        "        total_train_loss += loss.item()\n",
        "        loss.backward()\n",
        "        torch.nn.utils.clip_grad_norm_(model.parameters(), 1.0)\n",
        "        optimizer.step()\n",
        "\n",
        "        pred = torch.argmax(logits, dim = 1)\n",
        "        total_train_accuracy +=  torch.sum(pred == labels).item()\n",
        "        \n",
        "    avg_train_accuracy = total_train_accuracy / len(train_dataloader.dataset)\n",
        "    avg_train_loss = total_train_loss / len(train_dataloader.dataset)            \n",
        "    print(\"  Accuracy: {}\".format(avg_train_accuracy))\n",
        "    print(\"  Training loss: {}\".format(avg_train_loss))\n",
        "        \n",
        "\n",
        "    # Validation\n",
        "    print(\"\")\n",
        "    print(\"Validation...\")\n",
        "    model.eval()\n",
        "    total_eval_accuracy = 0\n",
        "    total_eval_loss = 0\n",
        "    y_true = []\n",
        "    y_pred = []\n",
        "\n",
        "    for batch in validation_dataloader:\n",
        "        input_ids = batch[0].to(device)\n",
        "        input_mask = batch[1].to(device)\n",
        "        labels = batch[2].to(device)\n",
        "        \n",
        "        with torch.no_grad():        \n",
        "            out = model(input_ids, token_type_ids=None, attention_mask=input_mask,labels=labels)\n",
        "            loss = out[0]\n",
        "            logits = out[1]\n",
        "\n",
        "        total_eval_loss += loss.item()\n",
        "        pred = torch.argmax(logits, dim = 1)\n",
        "        total_eval_accuracy += torch.sum(pred == labels).item()\n",
        "        y_true.append(labels.flatten())\n",
        "        y_pred.append(pred.flatten())\n",
        "        \n",
        "    avg_val_accuracy = total_eval_accuracy / len(validation_dataloader.dataset)\n",
        "    print(\"  Accuracy: {}\".format(avg_val_accuracy))\n",
        "    avg_val_loss = total_eval_loss / len(validation_dataloader.dataset)\n",
        "    print(\"  Validation loss: {}\".format(avg_val_loss))\n",
        "    training_time = format_time(time.time() - t0)\n",
        "    print()\n",
        "    \n",
        "    y_true = torch.cat(y_true).tolist()\n",
        "    y_pred = torch.cat(y_pred).tolist()\n",
        "    print(\"This epoch took: {:}\".format(training_time))\n",
        "    print('roc_auc score: ', roc_auc_score(y_true,y_pred))\n",
        "    print('F1 score:',f1_score(y_true, y_pred))\n",
        "    print()\n",
        "\n",
        "    training_stats.append(\n",
        "        {\n",
        "            'epoch': epoch_i + 1,\n",
        "            'Train Accur.': avg_train_accuracy,\n",
        "            'Training Loss': avg_train_loss,\n",
        "            'Valid. Loss': avg_val_loss,\n",
        "            'Valid. Accur.': avg_val_accuracy,\n",
        "            'Training Time': training_time,\n",
        "        }\n",
        "    )\n",
        "    print()\n",
        "\n",
        "    if avg_val_accuracy > best_accuracy:\n",
        "        best_accuracy = avg_val_accuracy\n",
        "        best_model = model\n",
        "\n",
        "print()\n",
        "print(\"=\"*10)\n",
        "print(\"Summary\")\n",
        "print(\"Total time {:} (h:mm:ss)\".format(format_time(time.time()-total_t0)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "Epoch 1 / 4\n",
            "Training...\n",
            "  Accuracy: 0.9014797507788161\n",
            "  Training loss: 0.01634137871407305\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9427570093457944\n",
            "  Validation loss: 0.009837308370891227\n",
            "\n",
            "This epoch took: 0:07:15\n",
            "roc_auc score:  0.94400661687523\n",
            "F1 score: 0.9460946094609461\n",
            "\n",
            "\n",
            "\n",
            "Epoch 2 / 4\n",
            "Training...\n",
            "  Accuracy: 0.9680685358255452\n",
            "  Training loss: 0.007335280389116394\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9532710280373832\n",
            "  Validation loss: 0.011293363998558103\n",
            "\n",
            "This epoch took: 0:07:14\n",
            "roc_auc score:  0.9521870311443786\n",
            "F1 score: 0.9571734475374732\n",
            "\n",
            "\n",
            "\n",
            "Epoch 3 / 4\n",
            "Training...\n",
            "  Accuracy: 0.985202492211838\n",
            "  Training loss: 0.0036151895493783406\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9579439252336449\n",
            "  Validation loss: 0.013102064621128864\n",
            "\n",
            "This epoch took: 0:07:14\n",
            "roc_auc score:  0.9586225468374744\n",
            "F1 score: 0.9606986899563319\n",
            "\n",
            "\n",
            "\n",
            "Epoch 4 / 4\n",
            "Training...\n",
            "  Accuracy: 0.9932502596053998\n",
            "  Training loss: 0.0020287746871342684\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.955607476635514\n",
            "  Validation loss: 0.014002995049824642\n",
            "\n",
            "This epoch took: 0:07:13\n",
            "roc_auc score:  0.9562703685995197\n",
            "F1 score: 0.9585152838427947\n",
            "\n",
            "\n",
            "\n",
            "==========\n",
            "Summary\n",
            "Total time 0:28:55 (h:mm:ss)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "229MWxgyFeRZ"
      },
      "source": [
        "# PATH1 = \"/content/drive/MyDrive/Colab Notebooks/'bert_finetune.pt'\"\n",
        "# torch.save(model, PATH1)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "r09QHHadsjHi"
      },
      "source": [
        "# 3.2 BERT+CNN"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "M5PrL7fVJC2y"
      },
      "source": [
        "PATH1 = \"/content/drive/MyDrive/Colab Notebooks/'bert_finetune.pt'\"\n",
        "the_best_model = torch.load(PATH1,map_location='cpu')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SYYJYMUt5sCE"
      },
      "source": [
        "## **3.2.1 Without Freeze Parameters**\n",
        "best acc: 0.9439252336448598"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gTQArcpeXU3m"
      },
      "source": [
        "class BertCNNClassifier(nn.Module):\n",
        "    def __init__(self, tuned_model, embed_num = 512, embed_dim = 768, dropout=0.1, kernel_num=3, kernel_sizes=[1,2], num_labels=2):\n",
        "        super().__init__()\n",
        "        self.num_labels = num_labels\n",
        "        self.embed_num = embed_num\n",
        "        self.embed_dim = embed_dim\n",
        "        self.dropout = dropout\n",
        "        self.kernel_num = kernel_num\n",
        "        self.kernel_sizes = kernel_sizes\n",
        "        self.softmax = nn.functional.softmax\n",
        "\n",
        "        self.bert = tuned_model.bert\n",
        "        self.convs = nn.ModuleList([nn.Conv2d(1, self.kernel_num, (k, self.embed_dim)) for k in self.kernel_sizes])\n",
        "        self.dropout = nn.Dropout(self.dropout)\n",
        "        self.classifier = nn.Linear(len(self.kernel_sizes)*self.kernel_num, self.num_labels)\n",
        "        \n",
        "    def forward(self, input_ids, attention_mask, token_type_ids = None):\n",
        "        output = self.bert(input_ids = input_ids, attention_mask = attention_mask, token_type_ids = token_type_ids) #16,512,768\n",
        "        output = output[0].unsqueeze(1) #16,1,512,768\n",
        "        output = [nn.functional.relu(conv(output)).squeeze(3) for conv in self.convs] #16,3,508,1 => #16,3,508\n",
        "        output = [nn.functional.max_pool1d(i, i.size(2)).squeeze(2) for i in output] #=> 16,3\n",
        "        output = torch.cat(output, 1)\n",
        "        output = self.dropout(output)\n",
        "        logits = self.classifier(output)\n",
        "        return self.softmax(logits, 1)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KhAjGZ6FkUtN"
      },
      "source": [
        "# Initializing model\n",
        "model2 = BertCNNClassifier(the_best_model).cuda()\n",
        "# set parameters\n",
        "epochs = 4\n",
        "learning_rate = 5e-5\n",
        "optimizer = AdamW(model2.parameters(), lr = learning_rate)\n",
        "criterion = nn.CrossEntropyLoss()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "iUXcJCtq-heh",
        "outputId": "ad140d99-cbaf-4783-deea-f5d4bddeb285"
      },
      "source": [
        "import random\n",
        "import numpy as np\n",
        "\n",
        "seed_val = 42\n",
        "\n",
        "random.seed(seed_val)\n",
        "torch.manual_seed(seed_val)\n",
        "torch.cuda.manual_seed_all(seed_val)\n",
        "\n",
        "training_stats = []\n",
        "total_t0 = time.time()\n",
        "best_accuracy = 0\n",
        "for epoch_i in range(0, epochs):\n",
        "    #Training\n",
        "    print(\"\")\n",
        "    print('Epoch {:} / {:}'.format(epoch_i + 1, epochs))\n",
        "    print('Training...')\n",
        "\n",
        "    t0 = time.time()\n",
        "    total_train_loss = 0\n",
        "    total_train_accuracy = 0\n",
        "    model2.train()\n",
        "    for step, batch in enumerate(train_dataloader):\n",
        "\n",
        "        input_ids = batch[0].to(device)\n",
        "        input_mask = batch[1].to(device)\n",
        "        labels = batch[2].to(device)\n",
        "\n",
        "        model2.zero_grad()        \n",
        "        out = model2(input_ids = input_ids, attention_mask = input_mask, token_type_ids = None)\n",
        "        loss = criterion(out, labels)\n",
        "        total_train_loss += loss.item()\n",
        "\n",
        "        loss.backward()\n",
        "        torch.nn.utils.clip_grad_norm_(model2.parameters(), 1.0)\n",
        "        optimizer.step()\n",
        "\n",
        "        pred = torch.argmax(out, dim = 1)\n",
        "        total_train_accuracy +=  torch.sum(pred == labels).item()\n",
        "        \n",
        "    avg_train_accuracy = total_train_accuracy / len(train_dataloader.dataset)\n",
        "    avg_train_loss = total_train_loss / len(train_dataloader.dataset)            \n",
        "    print(\"  Accuracy: {}\".format(avg_train_accuracy))\n",
        "    print(\"  Training loss: {}\".format(avg_train_loss))\n",
        "        \n",
        "    # Validation\n",
        "    print(\"\")\n",
        "    print(\"Validation...\")\n",
        "    model2.eval()\n",
        "    total_eval_accuracy = 0\n",
        "    total_eval_loss = 0\n",
        "    y_true = []\n",
        "    y_pred = []\n",
        "\n",
        "    for batch in validation_dataloader:\n",
        "        input_ids = batch[0].to(device)\n",
        "        input_mask = batch[1].to(device)\n",
        "        labels = batch[2].to(device)\n",
        "        \n",
        "        with torch.no_grad():        \n",
        "            out = model2(input_ids = input_ids, attention_mask = input_mask, token_type_ids = None)\n",
        "        loss = criterion(out, labels)\n",
        "        total_eval_loss += loss.item()\n",
        "        pred = torch.argmax(out, dim = 1)\n",
        "        total_eval_accuracy += torch.sum(pred == labels).item()\n",
        "        y_true.append(labels.flatten())\n",
        "        y_pred.append(pred.flatten())\n",
        "        \n",
        "    avg_val_accuracy = total_eval_accuracy / len(validation_dataloader.dataset)\n",
        "    print(\"  Accuracy: {}\".format(avg_val_accuracy))\n",
        "    avg_val_loss = total_eval_loss / len(validation_dataloader.dataset)\n",
        "    print(\"  Validation loss: {}\".format(avg_val_loss))\n",
        "    training_time = format_time(time.time() - t0)\n",
        "    print(\"  This epoch took: {:}\".format(training_time))\n",
        "    print()\n",
        "    y_true = torch.cat(y_true).tolist()\n",
        "    y_pred = torch.cat(y_pred).tolist()\n",
        "    print('  roc_auc score: ', roc_auc_score(y_true,y_pred))\n",
        "    print('  F1 score:',f1_score(y_true, y_pred))\n",
        "\n",
        "\n",
        "    training_stats.append(\n",
        "        {\n",
        "            'epoch': epoch_i + 1,\n",
        "            'Train Accur.': avg_train_accuracy,\n",
        "            'Training Loss': avg_train_loss,\n",
        "            'Valid. Loss': avg_val_loss,\n",
        "            'Valid. Accur.': avg_val_accuracy,\n",
        "            'Training Time': training_time,\n",
        "        }\n",
        "    )\n",
        "\n",
        "    if avg_val_accuracy > best_accuracy:\n",
        "        best_accuracy = avg_val_accuracy\n",
        "        best_model = model2\n",
        "\n",
        "print(\"===\")\n",
        "print(\"Summary\")\n",
        "print(\"Total time {:} (h:mm:ss)\".format(format_time(time.time()-total_t0)))\n",
        "print('best acc:',best_accuracy)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "Epoch 1 / 4\n",
            "Training...\n",
            "  Accuracy: 0.9749480789200415\n",
            "  Training loss: 0.02163551350898832\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9392523364485982\n",
            "  Validation loss: 0.023378211900452587\n",
            "  This epoch took: 0:07:24\n",
            "\n",
            "  roc_auc score:  0.9392280678614414\n",
            "  F1 score: 0.9436008676789588\n",
            "\n",
            "Epoch 2 / 4\n",
            "Training...\n",
            "  Accuracy: 0.9815680166147456\n",
            "  Training loss: 0.020834453210528642\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9357476635514018\n",
            "  Validation loss: 0.0238601181690938\n",
            "  This epoch took: 0:07:24\n",
            "\n",
            "  roc_auc score:  0.9340648167993889\n",
            "  F1 score: 0.9414270500532481\n",
            "\n",
            "Epoch 3 / 4\n",
            "Training...\n",
            "  Accuracy: 0.9761163032191069\n",
            "  Training loss: 0.02114686740877596\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9228971962616822\n",
            "  Validation loss: 0.024579555353271627\n",
            "  This epoch took: 0:07:24\n",
            "\n",
            "  roc_auc score:  0.9194928527855176\n",
            "  F1 score: 0.9309623430962344\n",
            "\n",
            "Epoch 4 / 4\n",
            "Training...\n",
            "  Accuracy: 0.9754672897196262\n",
            "  Training loss: 0.021134121124927623\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9439252336448598\n",
            "  Validation loss: 0.0232130039448493\n",
            "  This epoch took: 0:07:24\n",
            "\n",
            "  roc_auc score:  0.9448941794580099\n",
            "  F1 score: 0.9473684210526316\n",
            "===\n",
            "Summary\n",
            "Total time 0:29:36 (h:mm:ss)\n",
            "best acc: 0.9439252336448598\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Oeba2ZmuhDr8"
      },
      "source": [
        "PATH2 = \"/content/drive/MyDrive/Colab Notebooks/'cnn1.pt'\"\n",
        "torch.save(model2, PATH2)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bg68AMvDA1mi"
      },
      "source": [
        "## **3.2.2 With Freeze Parameters**\n",
        "best acc: 0.9591121495327103"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2YD6MCvkIBtX"
      },
      "source": [
        "class BertCNNClassifier(nn.Module):\n",
        "    def __init__(self, tuned_model, embed_num = 512, embed_dim = 768, dropout=0.1, kernel_num=3, kernel_sizes=[1,2], num_labels=2):\n",
        "        super().__init__()\n",
        "        self.num_labels = num_labels\n",
        "        self.embed_num = embed_num\n",
        "        self.embed_dim = embed_dim\n",
        "        self.dropout = dropout\n",
        "        self.kernel_num = kernel_num\n",
        "        self.kernel_sizes = kernel_sizes\n",
        "        self.softmax = nn.functional.softmax\n",
        "\n",
        "        self.bert = tuned_model.bert\n",
        "        self.convs = nn.ModuleList([nn.Conv2d(1, self.kernel_num, (k, self.embed_dim)) for k in self.kernel_sizes])\n",
        "        self.dropout = nn.Dropout(self.dropout)\n",
        "        self.classifier = nn.Linear(len(self.kernel_sizes)*self.kernel_num, self.num_labels)\n",
        "        \n",
        "    def forward(self, input_ids, attention_mask, token_type_ids = None):\n",
        "        output = self.bert(input_ids = input_ids, attention_mask = attention_mask, token_type_ids = token_type_ids) #16,512,768\n",
        "        output = output[0].unsqueeze(1) #16,1,512,768\n",
        "        output = [nn.functional.relu(conv(output)).squeeze(3) for conv in self.convs] #16,3,508,1 => #16,3,508\n",
        "        output = [nn.functional.max_pool1d(i, i.size(2)).squeeze(2) for i in output] #=> 16,3\n",
        "        output = torch.cat(output, 1)\n",
        "        output = self.dropout(output)\n",
        "        logits = self.classifier(output)\n",
        "        return self.softmax(logits, 1)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "79WPystSxBa3"
      },
      "source": [
        "# Initializing model\n",
        "model3 = BertCNNClassifier(the_best_model).cuda()\n",
        "for param in model3.bert.parameters():\n",
        "    param.requires_grad = False\n",
        "# set parameters\n",
        "epochs = 4\n",
        "learning_rate = 5e-5\n",
        "optimizer = AdamW(model3.parameters(), lr = learning_rate)\n",
        "criterion = nn.CrossEntropyLoss()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "xHI4n4brxPvw",
        "outputId": "3dd4eeeb-d6dd-418a-d28f-a75f4736005a"
      },
      "source": [
        "import random\n",
        "import numpy as np\n",
        "\n",
        "seed_val = 42\n",
        "\n",
        "random.seed(seed_val)\n",
        "torch.manual_seed(seed_val)\n",
        "torch.cuda.manual_seed_all(seed_val)\n",
        "\n",
        "training_stats = []\n",
        "total_t0 = time.time()\n",
        "best_accuracy = 0\n",
        "for epoch_i in range(0, epochs):\n",
        "    #Training\n",
        "    print(\"\")\n",
        "    print('Epoch {:} / {:}'.format(epoch_i + 1, epochs))\n",
        "    print('Training...')\n",
        "\n",
        "    t0 = time.time()\n",
        "    total_train_loss = 0\n",
        "    total_train_accuracy = 0\n",
        "    model3.train()\n",
        "    for step, batch in enumerate(train_dataloader):\n",
        "\n",
        "        input_ids = batch[0].to(device)\n",
        "        input_mask = batch[1].to(device)\n",
        "        labels = batch[2].to(device)\n",
        "\n",
        "        model3.zero_grad()        \n",
        "        out = model3(input_ids = input_ids, attention_mask = input_mask, token_type_ids = None)\n",
        "        loss = criterion(out, labels)\n",
        "        total_train_loss += loss.item()\n",
        "\n",
        "        loss.backward()\n",
        "        torch.nn.utils.clip_grad_norm_(model3.parameters(), 1.0)\n",
        "        optimizer.step()\n",
        "\n",
        "        pred = torch.argmax(out, dim = 1)\n",
        "        total_train_accuracy +=  torch.sum(pred == labels).item()\n",
        "        \n",
        "    avg_train_accuracy = total_train_accuracy / len(train_dataloader.dataset)\n",
        "    avg_train_loss = total_train_loss / len(train_dataloader.dataset)            \n",
        "    print(\"  Accuracy: {}\".format(avg_train_accuracy))\n",
        "    print(\"  Training loss: {}\".format(avg_train_loss))\n",
        "        \n",
        "    # Validation\n",
        "    print(\"\")\n",
        "    print(\"Validation...\")\n",
        "    model3.eval()\n",
        "    total_eval_accuracy = 0\n",
        "    total_eval_loss = 0\n",
        "    y_true = []\n",
        "    y_pred = []\n",
        "\n",
        "    for batch in validation_dataloader:\n",
        "        input_ids = batch[0].to(device)\n",
        "        input_mask = batch[1].to(device)\n",
        "        labels = batch[2].to(device)\n",
        "        \n",
        "        with torch.no_grad():        \n",
        "            out = model3(input_ids = input_ids, attention_mask = input_mask, token_type_ids = None)\n",
        "        loss = criterion(out, labels)\n",
        "        total_eval_loss += loss.item()\n",
        "        pred = torch.argmax(out, dim = 1)\n",
        "        total_eval_accuracy += torch.sum(pred == labels).item()\n",
        "        y_true.append(labels.flatten())\n",
        "        y_pred.append(pred.flatten())\n",
        "        \n",
        "    avg_val_accuracy = total_eval_accuracy / len(validation_dataloader.dataset)\n",
        "    print(\"  Accuracy: {}\".format(avg_val_accuracy))\n",
        "    avg_val_loss = total_eval_loss / len(validation_dataloader.dataset)\n",
        "    print(\"  Validation loss: {}\".format(avg_val_loss))\n",
        "    training_time = format_time(time.time() - t0)\n",
        "    print(\"  This epoch took: {:}\".format(training_time))\n",
        "    print()\n",
        "    y_true = torch.cat(y_true).tolist()\n",
        "    y_pred = torch.cat(y_pred).tolist()\n",
        "    print('  roc_auc score: ', roc_auc_score(y_true,y_pred))\n",
        "    print('  F1 score:',f1_score(y_true, y_pred))\n",
        "\n",
        "\n",
        "    training_stats.append(\n",
        "        {\n",
        "            'epoch': epoch_i + 1,\n",
        "            'Train Accur.': avg_train_accuracy,\n",
        "            'Training Loss': avg_train_loss,\n",
        "            'Valid. Loss': avg_val_loss,\n",
        "            'Valid. Accur.': avg_val_accuracy,\n",
        "            'Training Time': training_time,\n",
        "        }\n",
        "    )\n",
        "\n",
        "    if avg_val_accuracy > best_accuracy:\n",
        "        best_accuracy = avg_val_accuracy\n",
        "        best_model = model3\n",
        "\n",
        "print(\"===\")\n",
        "print(\"Summary\")\n",
        "print(\"Total time {:} (h:mm:ss)\".format(format_time(time.time()-total_t0)))\n",
        "print('best acc:',best_accuracy)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "Epoch 1 / 4\n",
            "Training...\n",
            "  Accuracy: 0.9780633437175493\n",
            "  Training loss: 0.0245809032429911\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9591121495327103\n",
            "  Validation loss: 0.022670826517811445\n",
            "  This epoch took: 0:02:42\n",
            "\n",
            "  roc_auc score:  0.9587407053237269\n",
            "  F1 score: 0.9622437971952535\n",
            "\n",
            "Epoch 2 / 4\n",
            "Training...\n",
            "  Accuracy: 0.9968847352024922\n",
            "  Training loss: 0.020661595914810742\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9567757009345794\n",
            "  Validation loss: 0.02231573055837756\n",
            "  This epoch took: 0:02:42\n",
            "\n",
            "  roc_auc score:  0.9561961760616403\n",
            "  F1 score: 0.960172228202368\n",
            "\n",
            "Epoch 3 / 4\n",
            "Training...\n",
            "  Accuracy: 0.9974039460020768\n",
            "  Training loss: 0.02022851565669011\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9579439252336449\n",
            "  Validation loss: 0.022223981567353845\n",
            "  This epoch took: 0:02:42\n",
            "\n",
            "  roc_auc score:  0.9574684406926836\n",
            "  F1 score: 0.9612068965517242\n",
            "\n",
            "Epoch 4 / 4\n",
            "Training...\n",
            "  Accuracy: 0.9972741433021807\n",
            "  Training loss: 0.020046465187831583\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9591121495327103\n",
            "  Validation loss: 0.02219131442709504\n",
            "  This epoch took: 0:02:42\n",
            "\n",
            "  roc_auc score:  0.9589330563478586\n",
            "  F1 score: 0.9621621621621621\n",
            "===\n",
            "Summary\n",
            "Total time 0:10:49 (h:mm:ss)\n",
            "best acc: 0.9591121495327103\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GnvR9pPchNxk"
      },
      "source": [
        "PATH3 = \"/content/drive/MyDrive/Colab Notebooks/'cnn3.pt'\"\n",
        "torch.save(model3, PATH3)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qQJRCGvxEeNq"
      },
      "source": [
        "#3.3 BERT+Bidirectional LSTM"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6T_bhYg8zKmP"
      },
      "source": [
        "## **3.3.1 Without Freeze Parameters**\n",
        "best acc: 0.9345794392523364"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jlNZeaLTEjGv"
      },
      "source": [
        "class BertLstmClassifier(nn.Module):\n",
        "    def __init__(self, model_tune):\n",
        "        super().__init__()\n",
        "        self.bert = model_tune.bert\n",
        "        self.lstm = nn.LSTM(input_size = 768, \n",
        "                            hidden_size = 768, \n",
        "                            num_layers = 1, \n",
        "                            batch_first = True, \n",
        "                            bidirectional = True)\n",
        "        self.classifier = nn.Linear(768 * 2, 2)\n",
        "        self.softmax = nn.Softmax(dim = 1)\n",
        "\n",
        "    def forward(self, input_ids, attention_mask, token_type_ids):\n",
        "        bert_output = self.bert(input_ids = input_ids, attention_mask = attention_mask, token_type_ids = token_type_ids)\n",
        "        out, _ = self.lstm(bert_output[0])\n",
        "        logits = self.classifier(out[:, 1, :])\n",
        "        return self.softmax(logits)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rx1ESjulxNOH"
      },
      "source": [
        "# Initializing model\n",
        "model4 = BertLstmClassifier(the_best_model).cuda()\n",
        "# set parameters\n",
        "epochs = 6\n",
        "learning_rate = 5e-5\n",
        "optimizer = AdamW(model4.parameters(), lr = learning_rate)\n",
        "criterion = nn.CrossEntropyLoss()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "TwhtA0_iAarb",
        "outputId": "aaf0d089-af60-4018-945c-337b9c9bb9ac"
      },
      "source": [
        "import random\n",
        "import numpy as np\n",
        "\n",
        "seed_val = 42\n",
        "\n",
        "random.seed(seed_val)\n",
        "torch.manual_seed(seed_val)\n",
        "torch.cuda.manual_seed_all(seed_val)\n",
        "\n",
        "training_stats = []\n",
        "total_t0 = time.time()\n",
        "best_accuracy = 0\n",
        "for epoch_i in range(0, epochs):\n",
        "    #Training\n",
        "    print(\"\")\n",
        "    print('Epoch {:} / {:}'.format(epoch_i + 1, epochs))\n",
        "    print('Training...')\n",
        "\n",
        "    t0 = time.time()\n",
        "    total_train_loss = 0\n",
        "    total_train_accuracy = 0\n",
        "    model4.train()\n",
        "    for step, batch in enumerate(train_dataloader):\n",
        "\n",
        "        input_ids = batch[0].to(device)\n",
        "        input_mask = batch[1].to(device)\n",
        "        labels = batch[2].to(device)\n",
        "\n",
        "        model4.zero_grad()        \n",
        "        out = model4(input_ids = input_ids, attention_mask = input_mask, token_type_ids = None)\n",
        "        loss = criterion(out, labels)\n",
        "        total_train_loss += loss.item()\n",
        "\n",
        "        loss.backward()\n",
        "        torch.nn.utils.clip_grad_norm_(model4.parameters(), 1.0)\n",
        "        optimizer.step()\n",
        "\n",
        "        pred = torch.argmax(out, dim = 1)\n",
        "        total_train_accuracy +=  torch.sum(pred == labels).item()\n",
        "        \n",
        "    avg_train_accuracy = total_train_accuracy / len(train_dataloader.dataset)\n",
        "    avg_train_loss = total_train_loss / len(train_dataloader.dataset)            \n",
        "    print(\"  Accuracy: {}\".format(avg_train_accuracy))\n",
        "    print(\"  Training loss: {}\".format(avg_train_loss))\n",
        "        \n",
        "    # Validation\n",
        "    print(\"\")\n",
        "    print(\"Validation...\")\n",
        "    model4.eval()\n",
        "    total_eval_accuracy = 0\n",
        "    total_eval_loss = 0\n",
        "    y_true = []\n",
        "    y_pred = []\n",
        "\n",
        "    for batch in validation_dataloader:\n",
        "        input_ids = batch[0].to(device)\n",
        "        input_mask = batch[1].to(device)\n",
        "        labels = batch[2].to(device)\n",
        "        \n",
        "        with torch.no_grad():        \n",
        "            out = model4(input_ids = input_ids, attention_mask = input_mask, token_type_ids = None)\n",
        "        loss = criterion(out, labels)\n",
        "        total_eval_loss += loss.item()\n",
        "        pred = torch.argmax(out, dim = 1)\n",
        "        total_eval_accuracy += torch.sum(pred == labels).item()\n",
        "        y_true.append(labels.flatten())\n",
        "        y_pred.append(pred.flatten())\n",
        "        \n",
        "    avg_val_accuracy = total_eval_accuracy / len(validation_dataloader.dataset)\n",
        "    print(\"  Accuracy: {}\".format(avg_val_accuracy))\n",
        "    avg_val_loss = total_eval_loss / len(validation_dataloader.dataset)\n",
        "    print(\"  Validation loss: {}\".format(avg_val_loss))\n",
        "    training_time = format_time(time.time() - t0)\n",
        "    print(\"  This epoch took: {:}\".format(training_time))\n",
        "    print()\n",
        "    y_true = torch.cat(y_true).tolist()\n",
        "    y_pred = torch.cat(y_pred).tolist()\n",
        "    print('  roc_auc score: ', roc_auc_score(y_true,y_pred))\n",
        "    print('  F1 score:',f1_score(y_true, y_pred))\n",
        "\n",
        "\n",
        "    training_stats.append(\n",
        "        {\n",
        "            'epoch': epoch_i + 1,\n",
        "            'Train Accur.': avg_train_accuracy,\n",
        "            'Training Loss': avg_train_loss,\n",
        "            'Valid. Loss': avg_val_loss,\n",
        "            'Valid. Accur.': avg_val_accuracy,\n",
        "            'Training Time': training_time,\n",
        "        }\n",
        "    )\n",
        "\n",
        "    if avg_val_accuracy > best_accuracy:\n",
        "        best_accuracy = avg_val_accuracy\n",
        "        best_model = model4\n",
        "\n",
        "print(\"===\")\n",
        "print(\"Summary\")\n",
        "print(\"Total time {:} (h:mm:ss)\".format(format_time(time.time()-total_t0)))\n",
        "print('best acc:',best_accuracy)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "Epoch 1 / 6\n",
            "Training...\n",
            "  Accuracy: 0.9637850467289719\n",
            "  Training loss: 0.022016916648404498\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.927570093457944\n",
            "  Validation loss: 0.02422226070661411\n",
            "  This epoch took: 0:08:31\n",
            "\n",
            "  roc_auc score:  0.9243895602855589\n",
            "  F1 score: 0.9350104821802935\n",
            "\n",
            "Epoch 2 / 6\n",
            "Training...\n",
            "  Accuracy: 0.9667705088265836\n",
            "  Training loss: 0.021657877548492337\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9217289719626168\n",
            "  Validation loss: 0.024612341459109403\n",
            "  This epoch took: 0:08:30\n",
            "\n",
            "  roc_auc score:  0.9180282371303425\n",
            "  F1 score: 0.9301355578727842\n",
            "\n",
            "Epoch 3 / 6\n",
            "Training...\n",
            "  Accuracy: 0.9607995846313603\n",
            "  Training loss: 0.022032600047061004\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9065420560747663\n",
            "  Validation loss: 0.02558620500369607\n",
            "  This epoch took: 0:08:30\n",
            "\n",
            "  roc_auc score:  0.9011040948785165\n",
            "  F1 score: 0.9180327868852459\n",
            "\n",
            "Epoch 4 / 6\n",
            "Training...\n",
            "  Accuracy: 0.9596313603322949\n",
            "  Training loss: 0.02209458248699442\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9322429906542056\n",
            "  Validation loss: 0.023964609115201735\n",
            "  This epoch took: 0:08:30\n",
            "\n",
            "  roc_auc score:  0.9327485862199727\n",
            "  F1 score: 0.9366812227074236\n",
            "\n",
            "Epoch 5 / 6\n",
            "Training...\n",
            "  Accuracy: 0.9502855659397715\n",
            "  Training loss: 0.02268608755055865\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9345794392523364\n",
            "  Validation loss: 0.023826131823463974\n",
            "  This epoch took: 0:08:30\n",
            "\n",
            "  roc_auc score:  0.9351007644579272\n",
            "  F1 score: 0.9388646288209608\n",
            "\n",
            "Epoch 6 / 6\n",
            "Training...\n",
            "  Accuracy: 0.9491173416407062\n",
            "  Training loss: 0.02276955447851188\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.927570093457944\n",
            "  Validation loss: 0.024264535475000043\n",
            "  This epoch took: 0:08:30\n",
            "\n",
            "  roc_auc score:  0.9268901235992723\n",
            "  F1 score: 0.9331896551724138\n",
            "===\n",
            "Summary\n",
            "Total time 0:51:01 (h:mm:ss)\n",
            "best acc: 0.9345794392523364\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xRbeNDj_kp4G"
      },
      "source": [
        "PATH4 = \"/content/drive/MyDrive/Colab Notebooks/'lstm1.pt'\"\n",
        "torch.save(model4, PATH4)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "AaWUyMnNzQOI"
      },
      "source": [
        "## **3.3.2 With Freeze Parameters**\n",
        "best acc: 0.9614485981308412"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OUytIdW6_5qq"
      },
      "source": [
        "class BertLstmClassifier(nn.Module):\n",
        "    def __init__(self, model_tune):\n",
        "        super().__init__()\n",
        "        self.bert = model_tune.bert\n",
        "        self.lstm = nn.LSTM(input_size = 768, \n",
        "                            hidden_size = 768, \n",
        "                            num_layers = 2, \n",
        "                            batch_first = True, \n",
        "                            bidirectional = True)\n",
        "        self.classifier = nn.Linear(768 * 2, 2)\n",
        "        self.softmax = nn.Softmax(dim = 1)\n",
        "\n",
        "    def forward(self, input_ids, attention_mask, token_type_ids):\n",
        "        bert_output = self.bert(input_ids = input_ids, attention_mask = attention_mask, token_type_ids = token_type_ids)\n",
        "        out, _ = self.lstm(bert_output[0])\n",
        "        logits = self.classifier(out[:, 1, :])\n",
        "        return self.softmax(logits)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "I0leuOXrzTFf"
      },
      "source": [
        "model5 = BertLstmClassifier(the_best_model).cuda()\n",
        "for param in model5.bert.parameters():\n",
        "    param.requires_grad = False\n",
        "# set parameters\n",
        "epochs = 10\n",
        "learning_rate = 5e-5\n",
        "optimizer = AdamW(model5.parameters(), lr = learning_rate)\n",
        "criterion = nn.CrossEntropyLoss()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "jZ_jyz063VpW",
        "outputId": "124bcfc4-c738-4320-af29-253a3e025566"
      },
      "source": [
        "import random\n",
        "import numpy as np\n",
        "\n",
        "seed_val = 42\n",
        "\n",
        "random.seed(seed_val)\n",
        "torch.manual_seed(seed_val)\n",
        "torch.cuda.manual_seed_all(seed_val)\n",
        "\n",
        "training_stats = []\n",
        "total_t0 = time.time()\n",
        "best_accuracy = 0\n",
        "for epoch_i in range(0, epochs):\n",
        "    #Training\n",
        "    print(\"\")\n",
        "    print('Epoch {:} / {:}'.format(epoch_i + 1, epochs))\n",
        "    print('Training...')\n",
        "\n",
        "    t0 = time.time()\n",
        "    total_train_loss = 0\n",
        "    total_train_accuracy = 0\n",
        "    model5.train()\n",
        "    for step, batch in enumerate(train_dataloader):\n",
        "\n",
        "        input_ids = batch[0].to(device)\n",
        "        input_mask = batch[1].to(device)\n",
        "        labels = batch[2].to(device)\n",
        "\n",
        "        model5.zero_grad()        \n",
        "        out = model5(input_ids = input_ids, attention_mask = input_mask, token_type_ids = None)\n",
        "        loss = criterion(out, labels)\n",
        "        total_train_loss += loss.item()\n",
        "\n",
        "        loss.backward()\n",
        "        torch.nn.utils.clip_grad_norm_(model5.parameters(), 1.0)\n",
        "        optimizer.step()\n",
        "\n",
        "        pred = torch.argmax(out, dim = 1)\n",
        "        total_train_accuracy +=  torch.sum(pred == labels).item()\n",
        "        \n",
        "    avg_train_accuracy = total_train_accuracy / len(train_dataloader.dataset)\n",
        "    avg_train_loss = total_train_loss / len(train_dataloader.dataset)            \n",
        "    print(\"  Accuracy: {}\".format(avg_train_accuracy))\n",
        "    print(\"  Training loss: {}\".format(avg_train_loss))\n",
        "        \n",
        "    # Validation\n",
        "    print(\"\")\n",
        "    print(\"Validation...\")\n",
        "    model5.eval()\n",
        "    total_eval_accuracy = 0\n",
        "    total_eval_loss = 0\n",
        "    y_true = []\n",
        "    y_pred = []\n",
        "\n",
        "    for batch in validation_dataloader:\n",
        "        input_ids = batch[0].to(device)\n",
        "        input_mask = batch[1].to(device)\n",
        "        labels = batch[2].to(device)\n",
        "        \n",
        "        with torch.no_grad():        \n",
        "            out = model5(input_ids = input_ids, attention_mask = input_mask, token_type_ids = None)\n",
        "        loss = criterion(out, labels)\n",
        "        total_eval_loss += loss.item()\n",
        "        pred = torch.argmax(out, dim = 1)\n",
        "        total_eval_accuracy += torch.sum(pred == labels).item()\n",
        "        y_true.append(labels.flatten())\n",
        "        y_pred.append(pred.flatten())\n",
        "        \n",
        "    avg_val_accuracy = total_eval_accuracy / len(validation_dataloader.dataset)\n",
        "    print(\"  Accuracy: {}\".format(avg_val_accuracy))\n",
        "    avg_val_loss = total_eval_loss / len(validation_dataloader.dataset)\n",
        "    print(\"  Validation loss: {}\".format(avg_val_loss))\n",
        "    training_time = format_time(time.time() - t0)\n",
        "    print(\"  This epoch took: {:}\".format(training_time))\n",
        "    print()\n",
        "    y_true = torch.cat(y_true).tolist()\n",
        "    y_pred = torch.cat(y_pred).tolist()\n",
        "    print('  roc_auc score: ', roc_auc_score(y_true,y_pred))\n",
        "    print('  F1 score:',f1_score(y_true, y_pred))\n",
        "\n",
        "\n",
        "    training_stats.append(\n",
        "        {\n",
        "            'epoch': epoch_i + 1,\n",
        "            'Train Accur.': avg_train_accuracy,\n",
        "            'Training Loss': avg_train_loss,\n",
        "            'Valid. Loss': avg_val_loss,\n",
        "            'Valid. Accur.': avg_val_accuracy,\n",
        "            'Training Time': training_time,\n",
        "        }\n",
        "    )\n",
        "\n",
        "    if avg_val_accuracy > best_accuracy:\n",
        "        best_accuracy = avg_val_accuracy\n",
        "        best_model = model5\n",
        "\n",
        "print(\"===\")\n",
        "print(\"Summary\")\n",
        "print(\"Total time {:} (h:mm:ss)\".format(format_time(time.time()-total_t0)))\n",
        "print('best acc:',best_accuracy)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "Epoch 1 / 10\n",
            "Training...\n",
            "  Accuracy: 0.994288681204569\n",
            "  Training loss: 0.020469989283310168\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9602803738317757\n",
            "  Validation loss: 0.022211515597093886\n",
            "  This epoch took: 0:03:16\n",
            "\n",
            "  roc_auc score:  0.9592435658582428\n",
            "  F1 score: 0.9635974304068522\n",
            "\n",
            "Epoch 2 / 10\n",
            "Training...\n",
            "  Accuracy: 0.9974039460020768\n",
            "  Training loss: 0.019768247409865863\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9614485981308412\n",
            "  Validation loss: 0.022178327125088076\n",
            "  This epoch took: 0:03:17\n",
            "\n",
            "  roc_auc score:  0.960515830489286\n",
            "  F1 score: 0.9646302250803859\n",
            "\n",
            "Epoch 3 / 10\n",
            "Training...\n",
            "  Accuracy: 0.9981827622014537\n",
            "  Training loss: 0.019703331251598715\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9579439252336449\n",
            "  Validation loss: 0.022269479825953457\n",
            "  This epoch took: 0:03:17\n",
            "\n",
            "  roc_auc score:  0.9572760896685517\n",
            "  F1 score: 0.9612903225806452\n",
            "\n",
            "Epoch 4 / 10\n",
            "Training...\n",
            "  Accuracy: 0.9983125649013499\n",
            "  Training loss: 0.019704241765931885\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9579439252336449\n",
            "  Validation loss: 0.0223606564452715\n",
            "  This epoch took: 0:03:17\n",
            "\n",
            "  roc_auc score:  0.9572760896685517\n",
            "  F1 score: 0.9612903225806452\n",
            "\n",
            "Epoch 5 / 10\n",
            "Training...\n",
            "  Accuracy: 0.9980529595015576\n",
            "  Training loss: 0.01972182229671894\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9602803738317757\n",
            "  Validation loss: 0.02224170618525175\n",
            "  This epoch took: 0:03:17\n",
            "\n",
            "  roc_auc score:  0.96001296995477\n",
            "  F1 score: 0.9632829373650107\n",
            "\n",
            "Epoch 6 / 10\n",
            "Training...\n",
            "  Accuracy: 0.9987019730010385\n",
            "  Training loss: 0.019687023454830282\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9614485981308412\n",
            "  Validation loss: 0.022172813451735773\n",
            "  This epoch took: 0:03:17\n",
            "\n",
            "  roc_auc score:  0.9618622876582087\n",
            "  F1 score: 0.9640914036996735\n",
            "\n",
            "Epoch 7 / 10\n",
            "Training...\n",
            "  Accuracy: 0.9980529595015576\n",
            "  Training loss: 0.019721307251157543\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9579439252336449\n",
            "  Validation loss: 0.022354588991849222\n",
            "  This epoch took: 0:03:17\n",
            "\n",
            "  roc_auc score:  0.9578531427409471\n",
            "  F1 score: 0.9610389610389609\n",
            "\n",
            "Epoch 8 / 10\n",
            "Training...\n",
            "  Accuracy: 0.9983125649013499\n",
            "  Training loss: 0.019701189890519357\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9591121495327103\n",
            "  Validation loss: 0.022315482809164813\n",
            "  This epoch took: 0:03:17\n",
            "\n",
            "  roc_auc score:  0.9589330563478586\n",
            "  F1 score: 0.9621621621621621\n",
            "\n",
            "Epoch 9 / 10\n",
            "Training...\n",
            "  Accuracy: 0.9983125649013499\n",
            "  Training loss: 0.019703309340857022\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9579439252336449\n",
            "  Validation loss: 0.022361176245123425\n",
            "  This epoch took: 0:03:17\n",
            "\n",
            "  roc_auc score:  0.9574684406926836\n",
            "  F1 score: 0.9612068965517242\n",
            "\n",
            "Epoch 10 / 10\n",
            "Training...\n",
            "  Accuracy: 0.9983125649013499\n",
            "  Training loss: 0.019704056569101406\n",
            "\n",
            "Validation...\n",
            "  Accuracy: 0.9614485981308412\n",
            "  Validation loss: 0.022188497759471428\n",
            "  This epoch took: 0:03:17\n",
            "\n",
            "  roc_auc score:  0.9607081815134179\n",
            "  F1 score: 0.9645542427497314\n",
            "===\n",
            "Summary\n",
            "Total time 0:32:48 (h:mm:ss)\n",
            "best acc: 0.9614485981308412\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "M2c13LGSJq-E"
      },
      "source": [
        "PATH5 = \"/content/drive/MyDrive/Colab Notebooks/'lstm2.pt'\"\n",
        "torch.save(model5, PATH5)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gnMc4cCPFsc6"
      },
      "source": [
        "# 4. Top frequent words in fake news (test dataset)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mpS_RH3Mm3Mj"
      },
      "source": [
        "PATH1 = \"/content/drive/Shareddrives/CS682 Project/'lstm2.pt'\"\n",
        "best_model = torch.load(PATH1,map_location='cpu')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "C9of3ukEoVeV"
      },
      "source": [
        "model = best_model\n",
        "#model.to(device)\n",
        "model.eval()\n",
        "word_dict = {}\n",
        "for batch_index, batch in enumerate(val_dataloader):\n",
        "        input_ids = batch[0]\n",
        "        attention_mask = batch[1]\n",
        "        token_type_ids = batch[2]\n",
        "        labels = batch[3]\n",
        "        # forward\n",
        "        with torch.no_grad():\n",
        "            res = model(input_ids = input_ids, attention_mask = attention_mask, token_type_ids = token_type_ids)\n",
        "        # prediction\n",
        "        output = res[0]\n",
        "        pred = torch.argmax(output, dim = 1)\n",
        "        for i in pred:\n",
        "          if i ==1:\n",
        "            for j in range(input_ids.size(0)):\n",
        "              for idx in range(512):\n",
        "                if input_ids[j,idx].item()!=0:\n",
        "                  if input_ids[j,idx].item() not in word_dict:\n",
        "                      word_dict[input_ids[j,idx].item()] = 1\n",
        "                  else:\n",
        "                      word_dict[input_ids[j,idx].item()] += 1"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zLaf4v9QFmYI"
      },
      "source": [
        "dict2 = sorted(word_dict.items(), key=lambda x:x[1], reverse=True)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "4Y4lqnoQMOcU",
        "outputId": "c47976b4-abf2-4ea2-b19d-d9621fbe4835"
      },
      "source": [
        "dict2[:100]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[(1996, 8249),\n",
              " (101, 7440),\n",
              " (102, 7440),\n",
              " (1997, 5349),\n",
              " (1999, 4686),\n",
              " (2000, 4605),\n",
              " (1037, 3765),\n",
              " (1998, 3358),\n",
              " (2522, 2539),\n",
              " (17258, 2406),\n",
              " (2003, 2378),\n",
              " (2005, 2195),\n",
              " (16147, 2080),\n",
              " (3572, 2016),\n",
              " (2024, 1975),\n",
              " (2008, 1649),\n",
              " (21887, 1480),\n",
              " (2013, 1431),\n",
              " (2057, 1407),\n",
              " (2015, 1348),\n",
              " (2006, 1338),\n",
              " (2047, 1275),\n",
              " (23350, 1271),\n",
              " (2007, 1205),\n",
              " (2031, 1204),\n",
              " (2004, 1117),\n",
              " (2023, 1101),\n",
              " (1521, 941),\n",
              " (2012, 924),\n",
              " (2038, 923),\n",
              " (2011, 919),\n",
              " (6677, 883),\n",
              " (2009, 868),\n",
              " (5852, 832),\n",
              " (100, 813),\n",
              " (2256, 806),\n",
              " (2045, 805),\n",
              " (2111, 794),\n",
              " (2025, 769),\n",
              " (2022, 727),\n",
              " (2062, 724),\n",
              " (2651, 706),\n",
              " (2193, 677),\n",
              " (4484, 664),\n",
              " (2163, 648),\n",
              " (2064, 640),\n",
              " (2042, 639),\n",
              " (2988, 630),\n",
              " (2102, 609),\n",
              " (2561, 581),\n",
              " (2019, 580),\n",
              " (2030, 576),\n",
              " (2001, 574),\n",
              " (2085, 553),\n",
              " (2035, 539),\n",
              " (1055, 534),\n",
              " (2040, 515),\n",
              " (2902, 507),\n",
              " (2740, 498),\n",
              " (2110, 493),\n",
              " (23713, 493),\n",
              " (1016, 485),\n",
              " (2017, 485),\n",
              " (2065, 484),\n",
              " (5604, 478),\n",
              " (2039, 476),\n",
              " (2020, 471),\n",
              " (1015, 462),\n",
              " (2097, 456),\n",
              " (2084, 454),\n",
              " (2634, 447),\n",
              " (2053, 430),\n",
              " (2021, 409),\n",
              " (2058, 400),\n",
              " (5022, 400),\n",
              " (2487, 387),\n",
              " (2027, 373),\n",
              " (2692, 372),\n",
              " (2115, 372),\n",
              " (2037, 371),\n",
              " (2951, 370),\n",
              " (2029, 362),\n",
              " (2575, 362),\n",
              " (3207, 361),\n",
              " (2055, 355),\n",
              " (2080, 354),\n",
              " (3189, 354),\n",
              " (6090, 353),\n",
              " (3231, 352),\n",
              " (4609, 347),\n",
              " (2683, 346),\n",
              " (7712, 340),\n",
              " (2420, 339),\n",
              " (1017, 334),\n",
              " (2553, 333),\n",
              " (2154, 332),\n",
              " (2549, 321),\n",
              " (17404, 321),\n",
              " (2050, 318),\n",
              " (2069, 318)]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 111
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "O1wKqeUX1iW9"
      },
      "source": [
        "word_list = []\n",
        "for i in range(100):\n",
        "  word_list.append(dict2[i][0])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "lYfi2kY-G88s",
        "outputId": "45149185-7584-4e7c-ea4d-1ece0f8b6d28"
      },
      "source": [
        "tokens = tokenizer.convert_ids_to_tokens(word_list)\n",
        "print(tokens)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "['the', '[CLS]', '[SEP]', 'of', 'in', 'to', 'a', 'and', 'co', '##vid', 'is', 'for', '##19', 'cases', 'are', 'that', 'corona', 'from', 'we', '##s', 'on', 'new', '##virus', 'with', 'have', 'as', 'this', '’', 'at', 'has', 'by', 'deaths', 'it', 'tests', '[UNK]', 'our', 'there', 'people', 'not', 'be', 'more', 'today', 'number', 'confirmed', 'states', 'can', 'been', 'reported', '##t', 'total', 'an', 'or', 'was', 'now', 'all', 's', 'who', 'hospital', 'health', 'state', 'amp', '2', 'you', 'if', 'testing', 'up', 'were', '1', 'will', 'than', 'india', 'no', 'but', 'over', 'patients', '##1', 'they', '##0', 'your', 'their', 'data', 'which', '##6', '##de', 'about', '##o', 'report', 'pan', 'test', '##un', '##9', '##mic', 'days', '3', 'case', 'day', '##4', 'vaccine', '##a', 'only']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IpdNXY1Z6iCh"
      },
      "source": [
        "#drow word cloud\n",
        "from wordcloud import WordCloud, STOPWORDS\n",
        "stopwords = set(STOPWORDS) \n",
        "all_text = \"\"\n",
        "for i in range(14,len(dict2)):\n",
        "    for j in range(dict2[i][1]):\n",
        "      words = tokenizer.convert_ids_to_tokens(dict2[i][0])\n",
        "      all_text += \" \" + words"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wKzkut4K7rkZ",
        "outputId": "8f7d06e1-1a7f-43f1-dae8-62f97c567cc9"
      },
      "source": [
        "wordcloud = WordCloud(collocations=False, width = 800, height = 300, stopwords = stopwords,background_color ='white',  min_font_size = 10)\n",
        "wordcloud.generate(all_text)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<wordcloud.wordcloud.WordCloud at 0x7fe621ff5a90>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 32
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 247
        },
        "id": "YlUjdO4u7wLJ",
        "outputId": "18ca92f1-b967-4f23-dd3a-c2ff47d06fa7"
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "plt.figure(figsize = (8, 3), facecolor = None) \n",
        "plt.imshow(wordcloud) \n",
        "plt.axis(\"off\") \n",
        "plt.tight_layout(pad = 0) \n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAk4AAADmCAYAAAAjiQpfAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nOy9d3gc13nv/5mZ7YsFFr03AgQL2KnCJoqiVa1iuclxt68tO3bsJL5xfjfOTXF+SRzHcZcdF1lyk2THVm+USIpFpFjEAnai97ZY7C6215lz/1gAIgiABEgALN6PnkcPsTNz5kw78533vEUSQpAiRYoUKVKkSJHi4shXugMpUqRIkSJFihTXCinhlCJFihQpUqRIMUVSwilFihQpUqRIkWKKpIRTihQpUqRIkSLFFEkJpxQpUqRIkSJFiimSEk4pUqRIkSJFihRTRHeR5alcBSlSpEiRIkWKPzWkyRZcTDilSJHiHBKqxh9ePcKAK4A93cwn33szkjTp85UiRYrrADWh8Ydf7iEnP51b71qKwXhlXp17t5/mdF0nAoGiKNy8sYZlN1RecntXy3Fda6Sm6lKkmAaaprF1bz1/2HKUV3advtLdSZEixRygahpbX6zj7T2NJOLqFeuHrMhEI3Gaz/ax5ZnDtDT0X3B9TROEAlE0TZtw+dVyXNcaKXmZIkWKFJeIMxxgW3cz51dgyDSauaWwEpvBeIV6luJ6ZO2mhazdtJDD+5r5/tefv+j6bqePpx59k499fhNZubY56OGfBinhlCJFihSXSId/iH9+eyuqJkj+l2RxZh61Wfkp4ZRiRhnjFnARDwEhBG1NDo4dbOGDn9owux37E+OaFk6a0OgINSMjU2QuQyfpCalBvHE3MS0CgEE2ka63Y1VsF/VFSWhxfIkhAgk/qoijSAom2UqGPhODbByz/ci+hdAosVSilwyjy8NqiIFIL6pIkK7PJNuQiyTJo9s5o32E1ABZhlwy9FmzdHZSpJgbPNEwoXiMfIsNnXxlZv/9sQhDsQgFFht6WZmz/VZnZPODDQ/gioRwR0Js627ilNsxZ/tPMbdIkkQkEsPRN0Q4GEXRyWTl2sjKsaEoY+/9eCzBQN8QPm8YIQS2dDO5BRkYTfox75J4LIF70I/XEyIeV1EUGVuGmbyCDHR6Zdo+lKqq4Rn043EFeXtPI0PuII2ne3AN+AAwWwyUVORgNOkv6bhSXOPCKaHF+VnLN5AlhS9Xf51AwssB107agvX4E8mbJE2XQYmlkvXZt7MgfRnSBDJdCMFQ3MVB9y4a/CdwRHqIqhEMsoFMQy5VaQtZm307hebS0e0TIs5v23+IJz7Il6u/TqV1wWh7LYGz/L7zpwRVP6szN/BQ6cMYpOSXZ1gN8lLvkzQHzvC+kk9zU9ats36eUqSYLRKaxpONddQN9vLNNfeQa7bOeR9UofFyRz3Pt57mO+vvoyQtY872bTeaeXf5wtG/PdEIZ9wDc7b/FHNL0B/h2d/uY9+Os/i9YTRNo6a2mD/7zEaW3lCJLCffD35fmK3P1/HGy3UM9HnRhCAnL521mxbxgU+uJy3dhCRJxGMJdrx6gteeO0J3uwtN1UCCgiI7dzywkgc+vIbpxp6EglFeefowR/Y109vlJhyK8uh3XxsVQJU1Bfz5395DQXHmtI8rRZJrWjiNEFaDnPHVsW9wG0HVj1Vno9BUSlD14497OeU9hCPSzUPyw1RZF41R8EII3HEnv+v8KR3BRiRkrDob2YZ8ImoId2wAx2APHcFm7i/6KNVpi5AkGRmFInMZnvggfZGuUeEkhMAR6SGihZGQ6Al3oIoEkBROgYQPb9yDQTaSZyy6EqdrThBC4A9GSSRULGYDRkPyVovGEsTiCVRVgASKLGM06DBM8mV1bjuyLJFhM0/6BaaqGv5gBE0IDHodaRbjmDY0TWC1GFAUmXA4RiyhopNlzCb96KAST6iEI/HRNiznfR2ey8hgoglBNBonFlfRNIEkSej1MiajHmWKFhghBPG4SiSWQNU0EMn29XoFk0F/0YErFk8QDMVQFAmr2YiiyAghiMYSRGMJNO2d863XKxj1uhkZDAcjQfb3d9Du9xDXroxzaSAWY19/Byfd/UTVlINritnjzPFOJFnir/7pPdizrDSd6eXR77zG737xJjW1xZitRtSEyuvPHeHpX7/FAx9ew5pbk++GPdtO8+rTh5EViU98cTMAik5Gb1BYefM8PvNXd2JLN+NxB/jdo7v5/WNvsmptFWXz8qbVR4vFyL0fvJF33becF546yLYXj/K3//Z+cvPTAdAbdWRmpU37uFK8w3UhnABe7fsfCkzF3Ff0YZZk3IBO0pMQcQ6532Sr4xmc0T4OunZSbK7ArFhGtwuqfl7qfYrWwFmyDfncUfBeVtjXoJcMqCJBa7CBrY5naQmc4fne3/Dpiv9NjjEfWZIpMc/jtO8oveGO0fZiWoSBaA8SUGyuoC/SRSgRwKwkv8RHhJNRMZNnLJzr0zRnxBMqf/ufz3GqqZcvfGQjH7x7Jcfre3htzxlONvTiGgoiSZCbmcaq2lLu3LCI2vmF6HXKpO3kZ6fz9COfmVTIdPV5+PK//hGPN8TdGxfzD1+8G4BwJM7fffsFehxD/P2f34XJoOPRP+yjoc1Bfo6N9925gnturSUQjPDH1+p47c0zhCNxaucX8pkPrKO2phB5gn0adArhaJyjp7vYsvs0Z5r7GfKFsVoMzCvN4fZ1C7jlhmoyMyzjtj2XWDxB3Zlu9h1t5fCpTgZcfhKqRma6hfkVuWxeu4CblpWTmT55O28daeWbP99KYW4G//SleygpyOR0Uy9b3jzDkVOduIaC6HUK2XYrC6sK+PB9q1lQmX+xy3hRugNeWnxulCuYksETDXHCdeHoohQpZgJ7Vhof//PbqFlSjCRJFJZkUn+yi63P1xGNxjFbjfh9EZ57Yj/rNi/ifR9bOzollpOXTkfLANtfOsZ7PrIGe6YVWZa57Z5lY/ZRVpWLa8DPo999jbZGx7SFk6KTyclLR9M0bBlJy1ZeoZ3CksxJt5nKcaV4h+tGOOlkHXcWvJ+lGTciD/sT6SUDN2TdQne4jQOuHfRFuvDFh0aFkxCClsBZmvwnMcpmNuXdy+rM9ShS8rToJD01tiVYFAs/a/0mveEOjnj2cmf++5CQKDCXABL9ke7RfoTUIINRB+n6TBbYltIdbqM33Em2MT9p+Yh7CSZ8FJpLR8XU9YoQAiGgtWuQ7fvq+cUf9zEw6E8ulAABnX0euvo8HDrZyVc+dRtrV1aOE0Yj7ZwfuTTZPjUhxq07YnVqah9gz5EWTjX2DvfNxaN/eIv8nHRO1Pfwxy11xOIJhIC3T3QgBHzt83dSkJs+bl+yIvPKzlP85vmDuL2h0d+j3gRubyenm/tobHfy2Q+undBSJgSoqsoft9Txhy1HGXQHEDBqmu8f9NE/6OPo6S5uW1PD5z60nmy7dWLL3HB7gVCUcCTO7kNN/Pz3e+kd8CKSxiZicZVgOIbD5ef+zUsvei4nO78j+9OEoHFoEGc4QIHFhjZ87s9HgknF7rntnb8Nk2x3/jatfg89AS8GRUEwvT5Mtv+L9eFqYeTZkKTJj+/c03G56/2pU1yeTV6hffTcyIpMdm468XiCeCxp7ezpdOFxBWg41c33/+WF0XU1TaOtsZ9YJI6z34s904oQAlXVaDjVw+ljnXicfkKhKH1dbmLRBPE5ShEwleNK8Q7XjXAqMJawJGP1qGgawSAbKbNUcdi9h0DcS1QLjy6Lixj1vuOE1CCl5nksy7hxVDSNadtUytKMm9jv2k6d5y025t6NSbZg12dj02XgjjmJaGFMsplAwsdAtJcq62IKTWXoJB3toUaW2m9EQ8MR7UFDo8w875obmLyeIF5PkNLK3Gn1fX9dKweOtaFTFD7+4E3csLQcu81M34CPrfvOsudQMz2OIR757W6W1BSRnmaalf5HonH+8FodS6oL+fE/fwjHoJ/v/WoHXn+Ex5/ej8cb4iP33cC6VfPYvq+e57ed4ExzH2da+sjPGR9c0OPw8Ngf91OQl85nPrCOmsp8JAlONvayZfdpGtsHeOGNE6SnmfjMB9aiKGO3TyRUnn69jl89e4B4QmXpgiLuu20pVWU5KLJMW/cg296q59DJTl7dfRpV1fjbh2/HZNAzGYFQlIPH23lu23EMeoUHb1/OsgVFWC1G3N4Qpxp7cQz6qa0uuKRz6IqGeb71FKfc/dR7nHQGhlCFoC/o575XfjnOMrfAnsu/3XwXVRnZY34XQhDTVBqGnGzvbmZvbxuOcACdLFNhy2JT8TzuKq0h35w2brrTG4uypbOeI85uGjyDtPs9JIRGIqHx/td+O876VWzN4F9vvpMVOWOnxiOJOL0hH4cHejjQ38kZj4OhaBhFlskzp7E0u4B3ly9kWXYhFt3k5/xKcvBAC089uY/Pf2EztbUl45Y/9eR+du44QzgcQxOCv/va/SxfXjZuvRdeOMorLx0jHI4Rjcb5m6++mzVrq+fiEK4pzBYDBtM77wgJCXn4/hzRnb6h5EeU0CAcjo3ZvqwqD4vViMlkSPrWuoP86pHtHD3QQlFpFsXl2WRmpxEOxZBliSl8K84IUzmuFO9w3Qincms1MhNH0xhlM7IkkxBxNPGOeo5rMXojnQDkGgux6e0Tbq9IOios89nv2k5IDeKI9FJhnY9VZyPTkIMr6sAVdVBsrsAVHSCQ8FFiqSDDkEWaLoOuUBua0NCEOmqdKrHMm+EzMPvse+MMLzy1n/9++svTclj0+iMU5mbwV5/cxIbVVaO+NdXluaxeWoqqauw62ESf08vhUx1sXrPgIi1eOqqq8dmH1lFZko0QgmP13byw/QT1rf3cuX4RH3ngRqxmA2kWI/vr2ujs89DdP4SmiXHCJxSOs3xhHv/nc3dQXpQ1KqwWVRVwQ20Zf/+9l+jsdfPC9hPcfcsiyoreiaAUQtDc4eSZ148RjsbZvKaGL338VnIz00bbqanMY92qefzno9vZsb+BHQca2XRzDbfcUDXp8QVCUX73ymFqKvL40sduZUFl3ugACHD/bUsJhqOjPmfTpSfg5Y8tJwknYghAP9y2JElY9AZ0590YFp1+wmnOmKbym/ojPF5/mL5Q0gppVHRomkaHf4g3e1t5pb2ev1y2nnUF5WOi9ZzhAM+2nsIR8o/tw/D+9OcJLYtejyKN9zWrG+zl64e20TA0OLq9UdGhCUFP0EfdYC8vtJ3hL5au5WPzV2LRGy7pnM0mXm8ItztAMBibcPmm2xYxv6aAgweaeWP7aRKJia0Ha9ZUU1KcRV1dOy+9WEc8npjNbl/TTBRgdC72LCtIEpvvXc57P7Z20qg0IQQnDrex/eVjfOjTt/Chz2wcndbb+eoJThxqm/G+X4iLHVeKd7huhFO6fvL525EXkWCselaFSiDhBSQyDBdOC2DVpaGTdCREYngbsCppZOqzcUS6GYw5KDKX0xluAaDIXE6GLpN0vR1v3I0v7sGsWHFEupElhUJT6eUc7jWFosjctmY+a1ZUjHNINhv1PLB5KbvfbiKhajS1O2dVOFWV5pCVYRm9J5bWFPHC9hPodQpLFxRhNSdfjvk5NiyW5L893hAJVRs3ABoNOm5ft5Cywqxx1qjK0mwevH0ZP/zNLvzBCHsOt/CR+zPPMdsL3jzcTN+Al3SbmQ+9e/UY0TRCmsXIB+5awb6jrUSjcXYebGTtykp0kwzGmiZQVcEXP7KRhfPyx7UnSYw6zV8KNfYcfrzxQTQhUIXGU43H+G3jUXJMFv5r7b3joupMikK+ZWziPVXTeKblJI+c2k8wHuWmvFI+VrOS0jQ7CaFxwtXL7xqPc9TZwz8f2sqPb3mQRZl5o8dSZrPzrbXvJq5pgGBLZyPfO74Hk6Lj32++e1xUnV6WKbCMT/5XaElHQmJpVgGbiuexOrcYu9GMJgQnXf080XiUJq+Lx84cYnVOCatyi646K/G69TVUV+dTVDzx+FdcnElxcSYed4Ddu85O2k5BQQYFBRlEo3Fe23JixvvZerqbg9tPjQ7AZQsKufG2xRhMV6cl73IoLs8mv8jOob2NbL53GfasNCRpZGpeAyHQ6RWEgIE+L2pCY/W6agxGXTKgIxKntamfcGhiMTxVJElCUZL7CQYiCJEMXBmZnr7a7uVz8boC7H/tBB5nMjreYNSz+f03kpk33mXiSnDdCCe9NP2vQYEgIZJfVroJpuhGkCQJWVKQJQUhBAktDoBZsZBtzEP1qTgjfYgMQVuwgTRdOtmGPDL0Wdj1ObTFGnBEeyg0leKM9pNrKMSiS5t0f9cbFpOBdSvnoddNfI7zsm3YrCb8wQiec3yFZoPcrDT0+nf6kW1Pvuh1ikxp4TsvH6NBh2HYUT0STUzoX2U26VmxqGTS6LQblpaRnmbCF4hwqqmPhKqNOr8nVI09h1sQQGFuOouqCiYdyPKy0ijKS6e1y0V3vweXJ0B+zuQDyM3Lyqkun9506lQx6/RUD0+7JTSNbFPSX1Any1SmZ1JkvfDAJoTgtMfBY2cPE4xHuadsAf984x3kmN5xfF+dW8zm4mo+t+sZmrwu/uPoTh7d9H5Mw9NlRkVHZXrWaHtHzb1A8jktt2VSlTG13GjlNjs/2/Q+Ci3p6GR51A9KCMGKnCKWZhfwmZ1P4wgHOOjoZEVuIcosfJVrmuDQoVaKizMpKclCCEFDQx9eb5iFCwvJyLAQCkVpqO+jqDiT/PwMmpr66e/3ggCdTiYrKw2z+eqziI3w2lP7efnXexhRTrc8sIpla+dfl8IpLc3ERz+3icd/sI1v/+OzbLp7KVabCd9QiLYmB/ZMKx9++FYkCSqqk1N3L/7PQSKReNIKfqCVw/ub0RvGzqCEglGc/V4i4TidrQPEYyr9PR7qT3ZjMOjIyrWRkWkZ89yXVuZgNOn4wy/3cNs9y1B0MiazgfmLijBbrt77peVUF09891XcjmEjRbqF5RtqUsLpakBGwiibADHG9+l8RsRSQktgUvQYlKQPjiTJ5BgLkSUFT3yQqBrGGekj31SMSbGgVwzkGgto9J/AFR0YjtRTyTcVoZOu7gFDCEHAF+GNl+ro7XJRXpVHOBgb81CGg1H27ThL05kejEY9q9dXU7uyAkV3np+ZXqG0MHPC6b2kKJXQKTJCJKPoZhOL2YByjtAZmX6TZGnU2jTar+EpH1XTJpzn1+sU8rMnLmMgSRIWk4G8bBu+QASvP4w/ECFrWKgFw1H6BpKDgtsb4luPbpu0z6FIHPew30QkmiAYvvCXaHV57qQWqSuNKgRvO7po93vINFp4qHo5Oaaxg70ElNsy+fiCVXz90HZOuPqpG+xlbUH5jPZlRGhN9LsELM8uJNdsxRMN0x1MOtnPxmyGpgl+8fNdrN8wn49/YgPhcIzf/uYtzp7p4W+++m7Wb6ihu9vD97//Ol/60h3k52dQX9/L3j2N9PYO4feH+b//9z3ceNPVOf0fCcVoONY+9uNjrpx3ZhAJmHgQG/0fkHSs3nD7YkxmPVuePcIvvreVSDiG1WairDKXJR+8MbmZJLFoeSl/9tmNbHuxjm//w7MYzXqWriznk1/czG9/unPM7prO9PDbn+zE7fQTDEQI+MJsf+kYB3c3YDTpeM+H13DP+24Y7YokSay8uYr7P3QzW184yuG9TRhMOlatqabkr7NHhdNUj2uu0FSNrmYHQ07/6G0ylcCgueRPWjjpZD05xgKc0T6c0X5UoaJI4/2kBAJXbAANFb1sIMuQO7os31iMQTbgiw/RGWompkWTwkk2IyFRbK5AI5ktXCM5tVBgKr3qhVM8luAn33yJvi43q9fPp6NlgBOH2kYLQcZiCX7x3dfo6XCx/MZKvEMhfvyNl/n4X7yLDbfXjmlLUeRZc/ieLjqdPKElRkIalwrhYsiyhOUCX/l6nYx1eHCKxROEo/HRZUO+MAk1WXhzwOXnxR0np7TPhJp0gr4QyQi+KTU358TUBIed3ahCI89sZXVu8STXA1bkFFFoseEIBTg00M2a/LJZsaJpQiOUSBBKxIhrKqomhn0SBbph36hwIs5suclKEsyvKaCvb4hIJMbAgA+3K0BBoZ3OThfrhGDIEyQcilFWnrT23XXXMm67bTHbt5/md0/tn5V+zRQdDb24Hb6xP15d78GLotcrfPvxz6DTK5jOsdTIisT9D93E5ncvG1MLzmQ2sG7zIlbcXEU0Ehv2kZTRG3RjLD3WNBPv/eha7nzPShJxFUmWsVgNGIx6Fi4rHbPuwqWl/N03P4jQJj55ljTjOJ1jyzDz0Kdv4d4P3EhCVZEkCZNJn1z3Eo5rLoiGY9QfbU9Oa16l/EkLJ4NspNJaQ4PvOM5IP/3hLootFePWi2sxzvqPAZBtyCPH8E7+m3xTIQbZRDDhozXYgCzJw2IqeWOWWCpRJAVXzElIDaKT9OQaCyYUaFcT9Se7aTjZw5f/4QFWrKkiGonz2Pde59CeBgDO1HVy6mgHf/uND1C9qAg1ofGrR7bx/BP7WHPrwjFtSSRFxGwzEiJ/IWRJvsA31PReyrIkXUSgSKOO0UIbmyIhaVlL/l1WmMnK2qn5vOXYrWTYLixCk75YV6dyiguNTv8QAAUWG9ZJHK4lSSJdbyLXbKUn6KM7MIQqNHQz/Nx0+D3s6mnlpLufZu8grkiIYDxGRE0QU1USIjl4z+Z7XpYlFtQU8MaO04TDcQYcPnQ6hcW1RXR3uQmFYvT3e8nNtWE2Jc+XwaDDYNBhNumv0iudRAhB04ku/EPBK92Vy0KSpAkFhCRJWNKMo0LkXGRZJs1mIu0Cz6skSegNOuxZ4103MrPH/mY06ceUSZkqBqMOg3Fi15BLOa7ZJhyMcvZI+5zvdzr8SQsnGYX5abUcMr6JO+bkkOdNsox5SWvRsK+DhsYp7yHaAg1ISNyUdSs6+Z2b16LYyNRn40946Qg2Y5TN5Bjf8VfJ1Odg1aXjjbsZig1i1dnI0I93Jr7a6OtyI0lQtSiZpNNo0lNSns2xA8lbpqW+F7fTx2Pfex3dsCga6BtCUwXBQATzHD9wQghUTSM2R3lPAFRNEIurk0aoaZpGNDrsQ6dTxvhWWc1GRsTNvNIcvvKp25iK2JEkrtppuKkghCCUSFreJhNNI+hlGaOcPGcRVSWmqehmqA6dJgQHHJ187/geTrj6UTWNdIOJeelZ5GVZSTeYMOv0vNh+Bldkdv3uAMoqcnANBggGo/T0eMjOSWPhoiJeeP4o3qEQHR2DlJXnjPN7udqJReK01/cSDccvvnKKFEBfxyDOHveV7sYF+ZMWTpIkUWapZlPuvbzc9xR7nK/jjXtYnbkBi2IlpkVp8J/gbfduNDRWZ25gmf3mcW2UWqo44N6BL+7Bbsgmz/ROvhhFUiizVFHvO05ci1FgLiXTkH1+V65izilPc86vQgjS7Rbueu9qDOcIB6NZP+NZZs91vlY1wWRVTLy+8Li8KbNJIqEy4PZTWjDeT0YIQTgax+kJAJBmNWI757xkZVhITzPhGgrS1e8BJAz6a+uleClIkoRJGS6/o1445D0hNGLDZVz0sjw6bXa5CCFo87n59yM7OON2UJmexecW38Q9ZQsx6/TJ5I/D9/0BR+ccCCeJzEwrOr2Ca9BPY2Mf1dX5lJXlEPBH8HiCtLU5Wbu2esyzdi3gcfppOdV98RVTpBjmyK6zyRJRVzHX1lM4C8iSzI1ZG4loYXYPvMqxof2cHHobo2IirsWJixiKpOPGrI3cnvceTLJ5XBvF5goianJwtekyyNSPFUZl5iqOet4aXX6h1AlXC4UlWQiRtCwl5+nj9Ha6RvO7VC8qQpZlcgsyWLyibNRCpyY0dHqF2AzlgZGkd/yIEqqG1x8mJ3Nis/PJxt45dZ2IxBK0dAxOKJwAuvuHcA0FkSWJ0oJMzMZ3LJV6fTL9wa6DTTjdAepb+1m2oHiuun7F0EnJ1AD1Q06c4SAxVcWgTCwYQ4k43lgEWZLINlnQz5C1CeCEq4+z7gEMio6HF9/EB6uXjcv1pAkx7Ns0u0gSpKUZyc9Lp6vLTX+fl/Xra0hPN2Gzmejr9+J2BSkosF9TleqFEHicPrqaHVe6KymuETRV4+SB5ivdjYtyTQsnCQm7PpuYFsWkjBc0I+hlA3Z9NoqkmzDtgE7SszHnbmrSlnDC+zZN/tMEVT8mxUyRqZzl9pupsNaMTuGdT5llHpn6HJBgnnXhmKk8gHLrfDINOUhIlJorLyl1wlyzcFkpC5YW86sfbmP1+nZCgQj1J7pGp5tqV5azal01v/juayy/sRKL1URvl4vKmgLe85G1M9YPWZIoL8riwLF2wpEYR093cfu6BWOSOgohcLoDbHnzzIztdyqEIzF2vd3EysUlpKeZxtwbkWic57YdR9MEZpOe9avHZorXKTJ337KYt0904A9G+cOrR8nLtpGfPT5DOSSn/cLRBEaD7qqaqhv14QJUcXFnToOisDKniN29rQyEAzQOOanNGp9vSghBu89Dd8CLWdFNuM75fQBIiItP1Qqg3e9BQ2BSdNyQVzJhgsx2vxt/PHrR9mYCW5qJwqJMmpsdRCJx8vLTSUszUVKaxbG6dmRZIjdvbp10LxsBjcc6Cfomj1hOkeJcetuc9He6rnQ3Lso1LZx0sp7/s/Dbo39PNrAutq1k8cKVk7YjSRI6SU+xuYJicwX3TFCN4kI+SbnGQv5x8SOTtl1uqeYfFz0y5rerHb1B4Qt/dx9vvHSMvm435dX5bLhjCaePdiBJoDfo+PRf3cmRt5o4e6IL96CfsspcVq+bP6P9kGWJVbWlvLTjJKFInN+9fBijQcfSmiKMRh2hcJzmDidPv1ZHv9OHQa/MmZ+TIsvsOdxMRpqJezfVkp+TjixL9Dl9vLrrNIdOJIs/r64tZcXC8eUwViwq5q4Ni3hpx0l2vd1EOBrngc1LKSnIxGo2oGqCcCTG4FCQrl43p5v7+fT711BeNLU8RbONJIHdmPxgiSTidPiHKE2bOPv+CDpJZkNRBU+3nqI/5OPZ1lNUpGeRdp6/kysS4onGo8Q0lcr0LG4prJy0zczhPqiaRvOQiwX23EnXHcGiS+5PFRqucIj5Y9ctk0AAACAASURBVHNmEohH+V3TcXyxyEXbmgkMRh3FxZm8+OJRCgvt5OWlYzLpKSvL5rlnD5NfkEHecA6bYDBKf98QoXCMjk4XsZhKS8sAZosBi9lAQWEGFotxdL1wOEZnp4t4XKWl2YHRoMds1lNYZMdiMRKJxOnt8RAKx2htHSAeV2lrc5KZacVo0lNUaMd6CVGxmhDU7amf6VOV4jqmvq79mhDa17RwmqoAmen1prvdtSCUzkeSJGwZFh782Loxv9eufCeXjslsYP3ttaw/L/3ATPdjyfwi1q6sZOeBJhraBvjO429QnG/HoNcRjcXpHfDi8YV5962LaWp3Ut86N1MDNy+vQFU1ntt2nEOnOsnKsCBLEq6hIN2OIRKqxrzSbD7x3ptHHejPPa70NDMff89NDPnD7DrYxP66Ns409ZGTlYbJqEfTBNFYAq8/zJAvhMGg46MP3DgnxzYVZCTKbXYyDCa8sQhPNtZRYLFRbkuKp4SmEUzEyDCYRqfZJElicWY+76lczM9OH+SZ1lPIksQnFqym0JosFHzWM8Cv649w0NGFQVZ4ePGNo+JoIkrTMsg2WXBHQjzZVEeZzc4Cey6SlHx5+2NRbAYjxmHfKglYlJmHSdERTsT5ffNxymx28szJKWBnOMjj9Yd4uuUkF4rBFEIMVyMQow6AgneKBqvDAQtII56C0qTFhiVJorg4k2gkTlGRnYzh7PZFRZkkEhp5uelkZCSThDY39fOrX+7B7Q4SDEYJhaM8+8whtmw5Tk6OjU//r40sWVJCa+sAv3h0F0OeEMFghGg0wdN/PITZcpxMu5WHP7+J2toSeno8/PiRbcOlW6LEYgleerGON7afIS3NyGce3sSqVRVTvCveOTfhQITGY53T2u5qZEyR8fOKIcNwCqThCNuRa3ulx/zRCF4x0v/h+3KCatojeZxG781z8kDNFSNuHi2neojMoZ/qpXJNC6cUfxrY08381Sc3Y7UYeftEB15fmJONvUiAyainICedh+5Zxd0bF/PjJ96cM+F0+/qFLFtQxFMvHeato62caeojFlfR6WTSrSYWVxfw6fetGS7+O/EglJ9j42ufu5Ml84vYcaCBvgEvXX0e4gktmcZBr2A1GygvzqK2uojcTOuE7VwJJEliWXYh95Qt4OnWk7ze2cju3lYyDEnrRCAeY3lOIf+x5u4xliizTs8Xa9fgjUZ4oe00v6w/zO+ajpOmN5AQGsF4jJimkmuy8umFN3Bv+aIL9qE0zc5DVcv4Zf1h9vd38MHXnyDDYEKRJALxGGU2O99edy8LM/NGt1uaXcCDlbU803qSF9vP8GZvK8XWDBJCozfoQxUa95QtIN1g4rGzhybcd7vfw7Otp/DGIvhjUfzxKGc8A2gIOvxDfO3Aa+SarFj1Bmx6I0uz87mjtGbSgsFr1lbz61V/jk4nj/oyrVlbzcpVFSiKNPpb7ZIS/v0/PjhhDklJkjAak8P64kXF/Mc3H5pkPTAO+9xVVOTwb9/44IRJBiWJKTmka5pGOBAl6A8T9IUZGgxwYl/jpGkIvO4g9UfaMF9GfrfcIjv5pbMXaBOPJfAM+OjvdNF8souWU130tDvxDgYIB6NoqobBZCA9y0peUSaVi4uoWVFOcWUeuUV2TFbjnAsoIQTxaAKP04ez10N7fR/t9b30tjnxOP34h4LEInGEAJPFgMliwGw1kZWfTmFFDiVV+RSW55Cdn4E9x0ZahhllmvntptPPgC95vwR9YdrO9nBsb/2EearUhErLqW4il1GGxmw1Uja/AL3x8mWPdJGMnFe3a3uKqxZV1Xh971mc7gBWs4H337Vi0kFkyB9m696zRCJx5pXlsGH1xEVs43GV+jYHHT1ufIEwsixjTzezqKqA4vzki/nIqU4aWh1j2onHVd7Y34DD5ae2upCVi0tQFBkhBP2DPrburcegV7jrlsVkZbxT+uO1PWdwDPqZX57LTcvK0ekUVFXj1d2ncXtD3HtrLdmZVhIJlbOtDtq7XfhDUcxGPUV5GSypKZpyXTghBIOeII3tAzgGfYQjcZDAak5mH68oziY/2zapc3B7j4v9dW3E4iprVlRQU5E3Z4N2b9DHbxqOsqunhe6Al5Aax6ToyDJZ2FxcxZeXriPXPNahXwhBMBFjS0cDW7saOe7qxxMNoZNl8s02lmYV8N55tawrKB8ttXIhnOEgf2w5wZaOBjr8HgKJGEZFh91g4qb8Ur66YuO4acSeoI+nGuvY2tVET9BLTFWxGYxUpWdxd9kCHqys5bTbwSd3/IEHK2v59rp3j0mHsL27iS/ufn408u9i3F5SzX+uvYds09Ujfi+VRELFM+Cjr32Qvo5BHN1u+jsGcfZ6cHS7cfd7Zz2B4Ye+fCef+rv7Z7xdIQTNJ7p469VjnDrYQtPJLmKRqQUJyIpMUUUOS9fO56Z31bJq48I5KyuTiCc4c6iN/VtP0nC0ndYzPUQvwXqjKDJ5pVlULiyidH4Bi2+cx+IbKkk7Z2ycLkIIAt4Qjk43/V0uetucDPS4cXS5cXS7Gez1EA7Orj/h/GWl/ONjD5NbNOXgrEkH0JRwSpEixWUhhCCqJhiMhAglYqhCoEgSBkVHht5IutE0ofM1JH2MvNEIQ7EIMTUxmq7AbjCTbpjeF3tMVRmMBAnGYySEhixJGGQFm96I3WhGN0Eei0gigTMSIJyIJzOFywo2vYEskwWdJBNRE7T53GQYTBRZ08f0xx+LDpdimdowaTMYR+viXcvsebmOZ3++E587aXkJB6NEw7FJM1rPFjMtnIQmcPZ5ePW3e9n1/BEGe4cuWfxJkoTNbmHFLQv4s7+8k7L5BcjKxFULLhdN1ehuHeDpn2znyK56PE7fjF0LSZZIt1spKMvm5juXcPdH1pM5zUziPW0D/O77r9N0opNwIDp6z6izXF7rfGZSOE3LZiWEwOUK0N3lxusJomoCi8VAWXkOhUX20ZuiqbGfSDhGXl46jY392DOt1CwooLfHQ0f7INnZaSxekvzqDwWjHHq7hdolpSQSKm2tA0QicbKy06iqyseaNvfmzhQpUkwdSZIw6fSUpGVcfOXzUCSZLJOFLNOlf82OYFCUixYZPh+TTndBh3azTs/irPwJl9kMRhYZ8iZcdj3T0dhHw9H2KQvGa4FEXOXE/iYe//cXZiTvlBACnyfImy8epf5oOx/60h3c/tDNGIwza32KReK8veM0j/7Lcwx0z3zSSKEJvO4AXneAzmYHqzYunLZw8rmDnDnUSl/H4Iz370oxLeE0OOjnu996hfa2QULB6Oi0wbzqPD7x6Y0sXVaKJEns3nmGI4famFeVx763mkhPN/HAgzewb28Dzc0ObDYTX/uHB6ldWoLbHeBb33iZP/voWk6d7KKr00U4HMNkMrB23Xw+87nbsF6BtO8pUqRIkWIyrg/RlHRKVtnz8jF++1+vzMrLfaDbzS//4yUi4Rj3feIW9EbdjBgD1ITKW1uO8fi/v8hg39AM9PTCVNUWM2/x+OjgP0WmJZzsditLlpXyvg/eRFVVPrIiceZkN9/51itsefkY86rySBt29uvvG+Ke+1bwZx9bx9e++nt+++s9/OVf30VeQQY/+M4W3tx9ltqlyYsQi8V54dnDfPbPN7NmbTWqqvHs04d46fkjFBTaeejDa2b+yFOkSJEixbQxmgzYMq0TOp2fSyKeIByY2G9Fp1cuy3laIlml4HLRNMHuF4/y868/i889eT09WZGwpJmxppvRG3To9Ml6kImESjwaJ+iLEPSHJ50iC3hDPPndLSRiKg9+dtOMiKemE108+v8/h2fAf4F+y1hsJixpJgxGHYpOQZYlhEgKr0RcJRqOjU63TpaxW1Fk7vrwOnSXUPJH0cmkZZixXSSwRWja8BTeBNOjUrIgsnwZjurWdDOSPDOzV9MSTnq9wkc/vmHMb2vW1zB/wVH6+4cIhWKjwsmaZuKGG+eRl5/OgkVFdHe6WL6yHLPFQFZ2GgPnVctevrKcTZsXYxp2pLv/wVWcOd3NjjdO8573rR6NAEmRIkWKFFeONXcuoXhe7kWNTs2nuvnjj7eRmCCvWs2Kct798fWYpxg8MQ4JSuZd3jSpEIKT+5t47F+fn1Q0SZLE/OVlrLilhuolpZTNL8Bmt4xGA4aDUbyuAN0tDhrqOnj7jVN0twxMKKBC/ghP/2Q7RRU5rL93xWX1PRqJ87sfvD6paJIkiYWrK1i+voaq2hIKy3Ow56RhthrRGXSoqkYkGCUUiODq89Lf5WKg203rmR7az/bi6HajnePfVVZTwOIbKi9J7BWW5/Cpv7v/ohFxPk+QFx7fTfvZ3nHLjCYDH/7K3RSW5Ux7/yOkZViwXYaD+7lMSzipqkZ//xAH9zXT2jqAzxsmGkvQ0thPUUnmmJtFp5OxDVeFtloMZGSY0ekUFCUZahs/72GqqMwdk+smO9tGfn4GJ4514nYFKJy6Q1eKFClSpJglSqsLKK2eIEvweRhM+jF1Js8lpyCDm961BJt9Zl5k00UIQXfLAE997zU8gxOLD3uOjQcf3sTG+1eRW2RHpx//ujRbjWTlpVOxsJA1dy7lzg/dzGtP7eflX+8hHhtfdso/FOLJ722helkZ+aWXXuz97JE2GuraJ1xmsZm47xMbuPuj68ktykQ3SQ1Ms8VIZm46xZV5LGM+mqbh94Twuvz0tDnZ+8oxju1tJOANsfKWheQUXTi57WTY7FZW3Tp5SpERnL0edr9wZMJlOr3C8nU1VC8tvaQ+zDRTFk5CE+zf28gjP3idrOw0apeUUF6Rg8lkwOsZr9YlKWkmHPm3osiMvUfGKnK9XhlbkkIno9craEIQi82t932KFClSpLh+iUcTvPSrNzl1sGVCy1nZ/AI+/y/vZ/n6GmRFmlKSY71BR+n8Aj799w+QXZjB/zyyFb9nfIHoziYHz/9iF5/+2v0YzdMvv6VpghP7mggMjW9b0cm8+2Pr+chX7sFg0k9LmMmyTEZ2GhnZaZTOL+Dm25fg8wQ5/lYjpdX56K+xAtOzyZTPhM8f5uUX6zCbDfzvv72Xqup8ZFlC0wR7dtcTiYw3w01HTPu8YTRNMFLvMxKJEwrF0CkyabPsHD4z0SHStI53rrnUY7zeIxqvp/NyPR3LpXC9Hn/ysC5vjLrqjvEKd6e9vpcdzxya8J5Jy7Dwqa/dz6pbF0673REBdfdH1uHodPHqE2+N89nRVI23t5/itvfdQM3ysmlfm2g4hqPLNWGqBLPVyO0P3XxJguz845AUCXuOjVvfs/qy2roembJwioRjBIMR0mzJwpOyLKGqGp0dg7S2OCgqvryptLqj7TzoCZI7HOrY2jxAW6uT8opcMrPSLrL1pZFQtWTJgkgMlytAS4eT3v4hHE4fLk+QYChGLJYgFk+g1+swGXWYTXpMJj2ZGVaKC+wU5KVTmG8nJysNs0mPxWLAYNCNKTx6pRBCEArHCASjDLoD1Df309s3RE//EJ6hIJFInEgsgSJLmM0G0m1mCvLSKS/JpmZeHoX5dmxpJixmw6Qm97kkkVDpdUycN0cC8nLTMU3RF04IQSSawOcP0+fw0tLupKXdSb/Tiz8QIRSKoaoaJpMes8lApt1CcaGd4gI7VRV55GanYTEbsFiMV7TorhCCYChGMBTF6fJT39RPb//wNfaGiETiRGMJFEXGYtJjs5kpys+gvCSb+fPyKMzPwJZmwmya22ssRLKcjMPpm3C5JEkUF9gnTfh5bjvhSJxAMIrD6eVsUz/dvR66ez34AhHCkTjxeAK9TsFs0pOVaSUnK43K8lyqK3IpyMsg3Tb3x38xhBBoQhAOxwmHY3j9Ybr7PLR3uujtH8I9FExe32iCeCyBLEtJh2ujnnSbCXu6hcL8DIoL7FSUZZNlT8Ni1mO1GlGuhjxSVzAoLxFXee7RnRPWRFN0Cg8+vIkbNi2+rH1Y0kw89KU7ObTzLP0TROr1dw5ycNspqpaUoJumw3M0Epu0npvRZKD4Mn2/UlycKQsne6aViso89r3VyJO/3ktFZS5Op48jh9vIzrn8qt2hUJSf/mgby1aUIzTB7l1n8fvC/MVf3jmjA5oQAq8vTH1zPw3N/Zys76G904XT5b9olMiFsKWZKC/Joroyj7LiLMpLky+mDJt5zr/2NE2jq9fD6fpejpzo5GxjL70O76QRE5ORn5vOkoVFLK8tZcWSUsqKM5Gv4KDr8gT51F/+isQEidMkCf7rnz/ATSsnLwY7gj8Q4dCxdupOdlJ3sovuPs+0zo0sS+Tl2KiuyGNRTSE1VfksqM7Hnj53/hqqqtHZ4+Z0Qy9HjndQ39RP38DUrvGIF4EkQUFeBrULilixJHmNSwoz50xANDT38+W///2Eywx6hV//6H9RXDCxX4UQgiFfmFNne9h/uJUTZ7vp6Zt+wsKCvHSWLipmxZJSblheQUFexhW1HAsh8AwFaWkfpKltgMaWflo7XXT3uklMFG00BZL3azrVlbksX1zCwvmFLJpfMKVSKtcjLae6klN0E1CxsJBb7lt52WU5JEkiuyCDjfev5A8/2jZuuRDJRKIf+MK70KVdQqTYJDepqmkkYolpi7EU02PKd4fRqOfDH1uHLMObu+vZueMMRcWZ3P3u5cSiCd7Yfnr0WiqKMmY+VKdT0BuU4UKIEnq9Mu4FvPn2JWiaxkvPH8HnDVNekcOX/upOlq0ou+yDHCnS6PEG2XuwmW1vnqW718PQUAhthpK4+QMRTtX3cqq+F50iY8+wUFRgp3ZBIRvX1rCgumBWLRMjhRwHBv28su0kbx5oon/AmyzdcYk4nD4cTh97DzZTWpzFuhureODOZeRk266qr3NIDkSDrsAFlgticZW6k5088/JRzjb14fNHLmlfmiboH/DRP+Bj/+EW7BkWVi0t40uf3UzmDEVtTMSIFWLA6ePF10/w1qFm+h1eItHxTqhTaw/6HF76HF72HmyitDibDTdVcf+dy8nKtF7Ra6wJQWe3a5xwEkKgqhpHTnTy++feprndifcyqqmPXMe9B5upLMvh7s1LeNctC0mbwzpjmhDEYyrtXYPs3t/I0ROdDAz6cQ8Fp/2xM2H7mqB/wEv/gJcDh1vJyU5jcU0h7713FYvnF6LTzU5G66sRTdU4/lYjQxM4hMuKzIoNCyiuzJ2x/a2/ZznP/mzHhJGF/R2DdDc7qFlRPsGWk2Mw6CeNRoyGYzQc62DZ2vl/Mtf0SjAtWV1YZOcvv3JPsuK3SKZjH6n5dcfdS0dV7sc/tYGPfmI9+mFv/i98+Q6EYDRq7h++/r5xbVssBh58/4189BMbxrR9uddeG852vudgE3986Qj9Di/qLJcGSKgag+4Ag+4AJ8/28HZdO9/4+/dSmD/9zMpTQQhwDwXZ/mY9z7x8hAGnf8YEIUA0lqC5bYDWdidbdpzikw+tZcPN1WQOV3C/WhhwTRwdI4Sg3+njyacPsm33mcsSk+ejagKXJ0g0liD9MgqWXgwhYNAdYOuuMzz3ah1OV2BGMzdHogmaWh20tA2wZcdpPvmhtay/sYqM9Lm3mELyZd/S5mTtDe/ULRRC0NM/xO+fP8S2XTN7HcOROGca+6hv7mfPgSY+94mNVFfkXnSq8HJQVY0+h5fTDb28+sZJzjQki0TPZkbuhKqNisUDR9p418aFfPR9N1OYn3FVPcuzhc8d5MzhtgmFjN6gcNPttTNa1DYzL53Cily6mvrHLVMTGicPNE9bOJmsRgrKspEVeUzKAIBIMMozP91BQWk2eSWXHrWX4sJMSzhJkoSik1A4fzCRxgww55sJz/9bP0F4pCBpUlaUmbtpNU2j7mQX//PCYQ4fb79kU/flIIRgUU0h+bnTKwUxnfab25w8+cwB3jzQNKvHqAmBc9DPjx/fyeFjHTz88Q2UFmXN2v6my0QWp5Hz89hTe9l/uOWypmMnQ5IkVi8vn7WXrCYEjS0OnvjjAd461DKrxVM1kbROPPKLHRw+3sHDH9tAUf6lhSFfVj80QUfP2BISbZ0ufvz4To6c6JgRS8xk+z10rB2PN8TnP7GRG1aUz4pPUCAY5aWtx9n1VgPN7c5x6VnmglA4xstbT+AY8PHZj97Cgur86/5FO+Ty014/Pk8QJMPmZzLcXZIkTBYDRZU5EwonIQStp3um3a4sS9SsKMOSZiTgHWttFQKO7Wng5//yHB/7m3uoXFR8yf1PMTnX5SR30vE3zstbT/DE0wcZ8oVm5YU5FXQ6mXs2L5mVaQ9V1ThyvINHHttJV6971l4m5xOOxNm9v5G+gSG++KlNLFtcMqtf5lPFeZ7FSQhBY6uD//rxVprbBmbtHjDoZW5aWTErbauqxv7Drfz017vp7vXMqCXxQoTCMXbuqaff4eULn7qV2gVFc36NHU4fwVAUi9lAS7uTH/7iDY6f6p4Tv+LmtgG+899b+aev3kftgqIZFxQ6nczufY2cneCFOpcIAYePdeDxhviHr9xLRWn2dSuehBAM9g0x0OOZcHlxVd5lR6Odj06nkH6BjNmObheJuDpprqXJWLGhhrKaQs4cah23LB5LsP/1E7Se7mbTg6t51wduJrfIPu30BCkm58q/7WaYEefvx57cy6NP7MXjvXKiCWDJwmKqKmZuznyEeFxl36EWvvHDLXR0u+ZMNI0ghKCh2cF//uh19h1quSoKfro9QRLD1hghoLVjkG/96HUaWxyzen5qqgvInoXIz3hc5c39Tfznj16js8c9Z6JpBE0ITjf08s1HXuPQsbkv6urzh3G6klPeP3psJ8fmSDSN0O/08b2fbaenf+brgBkNOu69Y9mVjsoHkte5qXWAHz++E6//0v3FrnaEgLb63knLopRW5c94YICskzFbJ5/Cj4RiBLzj8zFdDJPFyAe/eDtW28RtC03Q3+ni9z/cyv/3/u/z+Dde5O3tp/BewA80xdS54han7BwbX/+3D1BRmXvZVhkhBPGEymO/e4vX3jhFdILMrRdDkoZzWJD0szLodRj0CglVIxpNoA07YQsBCHHBgVyWJW5ZMx/jDEevaJrgyIkO/vuXu3BPkHz0Qv1JsxopL8mmIC8dm9WE3qAjkVCHw7l9dPV6GBoKDh/nxdvs7R/iJ7/ajdViZOXS0iv6RRONJQgEImSkm3E4fTz21F5a2p0X3Obc6y0rMga9Dr1OIRqLE4uryWssQDD5+Vi1tHzGo1g0TePAkVZ+8uvd03J+lmVpOMIzm/xcG7Y0E3q9jkRcxR+MMDDop7PbPWyFndo17u718ONf7sJiNrB0UfGcXWN/IIpz0M+rx9qpO9V1wXVlWcJo0FFemk1Bbjr2DAtGgw4hBIFQlEFXgLbOQVyeIEK78HN7Lm0dgzy/5Rif+fB6zDNojZAkiZtXVVJWkkXHFKraj9ynsixhTzcnU6Bkp5FuM2E06JEliWgsgWcoSJ/DS2ePe3S8mirHTnexddcZ3n/fqqsjZcFMIwR9bZOPB3V76vnXz/5iRsWspgk6L2BVVBPJWnHTRZIkVt6ygDs/vJaXfzVxlvIRPE4/L/1yN2++eITKRcXcsHkxa+9aRl5x5jlJqq8GCX/tcMWFk9lsYP0tC2akrXAkzhPPHOTV7Sen5TNgNOjIzbFRXGBnQXUBC6ryKczPICcrDYNeGQ39VFWNYCiKw+mns8dFa8cg7Z0uBgZ9eIZCBMPRMS+i8pJsVtSWzvg0XWuHk+/9bDt9Du9F11UUmcK8dFYuLWPzLQuprsjFaEyWQhgRDEkNKBCaIKFqtLQ72Xeohbfr2ujsdicFxAXo7vPwo8d38NUv3MnCmsIrlsMqFksw5AtjMOj4zR/2s+/tlglfHIoikZttoyAvg2WLi5k/L5/Sokyys9LQDwsgIQSxWAKnK0B7l4vm9gEamx04BpM5viLDjslpViPLFhWjzPA1bmh28INH32BgknIQ56LTyRTmZbB6eTm3bVjAvPJcjAbdpNc4ntBobh9g39vJa9zV67no89LR5eJHj+3kb754BzXz5sYXZsgb5NEn99DZ7Z7Q2iXLEtmZVpYtLuHWtTUsXVyM1WJEliQkWRp9AQoxHLkWV2luG2DnWw0cONyKw+m7qLBIqBrbdp9h7ep5rFo2/WSFFyLLbuHuzbU89tRbE/om6vUKWfZk3qnahYUsrE6mEci0JyMez72+MHyNtWTkZTAY5dCxdnbvb+RUfe+UxHcsprLljVOsXlbOvPKc6+5lKkSyrMdk9LQ66Wm98IfWTKOqYkJH9algMOn5yF/fjcli4JVf78V3gY9oIWBoMEDdngZO7G/iqe9uYeXGhWy4dwXzl5WRW5yZygw+Da6bM6WqGjvfauDFLcemLJp0isyKpaWsvaGKVUvLKC3OxDBBPaJzsaWZKMjLYHltCZB8WXf3eWhpd9LWMcjRk500tw0QT6gsqimkrDhzRgegQXeAnz+xZ0qiyWoxcP//Y++9w+O4rjzt91ZV54BGzhkgGMAIZooUSUlUDpYsy5ZtybacJuzM2Lvf7s6uZ9cTNnpmdsfjHdmecZTHkmzZEpWoQIkSJZFizgGMAAgQOXWjc1fd748GKZEEuhskGgQovM9DPmR3dfetuhXOPeF3NszntptnU1Wek1KOigWYN7uEebNLuO/2+bz57lFe2XyIjq6RhQovcPpsNz99+gO+86278aSxJD8RkajOoC/IkcbzvP7O0SseigIoKcpk3U0zWd5QSXVFLlbL6HF/h91CpsfBjOp8NjCbUChKy/k+jp/s4ODRVvYcbCE/1zXuFUndvT5+/NTWlIwmp8PCA3cs4JY1s6goy07JU2CxwII5pcyfXcK9t89j87vHeGXzoaS/13i6g5/9ehvf+fZdOBOEH8aLaMzg+CirdZNJ5ZbVM7l93Rzmzy69pM/laFjMGgvqS5k/p4SDN9Xx25f2sH3XGaIj6IJ9nP6BAG++e5T59aVo6vjNs6apLF1YyetbjtJ0rvfi6x63jfpZ5LIqmAAAIABJREFUxSysL2VmbSFV5TnYbeYxnWNWi4k71tezdmUdH+w6xW827ub4qY6kHsamc73sP3wufi7dYIYTSAZ6JlmoSsqrTiMQQuDMsPO5P72DoopcXvrZVk4dOpf0+/SYgd8X4v1X9rP99YNUziqmYe1MGm6ezewlleNaVXijMibDSUqJLxzmzROn2XLqDN5QGLfVwgP1s7l1RjUx3eClo8f5oKmFLt8QLouFO2fVcsfMGWiKgm4Y7G/rYOORY7QMDGBRNeoL8vjS0kW4LBZ0w+BwRxfPHTxCc18/OQ47n10wj4bSoqQPhLaOAX75m+14h5Jr8yhCUFTo4fMPLmXFkmoyXLarTnw1mzWqynOpKs8lGtX51F0LaW3vZ+e+JhrmlWFJUck6FaJRndfePszegy1Jty0pyuSJR29i1ZJqrNarG0NRgYfPPbiUZYsq+dsn3+Tkma5R81wksPtAMy9s2s8XH15+XZLFI5EYO/ac5f0dJ68wnq1WE3esm8On7lpISWHmVWnXWK0mZlTlU1uZx/qbZtLd68PrC5I3DgKwH9+Hl944yMFjiattBFBWksXXvrCaZYsqr+o8E0JQWpTF5x9axtKFFfztk29yprln9DmWsGPfWV56/SCfeWDxdQvnuJxWvvK5VWxYO/uq9JaEEMybXUJZSTb/7HyPV986lPRh8/7OU3zl0ZvGda4BKstyWLqoks5uL7nZLtbfNJM1K2rJzY6HWa/VW221mli3qo7qijz+7483J7136LrBex+e5O7b5qJabqxwnQSC/vD1Hsa4YzJrrPvUEuauqOXt53byyq8+YLDHl5InS48ZnDp0jrNH23jz2R3UL6/hnsdvomZuGVb72Iz1TxLqd7/73UTvX/KmISV/9+4HvHK0kVtqq7m5uoIMq5Uit4sSTwZRQ+ftU2coz/SwuqqCsK7zi137WFVRRo7DQa8/wLdf3ERNTjZ3zZpBUYabgVCI5eWlqEJwuLOLv3lzC9U5WdxeV4suJT/8cBd1ebmUZIxezh8KR/n+P7/NoSQPG4i799eumsG3v3kbSxZWjGs7EVVVcNgtFOS5aZhfTlGBZ9xOPCklx0918OTP38GXxDgsLcrkP/zx7SxvqLpmcTtVVcjOcrJobhnNrb10dHtHXbVKGV+xzpxRSEGee9wvOn8gzO9e3jvqQy4ciXHwaCsDl4Ul8nNdfPXzq3ns4eVkeuzD+mBXPzYhBGazhsdtIz/Xfc3fdwEpJYeOt/HPT73HUJIbfGVZDv/+j25n8YKKa/59VVXIzXYxf04pZ1t66Owe3fMkpaS5tZfZM4rIy3Fd0+92dnvZ9NbhMX0mO9PBHz+xnjvXz0noLUyGEAKrRWP+nBJON3XT2j56CAfi7X6KCz3MrC28qt8bDUVRKCrwUFddwNe+uJrlDZVkeRxYx7ECSghBhtvGvFnF7NrXlDQBvG/Az53r63Fdoy7Z+aZutr64d0T5jPK6QpbeWo/lKhd1V4OhS57/8Rb8kygBPiPLyc33N5BxjcUliiJwuKzUL6th5R3zcHrshPxhfAOBlORLpJQE/WGaG9vZ+uJezp/pJis/g4xs54R5oAK+ENtfP0hHS+8V75ktJm59eBlZadJCHIW/HO2NMXmcOnxDvHPqLH+wcikP1M+6YsVpVlX+aNWyixd8fUE+O5tbaezqoS4vXlmmS4MMm5WlZSXYTZdeNG80nqQow81XlzXgtlpZU1XOjpZW3jl1hoaSIrQRVrhSSvYdPse+Q8m9MELAulV1/OGX15KbPb4rx0t/Z/ytdAm8/OZBupNURdhtZp74/E3Mm10yrjfekqJMvv7FNfzV372c8CEz4A2y6a3D1FbmTaj68mhkuG188/GbWb2sdty9YOO9b4YhefmNQ/QmSfh32C187Qs3MWfm+JXJCyEoL83mG4+t4bvfe2nUHnIQb33z2pYjVFfkjjmEdC0oQvDwfYu5eeWMcUnGFyJeLPHpexZxprk7YahSNyQHj7Zx3+3zx73tUElhJiWF19brMxlCCAryM3j4vgZ+8JMtCQtnojGdk2c60ybYez0RiRbJAsQE1zmO57UjhAABheU5fPZPNrD6noUc2XGa91/dz4EPTqScSxUORXn3xT007m/mlk8v5cFvrJ/2Pl3GmAyn5v4BzKrKjNyRcykC0Sibjp1kZ8s5+oIhApEIZ/r6CA/nEGTabfzZ6pX8et8B3j55hhUVpdw9q46anCwkcKyzhyMdnTz+9O8unsC+SJhZ+blEdX1Ew8nrC7Jp86ErvAyXoyiCVUuq+cZja8hJU9PgdNJ4qoOt208m3EZTFR64YwErl1Qn3O5qEEIwoyafrzy6iv/5/U2jJoxLKdm28xS3rJ7JskXJ+8alE4tZ42tfWM3alXWTQmcqGUdPtPPBzlMJtzFpKp++ZxFLF1aO+41MEYJZtYV8+XMr+bt/enPU3B/DkGzddoLb1sxiQf34CQYmY/niKh64Y8G4VqkKIZg/p4SGeeW8tuVwwhygtvZ+evv9aV10pRNFCJYurGRm7TEOHGkddTspJY2nO1mzYsYEjm4CEPGE6tH4w795mBW3z41vOEGompJQ5+lqURSF0pp8iitzWXP/IlpPdfLOxj3sfec43e39BIZCCRstSwntzT088/3XOXvsPF/7r58ib5zzdacyY7oDWTUNQxpE9JFvqG80nuInO3fzjRVLmFdYQDim851Nmz/6MUXhtroalpeX8mHLOV5vPMV3Nm3m7++/k0K3C6umsay8lMcWL8CkfLSizLTbMI+gKB4PG/SxNwVvU0Gum8cfWUlB3tRbRcV0g81bj+EPJA7flJdmc8ct9VjHMa/q4yhCsGppNcsaqnjvw9GNOJ8/zKa3DrN4/viX6aeKEIJ1q+q4bc2sKWE0xWI6b2w5krSNSHVFLhvWzU5bg1ZFEaxZXss7H5zgwz1XiutdwDsUYtNbh5k/Z/w8m4nIznTw0D2LsNnG/9y2WEysXFLN2x8cJ5yg75/XF5dzmKqGkxCC/Fw3c+qKOHysbdTWUxd6GN5oCMA+iu4RgGZSyR7H9IrJgKIq2J1WZiwop3Z+Gb1fH+TAByc4sO0Ee989Tm9H4nmORXW2v34QVVP45l9/Gk+284Y6PlfLmJ4oNTnx3jcfnG3BFwoTjsUIRaOEY/GbzameXgrdLtZVV1GS4aYvEGAg+JEnKKrreEMhTJrKmqoKvrJkEUORMGd6+xDAysoy2r0+MqxW6vJyqMvLoTonixyHfUQPl5TxpM1kzVqFgAfvXkRNZd5YdnfS0N45wOFjbQkTWC888EqL0uvyt1pM3L52Nu4ENyCA3fub6EyhKixdFBVkcM+GeVedGD/RtJ7v58iJ9oQik6qqcPPKGRQVpLcFit1m4babZ+F0jNxI9AI79p5NGjoeLxbWlzKrtjBtN+15s4uTLjj8gTD9VyFWOJlQFEFdTQHmJPva2ze+vRAnBUKQnSD82N/tu65iyelGCEFOoYf1Dy3hG3/5EP/lp1/j89++k7ySTNQEVamGbrD99YNseur9UcVDP2mMadnqslj49s2r+OG2nRxs7yTHYWcoHKGhpIgvL13EzdWVbGtq4a/e3ILLYiGsx3CYPxKNa+4f4M9feYNCtwunxUKH10dFViZzC/MBuGvmDE739vEXmzZT6slAAr5whC82zGdV5ZWNEEPhKO980Jh03DOq89mwdvZ17fZ+tcjhPmVnR0iY+zgOu4Xb181Ju3dFCEH9zGLqagrYta9p1O38wQhvbT3GY59ZkdbxjISiCFYsrkpLq4x0IKXk6Il2zrUlFkJ0u6zcdvOVuYXjjaIIFtaXUlORx/4jowtP+oZCbPmgkc8+sCSt47HbzKxeMQOHfXzbYXycDLedksLMhHpHwXA0aWHGVKC8JAuzSSWYILshEo0RierjLt57PRECCsqyR32/vak7vhqfFHru6UMIgcNlo3ZeGTVzy7j7sZvY9KsP2PrSPlpOtI9oPEbDMTb+9F1W37uI0pr8iR/0JGPMTX7XVVdRnunhZHcvwVgMh9lEfUH8QDaUFPGd29bR0j+AWVOZW5BP15CfLHtc16c4w82frF5Bjz+AISWZNhv1BXlkWK0IIYZzoFZwuKOLdq8PVVEocruYnT+yp+jU2S76BxKvAFVVYf2qmTjT2Lk+nURjOsdPdRIKJw7hzKkrJHecS6VHI9Njp76uiL0HW0at2DAMyaFjbQRDEWzW9D3wRsJsUlm3auaUCNFBXIKg8XRnUqX7ubOK09LaZSSys5zMrivk4LHWUT2dMd3g8PE2QuEFaQsPQ1x+YEF9elXphYDiQg9HGkduAAtxgchgMIqUckoY5KOR6XGgJbk2dEMSi6XRcLoOh08IQfmMglHfbzqe2ON7oyGEQAjIzHXzyL/ZwJJb6vnXv3uVHZtHrnT1DwZ5d+MePv/tO6f0+T8ejPmq0FSFGbk5zMjNueI9VVFoKCmioaTo4mtlmR+FFWwmE6urKhJ+v9tqZWVFWdJxXGhOmUy8riDPzbw5JeOu7DxRRCIx9qeQw7VicfWE6eoIIViysJJnN+4mkKBdwPmOAVpa+6irGf1mlQ5Ki7Kom0KrokAoyqGjoyfrXmDVkuoJU2UXQrC8oYrnXt5LJIFBd66tn9bz/WkNg8+pK8LjtqXt+y+Qk53YKI23dBp7G6fJhj0FqQN5DcKMKXGd7JPS2gJsTgvBoSvzRdubuuk+309B2ZXPthsdVVOpmVvCN//6IRCwc/ORK4xIXTc4tucsfm8Q53USOZ4sTI0l+QhEYzFOne1KenGXlWRRWTZ12wcM+cOcbelJuI3TYaG6InfcG1QmorYyN2kOTG+/n3MJWhyki4YF5WgjFBNMVnxDoaT9ylxOK1UVE5ujN6M6P2l4rLvXx/k0NMH9OHNnF6f1+y9gT6EXXSxmTHmvRFz3K5Utp/Z+Xo4Qgqw8NxV1RSO+Hw5F2f3OsQke1eRBCEFBWQ73feVmMkZZRPR3++i9AQsHxsqUNZz6+wP0DSbWu1EUQW1lHrYpkiA8Ei1tyXvFFeS6yXDbJtQ4NJk1qsoTr8yCoQgdXYMpCbCNJzNrCibUiLxWWlr7knpOi/IzcE2wLpbFrFGZZPUd793oxTDSN8czqibGe5iKsW0YqTVGniaxZpKuG9cl0Tgz103NvNIRQ4WxqM7ut48ylCT940Zn9uIqMnNHFpwOByNpU1+Phw5HPmekvPqefulgyhpOXb0+BpJUuGiayuwpkiA8GicTdPO+QG6O65pVfseKAKqTeECkjBsFiUq8xxu71URRfnqrzsabk2c7k26Tn+fGYU/s4RtvhIDq8tyE21xQi09m3F8tLqeF7EzHhFzDU/cuMTmxWM2MtoIJDoWJRSc+7GmyaCxeN5uMzJE9Kkd3nWHXlqMYE7zYm0xYbCacGSOHxg1Dpu3YqJqCNkpOnTQkQ5OoonXKlkx4fUH8gdHzawBUJd6LayrT2ZXcLerJsGOf4ARsIKWy+O6+ISKxGHYmZnxutw2LRZtSxnKyBsrAxTYcE4oQFBYk1z3r6vERi+mQhgTxrEwnJtPUCbtOFFJKAsEIXl+QIX+YYChKOBIjFI4SjerxPzH9in9HojqRSCxpS5/xwJVpH/U67OkYIOAPM9F3ZyEEc5ZWUVZXwKHtV4rN+gYCvPSzrcxdXkNO4dRagI0X0XCMoVGqS80WLW4QpwGTxYRtlNSAWEznfFPilJWJZEoaTlLK4ZtFYsPJYtHIn6BKs3SRrGM9xPNfLJaJn8pUGp72DwSIxSZu9eZyWtMmDpkuUpljt9OKeYINCAHkpSD22DfgJ5amVajbaU1aAXYjIqVESkkoHCMUitI34Od0Uzdnmns419ZLR5eXQDBCTDfQdQPdkBiGcTGUKKXEkBJpyI/+LeMrd2OCYo25hR5MZo3wCAUk55u6aTvTRXFl7oQvchwuG5/6+nqO7TlLLHKlp/T4viZ++t838gd/9WmcntGNv6tFSoke01E1dczfLWXc4yOEQEnDdSGl5PjeJgZHuSe5PHYyctJT2Wuzm8kY5ZkSjcQ4vucsGx5ZhnWCPe8jMbWeMB9jyB8iMsJJ/3EyMxxTpiR9JKSU9A0kzuNSVWU4SXviPSwZrnheVaJk2YFBf9wbMUGYTdqUqqCUUtLXn1hEUlMVHNdpjt0pzHH/gD9teWxOh2VKX8Nj5YIn6UxzD42nOjjSeJ7GUx2c7xicMINnvLDYzJTVFXB055UK9LGIzss/f4/6ZdU4XOmvmLychTfNYPXdC3ln454rcq2kIXnvpX1IQ/LZP7mdshkF42I86TGdznN9HN97FsOQrLlvUVIh0iuQcHT3WZob25m7vIaS6rxxa8IrpWSg28erT73PwAiGkxCCkpr8UfOfrhVFVSipzsdsNRG5vIOChMM7TnN4x2kabp6VuOfgBDAlDSfDkElbUwATUsKcTnTdSFgKDvEEeJvFNOHJ0EIINE3BbFYT5jCFwrH0ljVfhqYqU0roVNeNpPlBqqpgvU5zbDKpmE1qQo2pYCiatkRfs2nsq/KpxgWjNBCM8MHO07z9/nGaWnro7PaO2hZlKiAUwfyVtSMaTgAHt53g5Z9t5b4n1o65ieyFY3a154bFZub+r67l7LE2mo63X/F+LKrz3sv7aD3dxf1fXcvyDXOxOy0JE5gvH9sFSQdv3xCN+5rZ884xTuxvpvV0F0tvq2fV3QvGPG6JpOnYef7lb16guDKXmYsqaFg3mwWrZmB1WOJyJWJsx+WCd7Ov08tP//tGtr9xaMR7tsmisezW+rQtZIQQ1M4vxe6yXmk4AT3t/Tz7/Tcoqc4nvzTr4mdS5VrPmY8zJQ0nKWVSbxMQzwmZwvfcaExPuspUhh9u1wNFEZi0xIZTJDqxhlN8JTJ1Jj0S1ZMaHReO8/VAUeIGcjhBVDwS1UnXFKvqjW04SSkZ9AXZtus0z720h+bWPqKTqHroWhBCsPTWejb9atuIHoxIOMav/+F1zhxt454vraGwPAeHy4rJYkJRBIYh42HIqE40EiMaiREJRQkHI4RDUYorc3FdZYNcIQS188r40n+8j+//h2foG6HEXo8ZnDp0jn/4/57m2e+/wfIN9cxfNYPMXDdWuxnNpCIUBUM30HUdPWYQjcQIeEPxUOTZbo7tOcvZo21EwlFiKVzrqRIJRTl77DxNx9vZ/NuduDLtzFpcxdzlNVTXl+Dy2LE7rVjtZkwWE5qmoihxz7GuG8SiOqFAhOBQiN7OQXa+dYStL+6lp31g1DHWL6umYe2scRn/aNTOK6N6Tgl7RpCFkBKO7DrNf338hzz8R7cyc2EFGdkuLDZT3Os2bKjGojqxqE40EiUSihEJRwkFIjhcVgrKclC1T6zhRNLybYiHbaYysZiBTBIBEYLrFsoQQiQV3bygeyPlqAU2n2hiMSOpWk48n+H6HLzU5lhPm7aRoogb9rwxDMnJM5385sXdvLvtRNoqE68XQghKqvNZcsscNv92x4gP5EgoytaX9rH3vUaq55SQV5yJw21D1ZSLhkgoGCHoC+H3BRkaDOLtG8I3GOA//+gJFq+bfdXjUxTB0lvn8Plv38lT33uZgZ6RQ+axSIzW050892Qnv//xFjKynHhynNgcFhRVJRaNP5wjoSgBX4jBviH0CcrrlFISjcTo6/TywSv7+eCV/ZitJnIKPOQUecjIduJw27DazKiaiqEbRKMxQv4wAz1D9HYM0Ha2e0QPz8fJL8nikT/egCnN+aOapnLP46s5uO0k0RG83FJCy4kO/uHfPU1pTT5ltQW4Mh1YrCaMYcmCcDBCKBC5eL4MDQYY7PGx8s75fOO7DyVs9JzyOK/5G64DElK8UU9dNzek9tCQMKEenct/PNk8KIqYtpgSkEpY8YLheV1IZY5TCF9Mcym6brD7QDPf/+e3aOsYuKprWBECVVWG/4jh+4VAGU4ctllMWCwmbFYTVqsJq0Ubbn8kee/DU0lb/IwHDpeVex6/iSM7T3M+gbTK0ECAAx+cSPl7hUj1GZCcWx9eisVq4qnvvUJnEiFaQzfo7/bS3528EvZ6EQlFOd/Uzfmm5FI2qZCZ5+bRb93JrMWVE3Kdz1tZy7pPLWbzczsw9FHaPUV1zh47z9ljo7dIupzxlFGYkoaTEKQUupjqKziTpiZtsSENJlxg8gKGlEmrqeL7MG07jYbZlMIcSya0MvHjpDTHJnV6fsdALKazbfcZ/s8P36S3P3Hxx+V43DYK8jLIznJSlJ9Bfp6b/Fw3mR4HGS4bTocFu80U7zE3PCmXT01P7xB7D7ZMiOF0IST2xHce4Mnv/Jae9vSqzI8VIQQms8b6h5ZQVJHLz//XSxzbc5boBGrPTVaEgKKqPB77d3ez+p6FE5YBYXNY+Oyf3o7fG2Tb6wevi1BqMqao4ZRaXk8oHJvSTqcLcelEGFJOyA3wci4kPkaSiNiZPgHJvddCSnNsxN3rE018jo2kYXGTaWrpZl1PpJScONPFv/zqvZSNJkUR1NUUsHJxFTNrCynMzyA704HdNraE6ouIi39NCPFcpznEojH+9e830XKyY1Ldly8cw5kNFXzr7z/Pey/uY+NP3klLaxFVVahbWM6qO+dfVdhLIMgtySQr301flzdtx1Eza9x013zueXw1MxsqJ7SKTQhBYXkOX/nO/SiawrZNByYs9JkqU9JwUoRISQywP0kp/2RHUQQZSSoDdd3AHwhfl47tgWA4qSfE5bB+osrJx8qFOU7U0y92Pec4EEk6x26ndUpVMl5P/IEIP35qK03nepNuazapzJpRyKfvbaBhXjlWizbcZ27qHWtNU1l11wJmLqzguR++xc63jtDf5U2aWzMSZouG3W0bdyFGIQT5JVl86uvrWHPfIt56bicfvnGQztY+hgaDVxXqMVs0XB4HrkwHc5ZVc/N9i6icXYTNcZXXjIDFa2fzt89/i/df2c+2TQfo6RhgaCBAOBi5+pC+AKvNjCfHRc28Uu76wipmNVRhsSVvCJ0uCstz+JP//TmW3VrPSz/fStuZLvze0JhDtIoqsDttuDyOcTMAp6bhpAhcTismTU24Gh7wBojpxpR+cOcmESCUUuL3h9F1A22CK68GBkdWl/04GRm2CR/XVCPZHBuGxB8IYxgSdYKTxPtTaHPg8dg/kSKVV8O7205w4PC5pNtZzBoP39fAQ3cvIjsrPYKDE42qKuSVZPGN7z7IXV9YxcHtpzh5oIWOlh562gfwDQQIBSNxcUhFQTOpWB0WHG4bGVlOsvLdZBd4KK7Mpagyl5q5peM+RiEEmkklvzSLR791B/c8fhNHd5/l9JFWWk910tHSy2DvEN5+P5FQvFJO1RRMFg2L1YTVbsGT4yKn0ENmnpviqlyq55RQXleEM+Pa+4leGF9heQ6f/oNbuOfx1bSc7ODMkVbam3roOt9PX8cgvgE/Q4NBQoEwkVCMWCxe3SyEQNUULFYzDrcVd6aT7IIMiipzKa0pYNbiSkqq8ybFPVsIgdNt45ZPL2Xx+tnsf/8EJw+00NzYTvf5fgZ7h/D7ghd72Gmaismi4cyw48ywkZnrJqfQQ25RJqW1+dTUl2IZp+4LU9JwAnDaLdisJqJDoxtO4XCM7l4fJYWZEziy8SUVde5BX5BgKIrLObEne0cq7WDcdkza9EM1ESkpsA8GCYWjE9qvTkroTCEJNjPDPqUXJxPFgDfAm+8eTarNZDapfOa+xXz+oaXY0zDf0pDI6xgrUzWViplFlNcVEvSHGRoMEhwKfVSyP+xZVYaNpwtGic1hweawopknLvzvznKyfMNclt46h4AvhN8bJByMEg5F0GMGhmGgKAqKKlBVNW7s2c3YLkgBmNMXxhZCYHNYqFtQzoz5ZRi6QdAfJugPEwlFiYRjxKI6uq5fVJAHUBQlbuyZNSw2M7Zh41SbxGkVGVlObr5vESvvmIdvIEDAFyIcihANxy42GP/4fpktJqwOMzaHFYvNhJKkMnisTFnDKcNtw+Gw4B0KjbqNbhicbemZ0oZTdWXiRroAvf1+/IHwhDf6bW5LXIECUFSQMeVaoEw0qcxxT98QwdAEG05IWlKY45LCzOl+cilw+Ph5zrQkrnQSAm5aVsvD9zdgs6WnJ1gkql+/Ks2PIYTA7rRin+D71tWgKMqwJ8N+vYcyInFPkjqpxzgemMwaWXlusvLSo16eKlN2mZid6SQjiVS/HjM4cbozbRozE0FVWU7SWHhXjw/f0Nhjv9eClHA6SbmrogiK8j1TXk8r3VSX5yStSuvsGryY5zRRSAlnUpjj4gLPpHDtT2ZiusGZpm68vtEXegA2q5l7bpuLxz3+PdIuEA/7Tq5k22mmmUpMWcMpJ9tJVhLVWN2QHD/ZQWCEJpNThUyPncL8xB3qu3p8dPUmbxQ7nvQPBjifpLTY6bBQmJ8xnTichCyPg4K8xHPc0eWlpy9xT7vxpq/fz/kklUUZLhv5ee6kkgqfdMLhKE3nepPqNZWXZDG/fvxzdz5Od6/vuslbTDPNjcCUNZwsZo2aitykD+VzbX2cae6ZoFGNPxazxvzZJQm30XWD/YdbJ1QI8+DRVkLhxBUxOVlOykqyJmhEUxebzcycuqKE28R0g/2Hz01oiGX/4XNJ23/k5bqmdCh8oghHYinliy2oL017ov2N1NZlmmmuB1PWcAJYOLcsaYigs8fH4eNtSUX8Jismk8acmcVJk2937WuasFWkPvwQTyYwWlTgoSjfMyFjmspYLRqzZxQmbW2yY89Z9AkKscR0nb2Hmonpiee4pDCL/DR1S7+R0HUDvz+cdLvS4vQuNELhKGdbelJqWTXNNNOMzJQ2nObMLCI7WbhON3jt7SP0jVGhd7IgBNRV51NSlHhVf66tj32HWyZkTM2tvRw61pow30ZVFVY0VE3nvqSAEIJZMwqThmSbzvVw8EjrhIzpTFMPRxvbE3q4TCaVZQ2V03OcAoaRmlBtOgs8pIROhSiYAAAgAElEQVTW8/2cPNOZtt+YZppPAlPacDKbVG5aVpN0u7MtPWzeemW35amAEIKykixm1RQkTCCOxXTeeOcowasQlBsLF7xNLW2jCzZC/AGwvKFquhVHCgghqCrPpbYqcXVdOKKzeeuxpCHSayUW09lzsJm2jsQ5bB63naULK6bnOAUEqfUlTKcnyJAGh4630Xp+crU9mWaaqcaUNpyEEKxdWYc7hW7Hz7+6j0PH2q5fQ9xrwGoxcfPKOqwJlHIlsO9QC3sPNmOkMRGmfyDAi68fIJJk9bx6WQ2ZSbyB03yEzWpi3U0z4z3GRkFKya79TRw8ktjbd6109w7x6uZDSfNg1qyoxeO+cUufxxNFVbBakovvDaYgKnu1DHqDvPDq/k98mM6QOmE9SFgPENYDU7rqOt3oMnbxOE0fq4+Y8oZTeUkWyxZVJi/n7vbyz09tpaW1d0pO/tKFFcydWZxwm95+P799cTe9fUNp2cdIJMYzG3clTbbPyXJy+7o5qNPVdGNi5eIqZs0oTLhNV4+PZzfuZiAFRe+rIRKJ8fTzO2lO0iU+L8fFHevmTHubUsSkqWR6khuZJ9IURguFozzzwi6azk3dQpnxojXQyKb2H/Ji2z/wUts/EjHSZ6xOdc4M7efl8/+PF9v+gdfaf3xdhVMnE1PacIJ4SGj9TTOTajoBHDrWxg9+toXefv+EGU9SynH5LU1TePDuhUlXrQeOtvLM87vQdWPc9lFKia4bbP3wBK9vOZJwWyFg2aJKqityJ60K7WTFZFJ58K6FWCyJda/2Hmrhty/uIRbTx32O33rvOG8lCWsrQrBqSTVlJVnTc5wiZrOakkL84WNthMYx3C6lJBbTeWfbCTa9dXhSCF9eT6Q0aPIf4vDAVo55t3PMu52YTG/oe6piSIMT3p0cGXyfY97tNPp2ThtOw0x5w0kIwbJFlaxckjzXSTcku/Y18Rf/4wV2H2gmHImlzYCSUuIbCtHU0stQCtU0qTB/Tgl3rJ+TsMJO1yUvvXGAnz+7Ha9vfEQxY7rBlg8aefLn7zLoTbw6Kyrw8Km7FmJPk+rxjU7D/HJuXTMrYT6Mrhv87tW9/Oq5HQz5x0cUMxozeP2do/z4V1vxJTlfy0qyuPf2+SmFnqaJYzGbqCzLTVod29nj4/V3jqKPUxVwJKrzxrtH+aefbkl67X4SCOpDdIROY/DJDlemgi/aQ3e4BaaNpSuY8oYTxJv7ff6hpSlpBkkJR06087/+8TV+8cw2zrX1j5uKrpSSSCTGidOdvLBpP//nR5v58//+fFKF7VQQQuCwW3jgzgWUJ9nPUDjGb1/czZM/f4f2JAKGyfD6gmzctJ9//Je36e5NLMAohOBTdy2ktip/2hNxFQgRb1794F2LKC5ILOMQDMZDLz/65VY6e65N/HTAG+D3L+/lyZ+9Q29f4upTRRE8dM8iqivypud4DCiKoLI8h6wk4bpIJMaLr++/5o4HUkr6+v08+8IufvCTLQxMG00A+GODdIUmpvp4qtMf7aQ/Ml2BORI3TC+MokIPjz28POWbRFePj2de2MU7206wcG4pt66ZRWV5DmaThklTUVWBEOKSh8OFG5luSAzdIKYbcX2WQJgTZ7o42nie/YfP0dM3xKAvSDgcQwjG1atVUZrDE4/exH/7v68mVEQPhqK8tuUIJ850cd/t81m9rAaXy4pJS9zI8ULIJhSOcezEeZ5+fhdHGs8nrdZTlHii/u1rp/NerpXqilyeePQm/uc/vpawgi4QjPDK5oM0nurgvjsWsGppNU6HJaU5jukG4XCMQ8daefaF3Rw9cZ5QOHHCv6IIbl0zi1tWz5qe46ugrjqfsuLspAuQ0009/P2P3uTPvn4rtVV5SefzAlJKDEMSCkc5cKSVX/9+B42nOkeUQdBUZcpq210tUkoGIh0MTBsDSZHSoCfUylAscfX0J5UbxnBShGDdqpkM+kL881PvpVSyHdMNWtv7aW3v55XNh8jJclJbmUdFWXa8F57bhqYqKKqCYcQNimAowqA3SG+/n47OQVrbB2jvHEgqBjleKIpgxeIqvvTZlfzi2e34A6OHVQxDcupsF3//wzd5+vmdrFpaw4zqfPJzXLicVqwWE+rwvkWiMYb8YQYGA5w828Wu/U00nuxI6eYqBCysL+Urj65KqcJxmsQoimD18lraOgb419/tSGgg67qk8XQn3/t/r/P07z2sXFLDjOo88nLcuJwWLOYLc2wQieoM+UP0DQQ4eaaL3fubaDzdmVJYSAhYsqCCL39uFQ77dBj2anA6LNx921wOHD2XUKxWSknjqU7+49/8njvW17NkQQVlxZlkehyYL+tgL6UkGIrSP+Cnq8fHidOdvL/zFEcb20etnqupzGXe7BJ+/8q+cd/HyYxEctZ/cDpMlwIxGaElcBTJJ8u4TpUbxnASQmAyqdxz2zx6e4f47Ut7xlR2axgy3vOtx8cHu04DcWNMVRUUVWAMe5n0SSBnoGkqd986F68vyG827k7JaGvvHOS5l/agaSqZGTZcTutHD1VpEI3qccPJGyAYHFuyZFlxNt94bA2lRZnT4ZtxwmRSue/2+Xh9IX73yp6UVOFb2wf4zYu7L1ZwOR2WS+Y4EtHx+0MMeINj1vuqKs/l619cTVF+xvQcXyVCCJY3VFI/s5j9h88l3X7QG+TZF3ax+d2jFBV4yMp0kOG24bCZUVV1eLETYtAbon/AT3fvED19voQJ4MWFHr7x2M1kZznZ9NbhtOu+TSYkkmb/4es9jClBzIjSFmy83sOYtNwwhtMFrBaNL35mOS6XlWc37mLgGnRRDCkxYjokF/ydcJwOK48/shKB4Pev7sUfSK2RcSym0907lDRckAqKIqirKeDPvn4LdTUF0w/UccbtsvHE51cBkpfeOJhys+poTL+4CLhWVEUwc0Yh//abt01XSo4DDruFJx5dxf/8/mtJBUYv0Nvvp3e484EQcQNMEE/ZHYsuXX6ui3/z1fUsnl/OgDdIZXkuRxvPX8VeTE0GIh30Rtqu9zCmBD3h1umQZgJuOMPpQhL1w/c1kJvt4qnfbk+qSTMVESLeAPiLn1lOVqaDX/9+J9291/6gHMvvr15ey5ceWUlVec70AzUNCBEXP/3y51aRm+3i18/vnNDWQUII1t5Ux2MPr6CiNHt6jscBIQRz6or48udW8YOfvD3mpG0pry5nsrwkm28+voYVDVUIIbDbzNRUpNdwihphBiKdDEa7GYr1EzFCxIwIBgaaMKEJM1bVgUPz4DJl4TblYFbSE+qPGVFOD+0jZkyuVbCUkqDuoz/SiTfaTUD3ETYC6EY8P1YVJsyKDbvmxq3lkGUuwKo603otRo0IJ327J630gCF1BqPd9IXP4431EtL96DKKQMWsWHBqmXjM+eRYitGEJS3H6oYznC5gNmncsmYmpUWZPLtxF9t2nxlXfZTJgtVi4t4N86gozebJX7zL6abucStlHg2P28adt9Tz6Xvjxuk06cVuM3P/nQsoK8niR7/cSlNLT9pDxpkZdu7ZMI8H716UtB/kNGND01TWrqpDNwye/Nk7aa14UxTB7BmFfPPxm5k9o/DiQ8Rq0agsy0HTlHFrDi6lxEDHG+3lmHcbp3178Ua7CRtBIkYQXcYwpIFEoqCgCBVNmDApVsyKFYfmodBWTbVzIUW2GiyqA4XUE+MlEkPqGOjEjAg94Va6ws10hs7SHWqhL9KBcVn4QGLwdPNfoYjU+y0KBA2ZdzIvc+1YD1G8+EbGCOo+zgWOccK3i+5QM0F9iLARIGZEiMnocG6RQEFBFRqaYsGi2LCpLsodc5idcRP51go0YR6zYRA/VsbwsTKIGEG6Q+foDjfTHjxDX+Q8veErPXNhPcAvzvx5vH9QimjCxIqcB6l1NYxpjCONOWIEORc4xoGBt+kKNQ0fsyC6EcVAR6CgChWTYsWqOsgw5TLLvZI693KcWiaKGD8RgRvWcAJQFYWZtQX8xbfvYdvu02x87QAnTnVMWGmuy2mhojSHrDQ/eMxmjUXzyvjef3mIja8d4O33j9PS2jfurVcyXDbqZxXz6INLmT2jMKXeW9OMDxazxrJFldRW5vHCa/vZ8n4j5873j7sOmcdtY96cUh59cCl1Nfko016mtGA2qWy4eTb5OW7+9fc72HfoHLFxboWSl+Ni9fJaHn9kBRku2yUPWCEEFaXZZGbYxyVsL6WkP9LOocF32d37Kn7dSyL9HwMdQ+rEZISQEfei9kbaaAkcYWfvy7hNOcxyr2Rx1p1kWRKr6UPcAGr07qDRt4OuUDN9kXaiRph4ZlPia+R88ORYdhWBYIZr6Zg+I6UkoHtpD57muHc7jd4dBPTBJGOTF49TVA8T1L0MRDtpD51iV9+rzHKvZFXOg+Ray8Zk+BnoHBzYwpmh/cMGZTu6jJHsWEkMWoPHU99pQBNmArFr640YNcK0+I/wfs9ztAaOD4915PHFpEFMjw578Tpo9h9me8/zrMp9mJnu5dg1N2Islt8o3NCGE8RvEKoquGlpDfV1RRw90cF7O06we38z3b2JEymvBkURFBd4aJhfzuL55cysLZgQr4wQgkyPgy8+vJwVi6vY+uFJ3vvwJM2tvde8jxca9q5ZUcuC+lLcTut02OY6IIQgO8vJlx5ZyYrF1bw3PMctbdceina7bKxcUs3qZTUsqC/F6UiPi3uaj1BVhYVzyygu9PDWe8d57e3DNLf2XbMx7HRYWN5Qxd23zmV2XSG2UXpcFhd6xsVwklLSGWritY4f0+o/fs1VaxKDwWgX+wc2M8O1OCXDyZA6x73bOTT47jX9drow0Hm369ccGXyfoH7tKRW6jHJkcCt9kfNsKPgKpfbZKV+vuoxxoP9tWgKJu0Bcby54mfb0vcaO3pfwxXrH/h1IBqJdvNHxE9oCjazL/zxO7do7HtzwhtMFLhgWK5dUsWheGf5AmOMnO/hw7xlOnu6kbyBAOBIlGjWIxmLoukQa8qLXRoj4SkNRBaqioGkKJk3FZNKw20yUl2RTV1PAorllFBVkYLOZsZi1CX/4qKrCjOp8qspz+fQ9DRw61sa2Xac4drIDry9IJBojEtGJ6QaGEW8HIwQoioKqxvfJYtawWU1UlGWzvCF+vLIzndispuv6MNVUhbLizIShhfwcF1oSdeapjqoqzKotoKYyl4fva+DgkVa27z4dn+OhEJFIjEhUR79kjgWKEq8SNZtUzCYNm81EVXkuyxdVsmheGVmZDqyWiZtji9lEWXFiMdfsrPTmc3wcl8uadDxul3VcNayEiHuGPnNfA3esm8OBo+d4+71GTjd34/eHCUdiRGMfzSXEq33F8FyaNCU+l1YThQUeljdUsbyhktxsV9LrNTfHxewZRSPqdxXkZqQU2pBS0h1u5tXz/0TrKFVYCiqaYkYVGgIFhkNquowRk5FRvRx5ljJKHbOTjmEqIBC4tCxC+uhGqkBBEyZUYbp47A2pE5XhEb0sEsn54ElePf9DPlf+X8gw56Zt/NeDiBFkS+e/srf/dWJypMIYgUmxoAoNBQVJPAwaMUJc7u2MGiEODrzNUKyPe4v/BKcp85o8TyLJ6mbCs8Ok1Onyv4oiTGTb1qEolrT/pmFIOrsH6ejy0j8YoK/fTyAYIRyJEYvpGIZEVRU0TcVq0XDYLbicVrI8DnKyneTnuDCbNaJGL32Bd8mwNmAzVaR93BcIRM/QH9yGMXxy2UzlZNvWX3LTDIaitLT2xquten0MDYWG989A0xQs5vh+ZWU6yM91U1KYOfyQmPY6TBUCwQgtrX109Xrp6vFdfPBeMscOC9mZzuE59uCa9h5OOqSUDPqCNLf20dE5SP9ggCF/mGg0hpRxqQqTScVpt5A5fA8qLczEk2Gf8PB5WA/yRse/sL9/8xUGkF11U+6YS761HI85H7uagaaYMKRORA/i1wfpj3QwGO2iO3SOvsj5S/rG3V/8p8zzrE/p/NRljEMD79CWIOzWFmikI3TmslcF8z3r0ZTUtckEUOdaTrVrYcqfAUlnqJnnWv43vZHWi6+qQiPLXESupYwsSwEeUz52LQOLYkMiCel+BqKdtAVO0Ow/TED3jjiiZdn3clvBl1MK2ekyys7eV+iLtI+6zdmhA/RFLi0cUIWJ+Z71iDHkCqlCoz5jDSX2upQ/AxAxQnzY8wJbu569woOpoFJgq6LYVkeBrRK7moFZsRKTYQIxLx2hs7QGGmkPnrriswKF+Z513FLwJRxaRrJhjHriTTqPk8SgzfsLVMWBx7ochfQbTooiKMz3UJifuM1FMgZCuzjd/z+o8PwpxQkMp3hyno5IMfExGTF9EG94P8FoM/7oSbKsK8i2rb9kG5vVRF1NAXU1Bdf8e9NMTuw2MzNrC5hZOznmWEqJHjMwDANpfMxzK+LaaIqqJD3/Dd0gFtURAjSThvgE5NUJIfC47Xhm25k/u2TMn9djOrFhbTfNpKFq6fHASikZjHZx3Lv9CqOp0FbNmtzPUmqfjS1BFZghDSJGAF+0j/5oJ01DBzjq3Y5ZWChz1Kc8FlWozPOsY65n7ajbbOn8VzpDTZeIOgoE6/O/iF1zp/xb8c+N9ZgKssxFVDsXMNDfgVVxUuNqYKZ7GVnmIpymLKyKfUSjREpJ0DNEs/8wW7uepjPcdPkWnB7ay8LwbeRZy5OORBUay7LvTZjPtLH1H64wnEyKhdsLvzqmfCoAZYzHSkqDU7497Op79QrDx6LYWZP3CHWuZWSYc1G4NKojpaSeeNXd0cEPeL/7N8MeqOH3MTjq3UaRrZZFWbePeV8uMOkMp6mM01xHjv12nObEF3xY7+C891eUZXwDTR3bBTsSLst86ixzCUabaez5D9f8fdNMc62EAmEOfXCCrS/s4sS+Zvo6B9FjBg63lfyyHBbfMocH/3ADZuvojYINQ/Lei3v42V8/T06hhz/63qNUzCqe9pAl4c2nt/GT7/6OSDjKv/3Bl1nzwOK0/dZZ/wGCl4Wf7GoGdxR+nRLbzKRzpQgFq+rEqjrJsZRS41zE6rxH8McGcZvGIoEhkj4ERyt0UISKKtL/KNSEiQWZt5FnraDG1YBD86CQfPEghMCuuZjpXk62pZhnWv76Co2lwUg35wLHybWUpXDMkh+r0b5DERrqVRobqTIY7eG97t/gvyypPNNUwIbCJ6hxLR51DEIIVDSyzIUsz74fp+Zhc8cvCOgf9WyNGEE+6PkdNa4GPOb8qxrjtOE0jthNVdTl/Lek2/nC++kLbqXY/Tga1244CaHEC1eFxphqRW8Q/EMh3nnzCJ5MB6vWzrzew/nEEw1H2fzMdn75PzbivSzxOOAL0t3WT25xVtIk6Egowu7Nh+lo6qazuZszh1upmFn8STzFx4ShG0TCUSKh6Lg1MB+N7tCVCuhZ5kKKbTPGbODGhT1VbKoLm3rjyZwIISiwVVJgq7zqz+dYSliWfS9vtP/0Es9ZVIbpDbeiyxiaGH0xMhU4OvgBnaGzl7ymCTMrch6gxtWQsuGmKSZmZ6yiPXiG3X2vXnK8BqPdHB7Yyk15D1/VGCe14STRiei9GDKCQKAqdlThvMKdacgoujGELuMuOVVY0RQX4opVhEQ3wuhyCGM4lq4IE4qwooq4mzSusxFAN4Ywq9nD/w4Mb2sb/t5Lfz+q96PLwPD/BCYlA1W5VIJASmN4jAH6g9uJGUNE9E4kF8ZhwaRkIoZPCikNdBnEkIHhsQoUYY7/PulN4JVSEgpGMZlV1BTCKen4fT1mEI3pWFNISO/q9PLK7/eQ4bFPG07XGSklrac62fjjt/H2DpGZ5+aWR1aw8OZZmK0m/IMBWk60M2NRJSZL4tuPqqk4PXZUTcViM+PMsE8bTZOMsHGltEu85PvGLtC4XggEFY55ZFuK6QlfarQORnuIyjAaU9dwGor1s79/8xWvF9tnMD/zljF7Bs2KjaXZd3NoYMtF2YsLHPVuY1HWBuzJc52uYNIaToYM0zn0Ip3+jYRjbYDEbWmgNOOrZFgWIYSKRBIzvHQNbaTT/yLBaDMAdlM1Bc6HyHXchfYxAyYU66Dd9zTd/teJGr1IwKRk4rLMpTbrLzCpmYCka2gjzQM/YEbOf6Pd9yy+yBEMGcJpnkVpxlfJsq3m43fwNu+v6PK/TMwYRJdBarL+E4WuRy7Zn6jRT8vAkwyEdhCMtWDIMIe6vn7xBpNhWUxt9l9iUj1IKfFHGjnn/Qne8D6ieh8gsGqF5Druocj1KCZ17JOdKv6hML99ahur1s6kdlbyUuB0sG/3WU43dvCpzy7DkiCcA5Cb5+b2exeQk3ft3rtprp1zJ9ppPdmBZlK5+8tr+dy/vQvlY5WOy+6Yn9L3xD9/MzanldziTOqX16RryNNcJVblSo26/kgHESOIRbVfhxHd2AgRr87LMV9pOIV0H7oRhfRG0tLKad9evLGeS14TKCzNugdNXF1z8UxzAeWOehp9Oy553RvtoS1wglr3kjF/56Q1nPyRE0gpKXJ9DqtaSCB2luaBH9Ay8CSzcv8ubuRISbvvWc4N/oRcxx2UZ/wRAD2BzZwd+D9IYhQ6H0EIBUNG6fK/xHnfrynL+CYO0wwMYgSjZwnGWhCXuTcjRh8tAz8k23ELRa5HCeudtPl+yZm+v8WSW4jDXHtx20LXZ8i2r2cwtIvmgR+MuD+qcJDvfIAcxwbODf6YocixuKGkxBPSNcV9iZdKlyEMGaHE/WWsWgmg0zH0AucGf4TdVEmO/fa0eIKklPR2+9j+3gnmLUqeaJgutm89QSgYSakXl9Nl5f7PjE2Qbpr0IA1JT/sAUkrMVjPV80pRtau7kwshKKkp4LE/v3+cRznNeJFjKb7itYFIJ0cG32N+5i0pK39PkzpmxYZthGT2mIxikN7QbDqJGRFaA42XJHMDeMx55Fsrr9rbLBCU2GddYTiF9CG6wi3UyMVjPkcnreGkKBbKPF8bLqtX8MjlhGJttPueJmoMoikewnoXrYM/I9u+lkrPt9GU+MnktizAkCFaBp4kx3YbZi0HKXUisU4UYSbbfgs2rQxQwHbzcIXb5V4Ng0z7KkrdT8S3Q2JSM2ns+U90+1/Dbqq+GLKzaPlYtHyiei+jibyoihWXpR4pY3QoOSjCjMs8D4uWd8W2QgjclvnMyv3e8LiG2yRopfjCBxkM7yHHvoHxjFtIQzI4EKCvd4j9u8/Sdq6XptNdF5N3TSaVkrJsnK6PekkZhsFAX4DeHh+RcAyzRSMnz40n035ZpQNEIlF6Or14B4MYUmI2qTjdNrKynZgt8cqIcDhK5/kBfL4Qez48TXFpFkcPnrs4htw8F/mFnovfPdDv5/y5vovtR9wZNsorL9UykVLS0+Wlv89PZU0+fT0++vv8GLqB3WmhsDgTi+XSuY/FdLo7vAwM+K/QjLLbzVTW5E+rpo+ClJLgUPzGpygCqz39VbHTXD/KHfVYFDthI3DxtagM83bnUwzFBpjnWYfblH3V1UvTXIkqNMwjyPQY0uA6KAiNG75YP13hZi7fhyJrDTbNdQ26S4ICa8UVrxrocYV5GcYsxtYjcdIaTlatBKd55kXjRKBh08oxZBhDhgEYCh9EEsVtXoCmfiSlrike3JYF9AXfxRc5TLa2FkWYybStpCf4Fid6vkOWbQ0e2wpc5tkoI7oABRmWxRdzji78XxEmgrEmdBlAE860HoOI3osvfIhw7Dy6DA7ne4XRR8gruFaiMZ33thzjnTcO03auj2hE5zdPbbsYJsvKdvKNP9vArLnx8mgpJft2NfH80x9y4ng7saiOZlKpn1/Kw19Yycz6j6qfwqEozz+zg62bj9Dd5b2oi1VUkskXvnYzi5dVg4Dz5/r5yQ8209kxSGfHAP19flqaei4aKXd/qoHPPLby4pjPnOzkmZ+/T1eHl+4uL0tX1vBf//dnLtkvKeGtTYd4/tkd/MG3bufF53ZxvrWfoD9CZraDO+5fyAOfWYrVZr64X2+/fpiXnttNMBBB13W62gfRDYPikiwWLq3im9/agKJc24NASsnQQIBTB1toPdWBr99PLKpjsZrJyHFSVJlHZX0JLk/idj19nYMc23WGjuZuAr4QVruZvNJsZi+pJrdkdDHHznO9vPv7nUhDsv6RFeQUehgaCHBs12laT3Xg94YwmTUKK3OZubiKvOKsEaUAfAN+Tu5vpr9zEG/fEN4+P/u3xtsyhEMRXv/V++zfeuySz+QWZ7HmgcW4sy69fqSUHN5+kt1vHb7idyxWM2s/vZSiyisXGpcjpWTHawc4tvsMtQvKWXnXQhRVIeALcWjbCVpPtuP3hbDaLWQXZFC7oJzCyjxM5vjt0Ns3xPNPbia/LJt1n16Gr9/P7rcO09/lpXxmEQ3r52C2mtB1g31bjnL6UAsmi4n5q2dSNafkkrDk5QT9YY7vPsO5E+14+/yoqoInz0XdokrKZxal5J3TdYNzJ9o5uuMU/d1eVE0lvzSb2UuryRtuxjwebSWSIYTAY86n1rWEw5cpdgd0L+92/ZozQ/upcy+jzrWUTHPhtPdpHLiQRH+jEdS9DEa7r3g9w5x3Tc2fhRBYVDuaMF2iEwYQiA0SNcJj/v5JazipIp4I/nHEheEOV+NEjX4EpuGE6Uv7MMVf0+JeIOKVZ5m2m5id+33Oe5+ifehZOoaew2mZQ4n7y7jMc69I+tYU12X/dyPQ0I3gRbHJdCClQZf/Zc55/wXDiGAzlWJSMjGIIa+xncFoaJrK6nWzWLS0in07z/D9//UqT/zxLdTPLwNAURUysz56iDef6eaf/m4ThUWZ/Ke/fhBPloP2tn5++k9v85P/9xbf+s/3UlSSiRCC0yfaeeYX73PHfQvZcM98VFVhYMDP0QOteDIdFx1nhSWZ/OG/u4NoROff/+EvmT2vhC9+bS3WYePt494ugDnzS/n3f/kp2lp6efLvX0+4fwN9AZ795Qfcft9C5i0qZ8gb4rLgWJ4AACAASURBVJmfv8/G3+yiqiafpaviodeWph5++aN3mLuwjEceW4XZorHljSP8/tcfsuHeBdz5/7P33nFyneXZ//c550yf2Z3tve9qJa16l2y5yt1xARuMMRAMJAHiBAhJCLwhgeSlvOQHyUvxSzEQDNjGuFuWi2TLTZLV26pu0fZeZ6eeOef5/TGr2V3trLZoZctkr8/HfNDM2ed5zjMz51znvq/7um9fjnoBzuRSSkL+MG8/f4Bnf/4qXS29BHxB9IiBaZoxN2ibBYfbTsXSIj7zzbsomJeTcIzXn9rD5l+9TntDN0F/GEM3UDUFm9NGVkEqN338Cq750Fqc5/QpA+hu6ePZn79KOKhTvDCP3vZ+fv71x2k62UbAFyKqG7GIkdtOeo6XD33hJq78wGq0c27sjSfbePArj9DbPoAeiRKNROPpVT0cZfsTu8cFRuevLGH5lQvGESeAUwfO8NRPtmKaMadsc7hhtdvrZOHasikRJ4B9rx3juV+8xoZblrPm+iU0HWvhV998klMHzxDwBTF0A0VTsdktVK4s4XPf+Uh8n4f6Azz2n1soqMimckUJv/n2Mxx4/TiRkI4rycGffepqPvzFm3j+l6/zxx++yEDPEIqqkFeayWf+7W5WXl01hmRKGTuP6ndq+MN/vUjt4Ub8g0H0SBQhBFa7BW+6h1WbFvGhv72JjLyUhARDSsnQQIBnfvYqr/5hFz3t/eghHTEc2UvL8XLvl29F1VSUi+TddC5sipNVqTfRHqobp7mRSBoD1bQGT7Ov90XKPStZlHwF6bZ8rIp9uBp4jkjB2CbFEgNzuBFvzGtJDt/yZPzfsT58fzqQUhI0hhjS+8a8LlCwKLbhNjUz/66EjSCqsI4jTiEjgHHOa1PBJUucpgJN8SKJYkh/vK0ExD6EqOlHYqCpI6aWirDhsS6mMv27hI0OugOv0O77I8c6/4YlWb/GaR1bJho1fef8ewiJgaLYUS5iyWfE6KK299s4LWXMy/r34bSiIBhtYCg8/ml8NqAoAm+qi+QUJ2dqYze0tHQPOfnjL+JG1GTzU/swopJPP7CJotJYeqyoNINQUOc7X3+S40ebyc1PASA0XBadlZNMbkEqDocVKSVLVxQDI54hdruF3PxUIuEoqqrgcMZSaQ5nYlGgzWbBlmEhHNKxWs//VRYCLr96AbffPSIENM0NfP1Lj1Jf28nqDeUIITh6sBHTNNl4zYL4ed10+3Kee3wP3Z2DuC7AaVtKia/Pz6//7Sm2/WEX4WAEzaqRmZ9KSkYSqkUlFAjT1zFAb8cgNoeF1KzxRQDhQITf/8fzPPXgVqK6QXpuCqWLCnB67IQDEdrOdFFf3cz/++qjNJ1u4xNfuxNXkiPhmsLBCK8/tZfTB8/Q0dhDQUU2aTleFFWhr3OQuqNNNJxo5Udf/i12p5UNtywfc/5JqW5Wb1pM0B8aPkeoO9LEqQNn0KwaK65cQFruWGPZ7ML0CdezatMiHC47QwMBfH1DHHzjBDWHGma03wA97f2c3FfPg//0KI0nWskpzqCgIhtFVRjqD9DZ1IPdaSMlwT63N3Tz3C9eo+lUG0suq6T5dDut9Z0899BrKKpg66M7ySvNpGxxASf21tNwopWnHtxK1bpyHK6xJP/NZ/fxs6/9gd6OAVIyk6hcWYI72UlUN+hp66PhRCvPP7SdM8db+fKPP0l2Ufq49ejhKL/77nM88/NXkaYkLdtLweocHC4b/sEgTafa+Mk/PsKCNWXxh8uLDSEEBc4FXJ99P6+0/5Ku8Hh7gqiM0BtpZXdPK3t7X6DQWUVl0hrynQvIshVNy7X7TwlSSsKmn/5IJwN6NwN6JwN6Fz69l6AxSNgMopshoqZOVMb+M8wIuoxM2Oj2/YwhvW+c4aXE5PXOR3i989ELHD3x78GQkeEU5/TwviZOHmsVoOCLHCNTBlCIVXEY0o8/chyBgtu6AOAcz5hYhVqe5z40JYna3m8xGD54DnGSDIb347Wvj7/iixzBlBEcWj6qmGnFiBK3PTBlJE74zq5PCEEo2oIh/Xjta3BoBfHjI9EOwkbnDOedPfgGgzSd6WGg389vf/EG2qin2/7+AFJCW3PfcEpOUFqRzeJlRfzxdzs5eayV1evLWb2hnKTkxDfPiwFFVVi6snjMTf9sFV4kHGtlIUTMdFEgUJThFLGI9QRj+L0LbcD6/C+388ojO9AjUbIK07jjL69lwZqyGHHSholT5yDNNe1k5KXiPIdgmKbJ60/t4flfbieqGyy9vJKPfPkWCubl4HTbCQUitJ/p4skHX+HNZ/fx4sNvkVOcyW2fuTphGigS1nnzmb2oqsJ9/3gba69fQlpOMkJV6O8c5JVHdvDEj18mOBTm5d/voGpdBd70kUhsQUU2f/HvI+lRwzB57AdbOHXgDHaHlTs/dx3Lr1wwpb0RQlBUmUtRZW78tf/+309Te2T8zXiq6Gjs5rfffZbBniE++fUPsOSySpJS3SiqYGggQHtDN1ablpDIhYMRDr99kr/5/scoXVTA/u3VPPiPjzLQ4+OZn77Kyk1V/PnX7sDhdvDYD17gyZ+8wrHdtfgHQ3HiJKWk/lgzD3/7WXo7BqhYVsTHv3o7xQvyxhCnl373Ns8/tJ1j79TwxI9f5v6vfwCHe4R8SSnZs/UorzyyA2lKiubn8pl/+xCli/LjxOnMsRYe+sYT7N16NBape5eCOUIIytwrsOU6eafnWU76dk/4FG9KgzP+wzT6q0mxZlPoqqIq+XKKXIv+RwjJY9ePWOPZU4O7qfMfoj/SzqDeM0Yn9j8NEjlBO5mRIy4GYmL66Y/9viZONi2HvKSP0jr4CFYlnXTXJgC6/C/RHXiFXM+92NSYM6ghfbT5/oCmpOCxLUBgIWoO0Rd8CwCHpWDc+F3+F7GqWXhsi4kYXTT2/wSL4iXduSmufTLMEFFzACl1wkYXSJOI0UtIb0YIFU1JRhGjUyUCh6WE7sDLdAw9QbrrekBBFXbsWh6gYdNyUYWbgfAeBsMH0RQvwWgdTf2/GOePIqWBbvZhyjChaFtMAyWDhKKx+WOeVolD/zNFOKQTDusgBJGITiQyMrbVqrFu4zyyc71xnbw3xck/fvNOdr99mpeeO8jDP3+d3/7idW6/ew2bblmC+13oiScEJHnHkl1FESBGLmYgqFpSAAJ2vnGSvIJULFaNt147jgDmL8q7oDRd44lWnv/ldvRIlLQcL1/95V9SvqRwXOuR/PJsqtaVD5O5sfvSWtfJlv9+g+BQmPJlhfzVt++heOGInszhtuPN8PDp7LsJDoXZ++pRtjz8JovWV1C+NIGrsIxFEO/+mxu546+uxTpKKO9027n9L67hzPEWdm05RMOJFtrqO8cQp0sd/d0+LPVdfP5797LmusVj9jojL5XiBXljotXnompdOYs2zENRBIs3VFK+vIg9Lx9BKIIb7r2MrMJ0hBBsvGMVTz74CiF/mM7GbtJzYlG2SEjn+V9sp7W+k5ySDD77nY+wYHXpmPncXicf/sJNdLf28eYz+9jxwgEuv20FSy6rjB831B/g1cd34R8MkpTq5tPfvIuVVy+MpwQdbjtpOV4sNo1vfepnDHT73lUSIoQg3zmfLEcJdb4D7Op5ls5QAyEzcVNbE4OeSAu9kVaqB96g0FnF2vTbyHPMw6Y4/yQJlClNfHoP+/te4kDfKwQN359k5Gim0M+ppruU8b4mTkKo5Hk+jsBKl38zLb7fAAKbmkVe0ifI9dw7SrekoBt9tPp+j270DxMfgUMrpCj5s3hs53rLCLLdd9Ex9DRn+v8LwwzhtBRRnPK3eGyL40f5wgdpGPh/6EYPutlPVPppGfwNXf7NKIqTgqT7x1TACSHIdN2KP3KCVt8jtPgejlX6Oa6lNPUf0IQHm5pJkfevafM9SnXXA6jCjkVJJct9OwPh/WNWqZsD1PZ+h4BeQ9T0EY52EIw2cLTzL1CEgyTbEspT/5nZNPewOyzY7RYys5L4yjc/gNN1/sopIQTJXiebbl7CxmsWcKK6lde3VvObn20HAbfdvRpVfTfErJMfU1KeyQfvXc/vHnqD6sNN2O0WdN3gpjtWcNmVk7eQmAhSSna8cJDBniGEgNs+czUVS4smFBKfjXidO0btkSZqjzahWVTW3biMwsrxglshBJkFqVx+2wqqd52m+XQ7R3edpnRxQcJ9zinO4PI/WxEXR48eJzk9ibLFhex+6TCDvX4Ge/3j/v5ShkCw6tpFrN60KGHETQhx3s+0tKogXpzgSnKQmhlL6WXmp5I2qsIzPdeLQCCRDPSOkIWW2g6q36lBSsmyKxZQtqgg4eeVnO5h8YZ5vPPSYXra+qk90sSi9fPin9fZ9CvAwrXlVCwrHifWF0JQvDCPqnXl7Hj+wHS36oIhhMAq7FQmr6PIvYi6oUMcH9jBGf+RMS0vRkMiiZghaob20Rw4QWXSOtan30GmPSZP+FOBlCZn/Id5vfMRmgMnx7hYJ4IqLDhUNzbFiaZY0YQFTbGiCgua0OgMNdKvd5x3jPcX5IxSZu8VLjniJFCZl/7vCFQ0xT3mnXTXtbisFTgshfFXLaqX/KRPkuG6kagZ+3FalGSsajbqKKW8KpwUJH+aLPcdw5ooE0VY0BQvNi1ruF3JWCTbV5Huup6I0QNILEoKdi2X0T9ol7WSkpQvImVi0XbMg2nsBcCmZlOW+jXC0XZMGUYRGhY1bVT6T5DjuYtUx2Xo8XPyYtNySHVcydnoCICmuMlPuh9TJmbrMW+o6UVJNEvsBuPzjWqOOCqV6ElyUDYvm/oXDnNgTz3rLp+HMnyBN02JYZhomhq/4ZxtOKpqCja7haUri8jNT+HgnnpOHmslqhtjIjmqpiAUQTiko+tR7NIyLp15MdHf56difg4fuHcdniQ7TpeN7BzvpEac50PIH+bM8RaiukFKZhKLN1ROu2GtETWoOdyIHo7iSXFRsbRowiosIQTlS4twe10EW3o5sbeOmz9xBapjvJ6kcF42qdnehPuqKCLm3m1R0cM6emT6Qsr3EoqmsPLqhWiWmV3qRmufVIsadzt3JTvH6JgsFg1FVTAMAz0ciyJIKelq6aO9oRuLJaZli+pRogOJowzOJAdWm4VISKe1rhMjGvtdSCnp6xygu7UPoQgKKrJISklcben2usgrzXpPOYdA4FA9LEy6jBLXUnojLZwc3E31wJv4or0TpvFCpp/D/a/SFW7kqsyPUupe9q5VCF5MSClpCZ5mc+uD9EXaxr0vULAqdrIdpRQ45pPtKCPJkoYmbGiKBQUVRSgIocYb5r7a8Rv6+/+UiJNI6AquCQtl7hWk26bf7HoqSLKkY1PPX7mcCJcecRIKbuv4thlCCKxqOlZ1vGhSVWw4leJJx7WoKcPu4FNeDXYtd5gsJcb0xzx7LqlY1cTl4rGLhRWHpYhzlRejSSMQ84OyLZzW/JMhLd1Ndl4KTz36DpqmYLNZEIqgYn4OyV4niqpw292rOXqoiZ/+58vUne6gtCKLSDhKW2sfHW0DfOi+9eQVpgHwyuZDnKnronJhLp4kB9GoQfWhJvr7/JRXZo/r3q4ogoWL8zm0r4EtT++ntCIbwzDJyUuhsCT2+eu6QWf7AIGhEG0t/QQCESwDAY4fbcaiqXiSHWRkJTHdO4iUsGdHDW6PnYH+AJGwTm/3EL3dQ+QXppKWkTQjD6ehgQD9XbFig8z8NLzpnmkTQMMwaauPadysdgtZw/s7ETLzUrEPC+vb6jpjdgcJZGUpmcm4PBOX4ypKLL0lTfluaY5nDYoQ5Fdkz/jvnaP2RYgR0m6xaqiWsaT17Mcph6sKpSnpbO4lHIxV4P7m28/w2+8+O+Fc0pTxisSh/sDIODKWcoyEdKx2C2k5KeeJVApSMpOwWDSi0YtTgTtVnG1O61AryXPMY0P6ndT7D1M98CatwdMM6j3jIi8SSWvwNM+3/Ig7C75EobPqfR946ou081LbLxKSJo+WRpV3I8u9m0iz5Q2TxJHsRCLE+tH96Qnqber4i5MiNCqT1rLUe81FmnVmX65LjjjN4b2FEIL8wnQ+8onLeOqx3XzvG89gsWqUlGXw+S/fRPKwTigjK4kH/uEmtjxzgG1bjvDYb95G01Qys5LYcOV8nO6R9F1KmpsXnt7Pti2HMQwz1qwyM4m779vANTcsRlXHR00+9PHLCIV0/vDbnUhT4nBaue/TV8aJU1/PEL968FXqTnUQDEYY6AvQ3ir41teexGpVWbaqhM/87Sas1qlHiaSU+IdC5Bem8sa2Yxw70hQnDA6nlcqqPD71+WspKc+cNunRI1EiodgN1O6yok1SBZh4gRA4ay6pKtgSRI9Gw+qwxm+wgaHQhMJ2i80yjgT8yUDEPKBmiokIilDEpBFDKSUB34jnmjRl3Kx1MphRc+TzkpJQIPbdURRx3sinEAKrLRb94j0mTmcR+60IHJqHBUkbKPespC1YwynfHg73vYo/QRpvMNrN9o7f8+Gir2GfQUTgUoEhDU76dtERqhv3nltL5abcv6DcvQpNef/2l5sNCARubXwAwpD6sJP4+VPq7zbmiNMcxsHusHDdLUtZf2UlesRAiNgTtnsUGRJCUFqRxWce2MRHP3UF0WjsOE1TcTitY+wBVq0vY+GS/LhXkRACzaLictnQLNo47ZEQgrJ52Xzlm3cSCurx6jznqCqjtHQ3n/u7Gyd8qrZaNaxWS0xPdPdqrr1pCSlpYy/AWdleHnr88zidVoQQDPYHePAHLzE4EOT7P/skmdlJwylCOFHdwg//zwu8svkQn35g07Q1WYoyUqlnGOaMq/PO7qs0JcYkN0YjasTn0azahM9WQnn/p0POhwu53l7QtVqIeDTV5rRyzxdvZu0NS6b0p+5kZ9wxH0G8clVK4t5WE+FSDgqe1UEVuarIc8xjSfJV7Ox5muMDO9DlWG+i1uBpTgzuZKn32kvqpjkdBA0fJwffGecfZBE2rsy8h3meNdN2VTelQfQi+gi+V3BqXqyKg8gog2dDRodtCkzUS8j0c444jYPAZZ1PftKfY1XPnwr5U4ZmUWPmlOeBELGn38m0P5qmkpQ8PfsGRRG43HZc7sQpJFVTSU2bmnO702VLKGBXNYWMUY2BG890U32wiQ9/4jKqlo6tslQUQXqGh8GB4IxIj81hxT5MPIf6AoT80zewUxSBd1icHNUNBnsTVyydha/Pjx6J6Wm86Z7zOlrPYfahCEFymidmc2GY2BxWSheNr96dDEIIXEmxSjMjajA0eP7OAcGhUFxXeOlCoClWshwl3JL7eTxaKju7nx7j46PLMPVDh1mYfPm0W2JcCpBSEjKGaEsQbcpzzqPcvXJGrWh0M0L4InSPeC8hhMCpJpFsyaAr3Djmve5wE2EjgFO7dKp5566k50AIQbJ9BWWp/4RNm7k2Yg7vP1htGlabRvXhJvp6/UQiUSLhKAP9AV7efIi2lj4WLs5PWPE2GdxeF+k5sVB0R2M33a190yZgqqZSWhUTSYYDYZpOd0w4hpSS1rqOeN+4ovm5cdH/HN4lCMjISyEp1Y0eidJc00E4NLNIQXK6B7c35vvU2dRDJJxYYK1HovR2DE4ajbyUoAkLq9NuId85Xts6qHcTiJ7P32c6ePejVoN6T4Iye0G6rQCPZeKWSOdDyBjCp/de+OLOi3d/r9yal1TreD1xZ7iBkDF0wR56s4l3NeKkR6KcONpMODxSVbJwScGk5exzmB30dg9RXzNys7U7rFQtHV8e/T8VpeVZ3PyBlWx+ch8PfOLnuDyOuO7J6bJx291r2HjtghmlbzSLypLLK3n9yT0E/WFefPgtFqwuGyM+ngxCEcxbXkx2UTrtDd3sf7WaDTcvw5MgMhgJ6ezdVs1grx9nkoPFl1XOTFc1hxlDCEFeWTZliwvZv/0YB988Tu3hRhasLpvWb+6svUReWRYn9tZxcn89HQ3d5Fdkn9NMW9LT1s/J/fUX43QuGoSI6VtK3EtoDFSPeS9sBmbN3ydR1Za8yGXwIWN8VFhBwa15ZxRtklLSr3fQHW6ejeVNCC3RXslYSxj1IjVsdqge8p2V1AztG1N52Rtpo3ZoP6tSb74o884E7+qV1DcY5Hv/+jQdrf1A7Afzk9/9JaXz5iI77waOHmjgu19/Mh7GLyhO56ePfe5d8VB6P8Bi1bj97tUsWJRPc0M3AX8ERRW43XYKitIoLs/CMsOojRCC1ZsWkT8vm5qDDezacpCnHszl1k9dTXKClKMRNehu68eb7omLwIUQFM3PZdW1Vbzw32+y77Vqtj2+iz+7/+oxlYlSSva8coQ3nt6LaZgsXF3K/FUlcwT5PUBajpeNd6yi+p0aWms7+f33NvNX3/4w+eWJr3m9HQO0nemifHEhtlGthjLz01i4toxTB85Qf6yFVx7Zwcf/6fYxZFiakjee3kvt4cZEQ1/SEAi0BG2s5CwqtmyKc9hrayyCxiAey/Qqo6eK2W4nY2JycvAdwubF9VOzK+MfxiQmIWMIizKzSNlkEEJQ7lnBOz3PMRQdG1E70LeVhcmX49LGt0Z6LzD3CDqHOYyCpqksXJzPwsWz7xviSXHxqX/5IN/77EP0dgzw2H9uYeeWg6y8porCylw0i8ZQ/xDNNR2c2FtPcrqHz377njG2A3aXjbv/9iZO7j9DzaFGHv7WM5w+cIZ1Ny0jIzeF/m4fB14/zvYndjPQ7SMjL5V7/u4W0rK851nZpYFQIEwkpMeb+5qmScAXAimRpmSgZ4ie9n4URSAUBUUROFy2mPD9EiWFiiK45q61nNxfz9ZHd7LvtWq+fs8PWXbFfOYtL8bpcRAORGhv7Kb2SCMNx1spqcrnb75/3xjiZLFq3PrJK9n90mGaazp4ZrhB9MbbV5GalUx3Wz+7thzknZcO483w0Nc5+D7QOY1AlxH6IuN9ieyqe9qd6yeCU4sVe4xlTjH7g0x70azMMW5ONWncayYGPr0XQ0YTRsEmgpQmbcEaDvdvn8UVJoYzAUExpUF7qG7GKcapIMNWyDzPKvb3vTzm9c7QGd7ufoIrMz4y3CD6vf29zxGnSxRR3aC/z49pStwe+1w6808AQgiWXDaPL/znx3n8/75E9Ts11BxqpOZQ4gjBonUVCcfIzE/lyz+5n4f+5Y8cfPME2x7bxbbHdo05TtVU5i0v5uNfvZ2qteXv+YVmKnjpt2/x+lN7CfpCBP1hgkMhAr4gpinxDwb5/gO/xuVxYHfZcLhsONx27vnizay8tuq9Xvp5YXVY+PS/3kVympttj+2ita6T1rpOXvj1G+OOFYqgcmVJQiF/Tkkmn/s/9/LTrz1G48k2Xvvjbl774+74+6qmsPH2VWy4ZTk/+19/oKe9/6Kelz86QG+kjQxbAXbVyUx1MVJKOkJnqPHtG/deiiUr4U18Jsiyl6CgYjDWgLR64E3mJ62ffdsDAS4tGZfmxR8d+1m0h+roj3SSZpvYI3A0pDRpC9XxYutPL3q0CSDXUT7uNUPqHBt4myLXolkjs+dCESpr026ndugAA3pX/HUTgwO9L6NhYUPGB2b8WUkpCRo+2kP1FDoXzDgiOEecLkFIKWlp6uXB/9iCrhvccc8aNl7iN4c5TA2qprJq0yLyyrPZ/2o1B984Tt3RJvp7hjB0A2eSg8z8VCqWFrHh1uV4M8ZXkgghKJyXwwPfv499rx7j7ef2UX+shaF+Pw6Pg/zyLNbesIS11y8htzSx55QQsbWooxzeJ4KiCFRNRUqJMgkBE4Cixo5XLdNr2tpa18Wp/fXjTDbPuqNHIwYDPUMM9IzoRno7E7fyOLtmzaJO31NAxDRpRoJ2LGfHVVVlHE0464WVqBWK2+vkni/ezOpNi3j7+YMc211De0M3waEQFpuFjNwUiubnUrW+gqWXVyZsOiyEYOnllXz5x/fzxtN72f3yYbpa+tCsKkWVuay/aRlX3LkKRVVwehwM9FzcfnV9kTZeaH2QZEs6Ze6VVHhWkWRJm5KJI8SucxKTtmAtW9t/xWC0e8z7FmGj0LUQyyylu1KtObi1FPr09jGvNwVOcKh/GytSbkAT1knXfBaT7a1AYFOc5NhLqRka2yqrI1TP0f7tXJZxN6qYOGIa2yNJg/8o29r/O2GF3sVAuq0Ap5o0pvGuRFIztI8TgzupSt44aVPm6ezVaKTaclideiuvdz6CPqojRtgM8E7PswxEu1iT+mdk20viOrHx4581640lew2p0xNupd5/iBrfPnoiLXym9PszJk5iEqX6rMrYe7t9fOH+h+Y0TpNASsnLzx7g+//2LBaryme+cAO3f2jNBY/7xivVE2ic5oor323I4fRTVDdivk5mrNGwEDFjxbM3/ckuOKZhEtWjGMOGiUIIFFVBs2qxlNZE7sOGSSQYQUqJZtGw2Ca+eOuRKPpwFZfVbj1vdZ6UkmgkOmyDILA5LBO2hTkXkZBOVJ9e01OrPTb+uSJpPRyNj2V32aZVCWmaZtwuwuawxtc/elxFVbHZLXGSJE1J0B8aXtPEeyRlzH8rqptIc9jPS4iYz5eqoGlqzFdrkhuSETWJRqIxp3EB6qjPHGJpT2nKST+vC0GT/ziPNP4bIWMIBRWLYifDXkCpaxnZjjI8Wgo21YkqtOGbrIKUJlGpo5sh+iIdnBjcSc3QPoKGb9z4ha4qPlL4z9jU6VmZTAQpTd7sepzXO38/TjtlETaqvBtZ5t2ES/MOkzWBxMSUBoaMEjUj6DJCyPCRZS8lyTK5XY0pTfb3vsgr7b8a51OlCgurUm9kifcaPFoqNtWJQMHEwDB1guYQfZF2qvvf5PjgDsJmIP53Xksm/XrHmCbBeY5K7i78R5Is47trTBdRM8LL7b9kb+8WzqUBVsXOytQbWZC0AaeaPGzeKZDSxCCKYUaJyghRM0LI9JPrqJiWNklKScQMsr3z9+zpfQFzXCNkgU1xkO+cT7l7BZn2IqyKI572qKcsPAAAIABJREFUNKWBQcz/qSfSQne4habAcYaG06MxywvBlyp/jfv82rYJf4RzEadLEKYpObT3zHu9jDlcRAghEKrAeoGkVVEVrOr0n5pUVcExgUfWubBYtXENgCeCEAKLzYLFNn0nZKvdMmL6eAEQQlzQWIoSi9hMZ1yhiIR/k2gMzaKhXcBpnjWQPR8hGt1H792AiUHY9NMcOEFz4AQANsWFx5KKVbFjETYUoWLIKGEzgD/ajz86MGGz22RLBldl3ItVmXxPpw7BgqQNVA+8Oc4rSJdhDvZt5fjADrzWLByqO05idDNExAwRMgKEjCEMGeWeov81JeKkCIUKz2qODb7NGf+RMe8ZUuednuc5PriTLFsxHksqqrAQlTphw09fpIPuSPM5VYWCeZ41rE69iaebf8BgtGc2NmYcVGGhKnkjNb699OudY96LmCF2dj/Nwb6tJFuysKsuBAJDRtHNMBEzSMjwEzL9aMLCR4u/MS3iJITApjrZmPEhQqafI33bx/h7gSRsBqgd2k/t0P54rz+LEpOz6GaYqNQn7Ik4G5gjTpcgdN3gyIGG93oZc5jDHOYwY4RNP+Hw9PU4qdZcrsn6GAWuhbOaahRCkGrL5fKMu3ml/ZcMRfvGHRM2A3SEzm/nIKZpf5hkSeearI/zeON38I0jOpJBvZtBvTvh346GgkqZZwU35HwKt5ZCsjXzohEnIQR5jnmsT7+TbR2/GePmfRZBY4hgAruF0UhUKTlVOFQPm7I+gUdNZV/fiwmjkhCr9gubgXhEbiq40G/VXI7mEkTN8Vb6+y6+AHAOc5jDHGYKh+om3ZqPMkutMOyKi0rPWj5Y8PcsSNqAKmb/9qQKlUXJG9mU9eekWfN4N4weYySkgtvzv0CBc8G0iReAU01mbfrt/FneA3i0NBShUuyaWvuemUJTLKxMvZGrMj9CsiXzos6VCDE38WSuyvoot+f9LSWuJbMgShd4LZlUJq2NR6hmgguOOI00ohwviBr9sDCbTw6jdVmJJFpnp5qNOSedK/4/Fz7f2bmqDzViRC+eKdtk859vT2P/f3bOc6K5Znu+OcxhDrOPVFsut+R+loZANXVDh2j0VxMy/UxPGitwqB7K3MuYn7SBItdCnGryRf3NC6GwyHsFKbYcqgfe5Gj/62NE0OdfrcClJWNL4HM02ZwlrsUk5T3Akf7tHOp/bTjKdP69sioOilyLWJFyAyWuxVjVkdRlkauKt7pGfKlGi/JnC4pQWZV6C1n2Uo4MbOf4wI4pR3bONu69kHSrELE+mhWe1WQ7SmnwV1M98CYN/uppVBcKHKqLXEcFRa5FFDkXkWEvvKB1XRBxiuoGXZ2DHHinlv3v1NFQ18WQL4jFqpGdm8Ki5YWsWl9OcVkmTpcNIQTqDNpVjIauG3R1DHDiSDOH9tZTd6qDvl4/hmHi8tjIzk1hyYoilq0uIa8wLT7vtM8tajDQF6CrY4DTx1s5Wd1KS2MPA30BQqFIrOmtx05mtpfy+dksXlFEfmEaKWnuac1nmpJQKEIoECEQiNDf62f32zUYw408pYTujgFqTrSdfyAB6ZlJk/aXG/MnZ4WtUhLwhzlT28mB3XUc2d9AZ/sAoaCOw2kl2eukclEey1aVUL4gh7Q0z6Sd4RNBSkk4pNPVMcjxI01UH2yi6Uw3fb1DhII6mqbgdNlISXNTUJxO+fwcSiuySMvwkOR1ok1RZDyHOcxhLAYDIXp8AQrSvWizVAyiCJVMezEZ9kKWpWwiYgTpjbTRGT5Db6Qdv96H3xgkYgQxiCKlRBUqdtWNR0shxZpDrrOCDFshDtU9aUXbbEIRKvmOSnLsZaxPu5O20GmaA6fojbQSMvyEzcCw4N2GW0shyZJGmi2PLHsxyZZM7OrU+mSOhhAK6bZ8rsj8MCtTb6I9VEeD/yh94TYCxhC6DKEJCw7Vg9eaRba9hELnQtyWVCxi/H2syLWYL83/zZhzsimzI6QfDU2xUOxaTL6zko0ZH6I1WENz4AT9egchw0/EDKGiYlUdsb3S0km3F5BpKyTJkj4rNg9CCJIs6SxK3kiFew0NPa0c766mO1qH0xNEx4cuwwgEVsWBVbGTZEkn3ZZPqi2XLFsxVsWOqlhRUC74ezbjqrpQMMKLzxxgy9P7aKjtnDBy4PLY2XTzEv7s7jV4khx89YGHqT0ZKwedTlWdlJK+niG2bTnM1s2HOFPTed7jk7xO1l4+jzvuWUtZZfaUN8owTGpPtrFvVy37dtVSc6KNYGDy/lJ2h4X5i/K54561rFxfHu9iPxEi4SgH9tRRd6qdtuY+2lv7aW/po7tzME6apgNFVfjs393Ibeepvju3qq6yKo8fPHQ/zY09PPHbnbz16nH8QxO3N9A0lUXLC/ngfRtYsaZ02tU6TWe62fzkXt5+9ThdHQMTfmdGw+myUVmVy4q1Zdz6wdU43XN+VnOYw2jUtfegR00q8zMmPGbrwdP84pXdPPjZD5Dink3B9QgC4QjHGjtYWJiF0za7jtlzmMNZBCM6P3jyDR5/8zAVuel865M3UZ574ZWECTB7VXVSSiLhKL/7xes8/8e9BBJ0eRdCxNMxfl+I55/YS92pDu5/YBPuafTmGj1nV8cgP/rOZg7urSccGquWP0uKRpPAwf4AWzcfpPZUO5/47NWs3lAxpbL7/p4h/uMbz9DS2JPYeVecDYnKMTf+UFDn4J56mht7+NhnrmLTrUvPGyEJ+MM8+budHN53JlZS/B7A6bJRe6qDn//XSxw90DhmHbE9HXuO0ajBwT31tDb38ukHrmPjtQsTGvWdCyklJ4+28IsfvkL1oUZMY+z5CiFiX1Epx5GpgD/Mgd31RKOS629bjpM54jSHOZyFlJLNe0+Q6naelzitqsinMMOLx3Fxfj9SStp6fTz65iH+/gNXzhGn9yGklOhRg6hpYrdYJvV3+5+MaRMnPWLw9KPv8OwfdhMKjhAYT5KDotIMsvNScDht+IeCdHcM0tzQQ2/PEEcONPDD72zGap1elEJKSWtTL//3289zaO+IOZ7NppFbmEZufireFBeKIhjoj6XW6ms6CQUjSAl1p9r5z39/lq9+626WrCyedD5Xkp2KymwaamMRLSEEySlOsrK9JKc4SU5x4XTZ0CNRujoGaW3qoa2lL046ujsG+fWDr1JUlsn8RXkTRrqEgGSvk5z8sfb1Ukp6unxjyKE3xYVrEsKpqmLapDQUivCzH7wUr+Bze+zkFaWRnZNCkteBaZj0dPtoaeyluaEnTkw72wb4+X+9TElFFoUlE1+sz6Kny8dPf/Aixw6PNKZ0umwUlmaQlu4hJc2NZlEIBXUG+vwM9gfo7vTR0+UjGjXQLCrrNs7DM4Vy7znM4VJHMKJT3dhBaVYqbX0+ugf8uBxWKnLSSXLaESJ2HegdCtLU1c9AIIQAMr1uynPS0FQVU0rOdPTS0jPIa4drqSrMYuuh0wCUZadRnJmCEIJ+f5CDda1ETRNNUSjNHt8uQ48aNHb309ITMxPN9nooy0lDVRQM0+REcyfJTjuRqEFLzyCKEBRkeMlLS0JVFHzBMHXtPbxzsomjDe28dewMyS47ihBsrCrBosau+cGwTn1HL92DMW1KktNOcVYKXtfc7/pSgJSSHz27gwN1LXzn/pvJS7s0+sJdipgWcZJS0ljfxYtP7x9DmvKL0vjI/VewfE0JqemeWMTJlPT3+zl9vI0/Pvw2h/aeoe5U+7RNfIOBCH98eAdHDzTESZM31cU9f345y9eWkVeYisUSO41o1KCny8fut07zxO920NYcKzft6/Hzm5++xlf+/YOkZXjOm7az2Sxcds0C9uysJSfPy9qNlZTPz6GgKI2UdDc2myUeUQsFIzTUdfH8E3vZ+vzB+Pr6eoZ45bmDzF+UN+E8Lo+dj37mqnFpwEhY56EfbuXE0RjJ0DSVTbcundQ5XAjIzJneF732ZDuRcMxcrHReNnfdt56qpYVk5nhRlNg5hsNRGuu6ePYP7/DqliPxNGJXxyAvPXOAT//tdZOmQffuqOHYkRHSlJWTzIc/uZFV68pJy/TEXJiH9zSqx7RlbS19NNR1cnj/Gdqa+1i8omhMI9s5zOH9il5fgB89/zblOem09Q0iJbT1DXL5whL+4oa18ajQo28c5J1TjSQ57AQiEfyhCH954zquWVKOlJLqxg4O1bfROTCErU3DHL4AaYpCcWbM2C8Y1jlU38aRhnZOt3bz/Nc/iccx8vBqmpJth2r4w9uHsFk0DMMkGNG576oVXL98HpGowe+2HyAcjeIPRdBUlQF/EIfNyhduu5wF+Zn0DwV552QT+2qbGQiE2HG8AbtVQ1UU1s8vwqKqRKIGj799mK2HTpPidhKJGgz6g3zq+jVcs2R8e485vPswTMlbx+oJhfUZyUX+J2FaxMk0Jc889g6tzSOdi72pLr70z7czf3H+mFSYUAQpqW5WbygnNz+FB//jRfbtqpmSruUspJTs31XL1s2HiA5XmaWkufn7b9zB0lUl41JhmqaSmZ3MzR9YSVaulx9+53k622JPUSeONLN18yHu/vhlqOrEN3ohBEtXlfDdn3yc1HQ3Lo89fmM/9ziH08b8Rflk5XoJBSO8te1Y/Px2v32KYPA6nM7EoXFNUykqHR+tCQYjuEbpeIQCmTne85KwmeIsaaqYn8Pff/MDFBSnjXFYFkJgt1uoWJDDJz+/iXA4yptbq+PnuGfnae68dx3pmeMbWY7GW68ej6vlbHYLd967nhtuWz5uX4UQWKwa6VlJpGclUbWsgGtuWoIeiU4acZvDHN5vqGvv4X9/7EZcdisH6lr59uOvsroin40LSwC4Y10Vd122BIdVI2qY/HjzDh598yBXLy5DEYLrl8/jykWl7KttZtOycj5yxXJADkd4Yr+rrBQPf3XTOl7cf5KfvLBz3Bo6Bnw8tHU3H7t6JVdWlQLwxI4j/Per+5iXl0F2Sqzlz55TzXzv/ltYkJ+JPxThH369mW2HaijPSSc3LYmPXb2C4qwUmrsHeODWDWR6Y39nH36oDUZ0nnmnmrsuW8Itq+YDEIro2K0Xbng6h9lBbXsP7b0+vO+yeer7EdMiTj1dPna9cTL+b1VTuOOetVQuyptQPySEIL8onQ994jJOH29loH/qJlXhkM4Tv99JJBK7wVutGnfcs5YlK4sn1A8JIVBVwar1ZVx13SIef/htpIxV4+184yRXXFdFbv75uzu73HZKKqb+5fGmuLju1mUc2d9Af28sDB3wh2mq76ayavYJz2wiJc3NR//iKgpL0s+TVhSkZXjYdPNSDu6pZ3D4MxzsD9JY33Ve4mSako62kQaXniQHpRVZU6qQU5RYlR3vUoNjKXWk0YE0WpFmN1L6QIYBHVABK0JxIZQUhJKFUPNAuC5xuwSJlKHYORkdSLMHKf3D5xUFNBA2hPAglHSEGjsvgTb9Hm9zmBYuW1BMdkosAr6oMIvy3HR2nWiMEyevy8Gxpg7aen0EwhF6fAG6BwNIQBECm0VDVQSKEFhUFWcCt/azx1k0NaHSdeeJRnyBMGc6+ugfqgaga9BP14CfM519ceK0tCSHxUU5OG0WHFYLCwuzaOsbJGoYWDUrdqvAqqkIAXarZdxaLKpCSVYqbx2rJzPZzbKSHDKSp1eBPIeLB1NK9pxsQo8m0PXOYRymRZz27qgh4B9JLeXkpbBibdmURNcLlxRQXJ45rVYix48001g30iE5Jz+FdRvnxVNz54OiKFxz81Ke/P0uosNfhrMVbDl5KbPuSFtQnE5mdnKcOBmGfF+YWFZW5bFkRdGU9mPewlySkh1x4hQKRujuOL//iTjHWkTXo/iHQvG+arMBI7IPfehBpBzvbqs5bsXi/EjCv4tptgyk2YMR3oYRfgMZbULKfqQ5BDJEjDSZwydxlmQ4QUlCKGmoluWotqtRrEuBs+XU7+3NQEoTCGPqpzDCWzEj+zHNLjAHzyFNJjEPXA2EHSHcCCUJoeagWNeg2a5FaMWA5YI+KyklZuQdIkM/IlGhrhAObN7vIZTz9o06z/hRosHHiQafP+9xFtfHUW3Xz+hcpJTI6Akivu8j5fiHP6F4sCV/e1rnkDzqyd5m1Uh22unxxcYORqJ898ntNHf1s6Agi/Sk4ZJuScyjZJZ+Ox39PqKmSdfAEAP+kd/PDSvmkeUdKbnP8npQR4mFrZqKHjWnnEGwWy18/pYNPLnzKD9+YQepLgfXL5/HHesXYZ3gIeqXL+1m96kmbl49n1vWLACgpWeArftPU9vWy1AoTLLLTllOGpcvLKYkJy1hE2opJaaUtPUO8saRek40ddLvD+K0WSjKTGHj4lLm52eiTtAjsKa1mwef34Gmqnzlw9eQ5LRR09rNa4dqqWvvRTcMclOS2LiohBXleTGSOkmvQT1qcKqlm7eq6znT0UcgHMHrcjAvL4OrlpSRk+ZBSdBkerb2R0pJ/1CQI2diKdzTLd0crm8japr0+AL8y8MvYzunMtyiKnzkquVsWFg84XldyD6PHmcoGGFvTTNvH62ne9CPw2ZhQUEWVy8pJS3ZNWmz8YuNKRMn0zQ5erAhTkIAissyKSieOFIxGharxprL5k2ZOJmmyb6dtWOq9krKpyZGPousnGRyC1JorI/Z2UfCUU4ebWb56hLEedJ1M0Gy14lrVO8vKSWBofEVh5cSNIvKmssqYlGdKcCb6iLJ64SGmM1/NGrgn+QchRAUl2bGCfBAX4Ctmw9TXJZFTv7sEFhp9mBEdoIcb/9vCHtC4iSljozWogceIRraAmY3cL6nLQlEQEZikSizA8lpzMhudP8vUKwr0Jz3oFmvACXtPXmSltJEGi0YkZ1EA7/H1I8xQvwmghH7T4aRcgBptkD0OEZ4O7rvB6i2jWiOu1Bta0Ckzvy8hBVT3w8ywcOEcGBGT6Fa185sbDmIEX4NM/LWeQ+LqtmotutnNgdgRE9jhF9O+J6iloM4f8r6XPiCIw+hhiEJ69E4mXrjaB0Halv45w9vYkVZHkLAz17UaeruP2eUC/ueeRw2MpPdfPr6NWPEwIKY3CI83ChZU5VJZzrfd0MRguLMFL5w2+V8/OoVbNl3koe37yccNfjY1SsS/k1NazfvnGgkPy2ZKxaX8uzOY/z6lT0M+ENxPZcg9pB8pr2X/3XvpoSEcsAf4g9vHOLxNw/TOxRAmiMtfhUh+N1rB7hxZSV/fv1q8tKSxp2HLxBmz6lmFEVwvLGDmtYefr5lF4GwPrIOAU/uOMLVS8v57C3ryU+f2MSzvr2Xn23ZxZtHzxCK6GPORSiCX72yh49ctYwPXr4Yr8sx4TgXsj9SwraDNfzX029iDhOes9GmSNTgWGPHuHmtmsoNKysTrmU29vksTrV08/898XqsqCFqxMd4ed8pHt62j8/euv49b0w/ZeI05AvR3TkSXVBUJW4wOVWUz5/cr+ks/L4QzQ3d8Wo1RRGUVeZMqfz9LFRVISvHGydOAA31XZhSznqvGU1Tx4qXpcQ0L22BnaYqVCzImfLNUAgxRrNlGpJwePJGipdfs4A3t41oo3a+cZJQMMJNd65kxdqyMZqu2YZpNCBlCCFGk9og0cAj6IFHkdFTTM/peNwMgBmLqOhHMGxXYXE/gGpZwLsZeZIyghF6ET3wMGbkEHChpD0WtTLCWzEiO1FtV2JxfQrFshwhplcZK4RAKF6Emo+Mnhx/gIxiRmtmTJyk6cPUayc9ztSPEiOKM/H9NZH6qQnfFZaFTLeD1f7aZu69chlWTaVzYIjath4+ce1KAAJhHaumkuV1o6kKPT4/x5s7iZ4j2lUVgVXT8AXDsevaNIntyrJ8fvvafo40tJOfnoyqKLHCFz2KbZqfs1WLicB1wxgXUTbN2JhOm4WMZDcfvWo5B+paONF8fj8+gO5BP8/sqOahl3Zjt2pcs6yc9CQXhmnS3D1ATWs3S0pzE5orR/QoD720hz+8cZBI1GBxcTbLyvJIcTsYCoY5cqad/TXNPLXjKAOBEP9y33W47YmvR6GIzq9f2cupli6Ks1NZPjxO31CQPScbOdnSxUt7TyKl5F/vux5bgsxIc/cA3/vj6+w60YCiCNZUFrCwMAuXw0rvYIB9p1s42dLJQy/uptcX4It3XoFlElnDTPZHCFhRnseX77oq/tqzu6rZX9NCssvOJzatItUz1kxTUQRLSnISrmG29rmzf4jvP/k6e041IQSsmlfA0tIc7BaNtl4fu0408OPndpCRfOGmmheCKV9B+nqG8A2MhHI1TRlXSj8ZsoartabiWzTQH6Cne6SpX0yMbR2jl5kMesQYR7QG+wPTEqhDLHpkGCZ6xCAaNTANE9OUmKaMhfCHva3Oiq3fL1C1GPmdDkbvp0zgu5QIy9eUsnZjJbvfOhXbN8Nk365aThxtJr8onRtuX86q9eUke53Y7BeWFhoHcwhptCC0suH19qEPfo9o6ClIkNq7IMgARmgLZrQOq+cfUG1XTZtkTHtKKZFGE7r/QaLBzSAHLsIk/th5RQ5hcd2P5vwwQvFMawghklHUQoxExAkDUz89oxSUlBJpdiGNyZtiS7MbaTQhtJJpzRFfY/T4hO+qlqXTHrF70M/Xf/cyhRle9tU0k5PqYV1lERC7qT28fR8PvrCTyvxMjjd1MBgIjbuJCiG4bGEx2w7XEB4mJivL81ldUYBpSk62dNLY3c++mmYCYZ0t+06Q6nFSmZdBQbqX8pw07lhXxa+27uHImTa8LgedA0MYpslX7rp6Wg9VOSkeclOT+I8nX6eqKJuoYfLp61ZjtWgMBIJ88RfPUZaTRmaym/Y+Hw2d/XzmhokNe8/icH0bh+rbuLyqhPtvWE2W14NFU5FIwpEovmB4TNrzLExT8uK+kzyz8yimlHz4iqXcf8MavC47mqpimCa+YJiHt+3j4W37eONIHU++fZT7rl6R0MMorBscqG3h5tXz+dytG0hxO9BUlahh0HvVcr712DbePFrP9sO1bD1wmptXzx+zf/5QhN9s3ceeU4247FY+deMabl9XhcdhQ1UUdMNgwB/iwc07eG7XMZ7eUc28vAxuW191XkI8k/0RQlCak0Zpzsj1/1BdKwdqWnBaLVy7rJzCzKmlnWdrn6OGyVM7jnCgpgVVEdy1cQmfvmFtzNpCEUR0g9beAf7l4ZepbuiY0touFqZMnEJBnfAoYqAoCilp07Odt1g1XB77GAI2Efz+MEODIy7WhmHyo+9uhu9Oa8pxGPKFJm6QNgpSSoyoSUtjD2fqOmlu6KG1qZeeLh++wSABf5hwSEfXDfRIFD0SjVf+vV9gs1mw2y9+VYsn2cH9f30tFovKju0n4qWu/qEwJ6tbOFndQkZWEqvWl7Pm8nlULsojdZqtayaClEHM6BkUtRRpdhIZ/DZG6DliGp+LgWEtzMBXsHm/j2rbeJHmic1l6oeJDP47pr77Is4Tm0uaLUR838I0mrC4/xpFnXraHCUpppcKx4xVx8JExiOD0/f0MSMHiaUkJ4EMYurHUGZEnEzM6IkJ3lNQLIunPeId62L6ntOt3VyxqJTrllWQ5Y1974syvHzj3uvZfqSWHp+f65fPozDDy9vHz4z7Xfz5tSvJ9rqpa+8lpEfjlWzmsGXB8aZYVOe6ZRWcbI6lzO0WLd5+5ZObVrOsNI/dp5oYDIbITU1idUUBdouGYUrWzCvAbrXEK24VRbCkOIfizPCY9i2FmSl85a6r2X6kll5fIKbLGl6r02bljnWLqO/ooccXIMvr5qsfuoaVZZMXz/QNBVm/oIgvffAK0jzOMefvsFoSkiaIEdMte07gC4ZZPa+AT1y3iozkkWIORVFJ9Tj58JXL2F/TwuH6NrYdOM11yyvISU2cdi3M9PLRq1eQ6R25PlkVjawUN5+7dQMnmjrpGvDzyv5TXLm4FPewtYSUkprWbrYfrsGUcO2ycu6+fAku+0ibGZuikZHs4q9uXk99ey+H6tp4Zmc16+YXkj3Bei5kf2YLs7XP7X2D7DjWgG6YLC7O5t6rlpOWNHI+DptCaXYan71lPX/38+fjaeT3AlMmTpFIdIy+SQiw2acX8haKwGaz4GNy4qRHjCmlgaaLycjN2QhSQ10XTz+yi5PHWmlv6ZtS25X3G+xOK+9Wd/DCkgz+6u9upKwym81P7KW7c3AMf+3qGGTL0/t5+7XjFJdlcvWNS7jyuqp4i5UZkygZQhpNSKLoQw9ihF5gaqTp3PmmF6aUZicR339gU3IQWtksa55ikT5TP0hk8Jsx7dC0cCHnZhANPAIYWN1fAmVqOjUhLChqIQjbsOj+nBWYvUizG6EUTGMtABJDPzDFQ8OY0VNIefO0P49YtKor4XtCyUSoGdPWa1s0lTvWVSVcixAxcrKkeGxqZF7eeLLqttu467Il417XVCXh6+fOY7NorKssZF1l4bj3FSVG8EZDVRSuWzZv/LFCMD8/k/n5mePes1k07lh3fi+6iaAIwa1rFowI5BOcw7mQwyLlk81dqIrC0rJcsryJPfy8LjsLCjI5XN9GR5+Ppq7+CYlTeU46eQn0S0IIslM9LCnJYdvBGlp6BmjrHaRi+POSEo41dtA14Mdps3DZwuI4qTp3nLQkF+sXFHP0TDsNnX2cbu0hK2Vi/8GZ7M9sYbb2WUoZq+Ts6EUIQVVRNrkJdFBCCEqy0yjJTuVE0+Rp3ouFKTMfI2pijsqvCyGmVN02GgKmbGJoGuY4Ey6bTUNcYJNg2yQRlmAgwnOP7+GpR3bx/7P33vFxXOe99/ecmdmOXolGACTYmyiK6oUS1WXLRS6yHLfYjnPjOO9NXOKbOMnNdfLeG79O4uvEjuPYTuzYciyrWL130ZLYxN4JAiB6L9tn5rx/zAJE2V0sgAWb8Pt8KJE7s3NmZnfn/M7z/J7fM9A3MmFy13UNf8CN1+/GMDQMl46mCUeoJgQtp3oIDqfu9Xa+wTnvszPWqKXBhz95DTfJgMTJAAAgAElEQVTevo7nHn+H154/SGf7wARSOjQYZu+uJg7uO81TD+/kQ5+4ms3XNODxzraFQwTbPIoZfhAz9AtSa3/cCK0cIQuRWjVCKwfhBzRQQWy7HWU2JSbQdjIhX3Z8L/Hgv+HK/TMQM0ttpYNSYJsHiA5+HZUyCjIebmdilyUIrQKhLUKIXMckTEVQVje21YqyO1BWF6hh0pOpqHMvhRtXzp8AmTUWlfoSEIEUxKkXZXcAMyVOMez4Oxnu62ipUEHnPGYAR2ifvHhA6EsSn+9Caf18QErBpmUz+14ooLlrgP6RMG5DA6V452Rb8n2VIhxzFulDoSgDI8kX9lIIygpykto+AAQ8rrH0Vv9ImO7B4BhxspU9ll7yuAxWVE8ll6PQpKSuvBC/x0X/SJiW7n6Uqk1JzGdzf7KFbN7nzv4RhkJR3IZObVkBupZc5uD3uKgqzrswiJOuywlKdqUUZmzmobLJfcpSQUgxwYxR0yQf+PhVLKqcXcnyKHLyvCnJm23bPPbr7fz8hy9PSEv6A242XdXAspWLqKguoqgkB5/fjc/vxuXWMVw6sWicv/nTB3hne+Oczu9ixqjHVtmifD72u9dz4+3r2LvzFG+/foy9u05NSOGacYujh9r4h28+yoc/eTXv/fDmCVWLM4EVeQkr8gzJSZPLKb33bEUam5B6HUKcSTGMQSmUCmKbR7Gir2CGH0dZJ0hPMGzMyNPo3vciXVdmb+Vn9xAf+jtUGs2NAx3pugzdcyvSWI/UG1Jqk5QyUVYzdnw/VuRFzOhzSasUz8DCDP0nmt6A7v0QZKDlEnotQgRQ9EzZpuz+lBGddLDNEyg7me4xWUoQlNWKsrsRcqbE6QipqhOlXoeQ51asejHDZWj4PTNbOCml6Bl2KjijcYt/e/pt/u3p6dPZccsimsLLSAqBP40GU9c0Al43QkAsbhEZNz8qBf0Jqwldk+SnabQshNOOxmXoQJSBkQiWbSNl8t/YbO5PtpCt+6wUDIWcBZUmRdr7o2uSwDm63rFzyHRHl1tHN858cEpBJDKzVJptO21KMhrPpeMZlwoUQrD2ksVcesWSGY2ZKZRS7NvVxGO/2j5GmoRwfI4++YUbWba6Ap/fPYHMjcdsSOS7GVIKKqoKWVRZwFU3rKC9tZ/XXjjIi0/sZWAgiJVIqYaCUR746Rv4czzc8f5LMzLOnAxlJ1sBCYQsw8j5E3T3loSFQJpophAIEUBzXYI01qB77iI28i2syPOktTFQ/cTDv8btuhzHRHNuUCpEPPivWLFtafcTWhWG//fQPLc4qaR01wYIoSP0eqReh+a+AS12B/Hh/4tt7iclOVQR4sEfIfTlSGPDtMRQyHKEVoKyTiU5VhBlt6GUNSNBvR3fB2rqc0jz3I4VfRkmeS7ZVqtD0LTajIXoSilsMxVx0pFaLZC5Nqs0L8Bf33cruSm6CixgIvRZZhlGy+sF4Pe6UvpFjYcm5TT7pfMfcuoqBWKsxH9sG05Lk1FMV/0oxZmRTNtOuzyb7f3JFrJ1n61EFbpApG0wLIWYoK07F8iYOHl97glpLtuy6ekeTvOOqYjHLEKhzMqk/QE3OXk+SPSbU0pNa7Y4F8RjJttePkxP15mqpIqqQr74tTtZusLRGaSbGMwkqcUFTA8hBLl5PnJyvSxbUcFdH9jE07/ZxdO/2U1/rxPxCI5EeeT+N9l8VQNlFflZiNxIpOtSXDl/jjTWTUsqJp0xQrhAX4o77++I8hdYkSdIl7qzIy+gctoRWtWczlopGyvyCmb416QTQ0t9Da68v0Uajr5lZvdLIGQOmvtmpLGS6MBXsdOQNNs8ihn6d1y5f5tIbaY5stCQxlrs2PYkWxW2eQKIkSkJUcrCju9NvGciNPcWlNmcIH7jT7gbZTWj2ITINLWmBlBWa/JtMi8RScv8Hhu6Rk1Jfsb7L2DmEAhyfU6E2udx8Zf33cIN66ZfdDu+R8k/S1spQtFYSvNeJxVlYiuFS9ec1NW4446KtG1bEYzEkmqcRhGKxsesJ3K8LrTz1GE9a/dZMBY1s5UinEbfbNk20fi5dTjPeMYoKPKTk3vmgWaaFp2t/Qn35czQ1Tk4FkmYDrn5vglVe7atOHGsY0bjzQSDA2Eaj3dN0DRdcvkSlix3fI6mezBGQrELzo7gfILj9SMoryzgvs9ez+9+cesEj7DW5j5OHO3Izlh6Ha7cv5wFaZp0viIPV+APkPqKtPsqFcKKvjmrcSYeKEQ8+GOU3ZdyF6Evx5X7DaSxNqPvbYqjJD6PKlw5X0NMc31m5Hls80BGR5Z6aoGwbZ4CNYMiDHsAZTYzNSLmQRqrUtgOKCzzEOnNTie9w2pH2cltHoTIRc6REC8g+xACFhXmYmiSuGnR0T+MJp1IRbo/yfqSjsJWiv6RMLEUqbxIPE73oLPY83tcY4QCHJJQW+bITGKmxeme1LYhSim6BkYIRWO4DZ3S/EDaCMy5RLbuswCKcnzOcSyLzv7hlHN9zLToT6FDO1vIeNbwBzyJViXOv21bcepkF8NDmV/AkQMpVm1JkJPrZcny8jE9klKKowdbx6IQ2UY0EmNwUouUVeurM/rCKuX0Y8vGuYmx/0wYYM7HvZCgGxpXbVnBFddNdKltPJ4FMaDIwZXzF0h9zaxJ05ljCYTegO7/OJCu6CCOHXsz0QpltlCOl1J8Z5rzycMV+COk67K5XxsOOZTGWlz+z6YXt6sR4sH/QKnpFw6asY5UKUtlHkepzI07bbsVO0kkSOh1CFmK1BtIllpxxOSZL3JsqxVUcv84IQsR2tRqtAWcawiqi/OoLS8kZlpsP9pC/3B4zgvvxo4+OvqmTupKKXqHQmMC8LKCHCqKzlTmSSHYsKQSv8dFKBrjzcNNxK3kBGw4HGXX8VaicYuKwhzqys9OJwI9USwUiZtYdmYefdm6z0IISvIDVBTlYVo2B5u7GAwmKSJRir7hEMfbpuokzyYyfroKIVi7cTHauLxk08luWhp7MrpJ8bjF9jeOZXxiQgg2Xbl0glN1S2MPu98+OS+O3LatJqbaBHi8mXkcWZbN4f2n6euZO3HSDW2CjkfZXJRWCOkghGNbUVqeN+F1c86+HRLd+0E09+asPYiEkOie2xGyLM1eCts6nXLyzQTK7ice/hXpIiW6Zwua59asmm4KoaF5bkUa6UvJ7dj2hA5omuNpJQiZvIOAsrvB7s3sxJRCma2JSryJkHoDQvgRek3S9KEdPzJF+5R2HKstZcRJ6vUXvTB8YCjMiabuGUsRlFL0hIM8eeoI/7pvO/+y9y0eOn6AtpH5k1yMQgioLM7j2jX16Jpkx9EWfvXangmC7YnnCj2DQVq60/9Gj7V28+Ke4xP0SuBEo57deZTjbT1oUnBpQ9UEewAhBMsqi9m8vBrLsnlu1zF2HWudoIMaPc7Le0+y7eAppBBctqyaJYtmZlI8W5QkesANBiOcbO8lE5uSbN7nisJc1taVI4B3TrTy+oHGKXO9adk8+fZhOvpnJhPKNmbkJ3DplUvJzfOOEYTujkG2vXyYhpWLcKUo0QTnB7RnRyOnTswsYtCwchGrN1Tz5qtOq4OR4Qi/+a+3WbW+hvIZaF1GiZ1SKqW42zA0vL5xSn0FXe2D0zajVUrR2TbAc4/vmeBzNVtomiQwrnrMsizaT/djmtashNHnGratxqJoM/m8YjGT7q6Jk1Vh8dxK+oVWhe69CzEDIW9mB85Bc1+JGX4g9T72AMrqRsiZue1Dwow1ttMppU95Dn5036fIhgB9yqFlriOGj6VONyp7ADv2W6S+aprP2YU0lmFFk0WfFbZ5FGmsnPacFJajiUpCgBzbAxdSq0LIfJQ1aUGjQtjxo2juKzMYJ4aymkkeoZJIY02S1y8eKKU4fKKDnfua+OxHr8m4R5hSitaRIb7+xjPs7e1gOBZFAX7dYFVRGX+66XouKcm83dNs4DZ0PnztOnYfP807J9v42Qs7OdzSxXVr61lcWoDb0BiJxGjuGuBAUwcn2nu5eeMyPrl1U9LjCQFSSH72wk4GRsJs3dhAeUEOvUMhnt11lIde34etFDUl+bz/qjVTrq0gx8d9WzZyqLmT5q5+/vaXL/CR69ezeXkNeX4vnf3DvLz3BL958wDD4Si1ZQV8/KZLk7ZumQ9sXFqFy9AJR+N874nfYivFmtpyXLrGUChK/0iImtKCKX5R2brPfo+LOzev4s1DzfQMBfmnR99gYCTM1avr8HtcnO4Z5NmdR3hu91FyfG6GM9RLzwdm9Ink5nnZeud6HvjZNlSi5chjD7xN/bJyrr1pFS731MMppThxpINf/vg1hgYyXOWNnpyuce9nruPooXb6EkL0Iwda+fb/fITPfekW6paV4XKlvoTRCbinc4jG413k5HpYvym5a3Ag10tFVSFHD56pwPrty4e58fZ15OUn96mxbZvW5j6+962naDmVndChEIKlKxbx8nP7sRLdx/fvaeLw/tOsXl9zVkK22UTTiS6aGrupW1JKWWWB48U1zTXYtmLXWyd4+7UzEUq3x2Dl2rloSQTStQlpbJhxW4/pIZGuyyENcVJqGDXrdigRrOhrkEbbpLlvQGbdaHPc8T03wfD/SXg8JUMUK7Yb3TsEIi/FPjiERl+NFX0p6WYrvh/de/f0J6Siyf2bhCfhDK4jtBqELEJZpyftZGPFd2ZEnFAhLPNkio3C+T7NAYPDYRqbe1hWX0ZTax+9/SP4fW4aaksJJDR+3b3DNLb0EoubFOb7qa8pxuM2CIVjHGvsoqaykObWPoaDEYoKAtTXFI91trdtxanTvXR0O1GeyrI8KsvzxxZhrZ0DDI9EWFSax4lT3YQiMYoLAiyrL2VoJMLJ5h5e2naE0+0DvPb2MQxdY1FpHg11pWm/axHL5O93v85rbacmxC2G4zHe6mjhu3u28a1rbqfYO7/RutL8AH/9iVv5x4df463Dzbyy7ySv7m+ckMBVCUNZt6Fz66Wpr8lt6HzgmrWc6uzj5y/t5j9f3IUQTlxG2c4Ce8miIr5yzw2U5U+1u5BCsHFpJd/42M38w8Ov0djRx//34CuOnhAnGqOUQtckmxqq+NL7rqGqOM1vKcvYsKSCuzav4rG3DnCyvZc//fGTEx6VLkPnf3/mDq5bUz/lvdm4z0IILl9eze/fdSXff+K3dA6M8PcPv8o/PvKa837lWC/cedlKygtz+OFTb2X7FmSMGREnIQQ33bGe7duO03jMyeVGoyY/+u5znG7uYcuta6muLQaci+zrGWbXWyd54sHtHDnQhhACl0cnGs7MxkAIQf2yct7/0cv5xY9eHUtZ7d/dzN/95UNcdlUDl2yup6aumJw8L1JKotE4QwNhutoHaG3u5djhdk439dByqocPfvyqlMTJH3CzfE0l2145PCbyPnKglft/9Cr3/M5VFJVMjHYEhyO89uIhnnl0N0f2O52zXW4jY7uFdNiwuR632yBkOoy6tamPf//ei9z32etZvaEGY7wthK0YGYkw0BeksChAIDfL0ZQ5oq2lj+9/6ymKSnKoXVrKyrXVLFleTmV1IbmTCGksanLyWCc7th3nucffmaCfu2RzPeWVmblUJ4eG7t6KEPPTYkZq1Tg6pxTfbRVG2cHk26aBsvqw07pjG2iuy6atapsLhMhB6iuw48kq4hwoszHh/p3uYe/YHjiPnqlRHGUenTbK6yCGHZ8qSBeiECHLnPfLohQpVIUd35fROEpFUGaKPngigNQXT3Oe6dHS1sf//clL3L5lNXsOtTpeZ1LwiQ9cjt9XTGNLLz/+1TbCkThet0F33zA3XrWCu29ZR0//CN/72SssWVzCcDBKPG7S0TXEPXdu5NbrV6Frkte3H+eXj+4g4HcjhCAYivLR927iykvr0aTkrd2NvLTtKGuWV9DaOUAsZlJRlk/94mL6BkLsPtDC8aZugqEoO/Y2o2mSDauqWFpXmrYmsWlogP29nSmTPbu62ugIjVDk8c3rYlAIQWVRHl//yI28daSZl/ee5EhLF91DQWJxE4/LoLwgh6UVRWxYUsm1q1O347FtRVl+gA9fu45H3zzI6wcaOd0ziGnZlBYFuHLlYu66fBUrq1OTSiEEl6+o4ZufvJUX3znOGwdP0dw1QCQWJ8fvoa6skGvX1LFl/VKqSqY6lM8npBB84c4rWFyWz4vvHOd4Ww/BSByXoZHv91JXXkBpXnL/s2zdZyEEd12+iqIcH8/tPsaOoy30j4TxuHSWVhRz4/ql3LF5JbtPtOJzu86Z5+yMiVN1bTH3fPwqfvidZxnocyaC3u5hfvUfb/DEgzvJL/DhD3gIBaMMDoYIB6NEwnF0Q+OW92wgHIrx0tP7Mh7TMDTe86HNRCNxHvrFm4SCUZRSTu+403088+huXG4dKaXD/hNaJdO0iMcsYolc63QhZikl1960mlefO8Chfc4KNRKJ89ivt/PGS4dYtqqCopIc4nGL7s4hmhu7GegLEo3E0TTJdTevZnF9Cf/5w1cw51gqWVNXzHVbV/PsY7vHGgnv29XEN7/2K/LyfeQX+hFSEgnHGBkKE4uZCCH4k79837z5XM0FcdPixNEOThzt4LUXDuJ2GxguDX/Agz/gwTA0wuEYQwMhwqEY4VB0QmucxfUlfOC+KwnkzKHnkgygua/KwtVMhTNJ+x0BtUoeFVIqTrKy+engNLHtSBgwphhfliD1FVnVNk0dxEAay9ISJ9tqcRzAVX3aiUNqFQhZnFSfZJtNGTl7O2NNfb/Qih3XdxxvKmmsxoo+z2QPJmW2gN0NWmoHZ6UU2EMoqyXpdqkvBzH3PmDhSJzT7QP88Wdvwu3SsW2F12MQisR58Mld5PjdfPnzWzF0jYPH2vn2D59nzfJF5AQ8mImG41/5/FaEFDz36iEeeGInG9dUE4tb/PyRt9l6zUpuu34VCnj21YP8+L+2sWJJOcWFAZSClrZ+7tiyhnvvvmzMX8jQNRZXFvLR92wiHrdo7xrk93/nOtwu3amImuaaesJBBqOpuygMRCP0htNnIL76oS186X3XIoXAm0YKMh1G25jctmkF169dQjRuYlr2GHE2NInL0PG49LT+SrZSROMm1SX5fP6OK7jvxo3ETQuFwtA0fG4DtzF9RF2TkmWVJdSVF3HvDZcQM60xGYlb1/B5XBn5FGXr/oxCCEFBwMdHrtvAXZtXEYs71gqjZN7QNfzu1MaT2brPLl3j2rX1XLa8mlA0jmXZSClw6zo+jwtNOo2t/+t/fBwpxbz34kuGGSdPNU2y5dY1hEMxfv5vr4xVkplxi8H+4JTKNACvz8XWO9fzO5+/gWcee4eXntmXcXssIQRen4t7P3MdeQV+HnvgbVpO9ThtJyxFKBglFJw+1yk1iXcaw7mSslw+9//cwj9+8zGaG7vHrqurY5CujuRpFsOlccPNa/j0F7cSj5n86j/emDNx0jTJRz9zLX29w+z87YkxUebIcISR4QitLVMnZ12XWdFYZRtSm+iWE4uaYxG96cT0UgoaViziM394M+s2Lp7T6kvTV8+4zcZMIDAQ0otK+RFYkEHVWTLYsZ2kbhWTIAtzjHxMD336sns1hLI60xcYAkIrQ2ilSYkPKoRtNaPJVWmPYcd2kMyQ0mkrcybK5DTflVP2VXY/ttmIloY4AQldWXLC6wjm564/EQJuvHo5hfkTI4ZdfSPsP9rOsrpSnn75IAAjoQgjwSitnQMs9zsE8YpL6sgJeBBCsH5VFfc/uoP+wRD9g2G6e0e46erl5CR0kxtWVfOb5/bS3NZHcaHzeygryeHStdXkTnLm1zSBK0GUpJS4XTqeDCdoa5IBZDKY0xT5pHOPng0c52/XnFy2lXLmJJeu4ZrD+WXjGNm+P+B8F3VNzomMZOM+SyHwuV1OVCkJRm0azhVm9avXdI3b7r6E0vJcHvrFmxzc04JpWtjjKg2claWgpCyPj3zqGq7ZspKcPC+LKgvIL/ATCkZn5DOjGxq3v38jqzfU8MITe3jpmf0MDQSxE1qr8RAJIbKUAo/PzfJVFWy5dS0br5iam52MlWuq+O/feC8P3/8mb712lHjMTDT+PbOPlM6xS8rzec+HLuOm29eSV+DHMm0W15dw4mgHLo+RUog+HYRw2pL8wVfu4JH/eovnH99DKBgdiz6d2dH5ggkhnL5504wnNadabdRiwUijD0sFw9DGtGyaJtGn6T24al019332ep597B3aTvdhxq2xRsqTn6tCnGm1U1KWy9Y71nPt1tXU1BXPOWQtjFXMoIh0FpCkF2YrUrXsSA81vUeSLEDI4lkceybQQEzf7sjp46dIF0MXshQhkxMWpSKOGNtITZwcsXyy1KVEaIsR46JAjtBcY3JaUKkBbKsJqdJXWNpp2tpIYwVZEeMLQV7O1EnQsmwikRiRaJzu3pGxW3rr9aupKi8Yu8OeccbEmnR+k7GYRdy0EHKiy7JhSFyGRnBcpa7L0PFmuYVFwOXCq6cmWR5Np8Bz9iMFC1hANjDr5ZLh0tl8zTI2Xr6EQ/tPs3fHKU439RIciaC7dMoX5bN6QzXrNtYSyPWOTdZX37CCy69ZhkqEnGbSKNgwdOobyqj90s185FPXcORgG4f3naazfYCR4Qi2rfB4DIegVRVQU1vC8jWVBAIepCYz8mSSmmTl2iqWrfwAjcc7eWdHI80nexgaDGHbikDATVllAQ0rFrFhUx1ev2uMAGq65Fs/+JQT3oQJ1g0zhZSCsop8Pvelm3nfRy9n364mjh5sY6AvSDgcQ9cl/oCHkrI8qhYXUru0jLol6VfPV16/gl8+0zAW7BsllzPB1//2ngklotOFlPMK/Lzv3su5657L6O4c5PD+Vjra+unpHGJkJEIsamJbNh6vC3+Oh+raYuobyli+uhK3xyF52cjzS30p858QT3P8WfvHKKwkWp7xY0qtYt60W2dGASG9ib+lvhaHONmkJanC7+icolMjQagQttlIWvKlhrCTNjfW0YzVE94nZBFSX4xtHk0yznEcTVpq0uA4kye5BFmA1BanPscZItl33OsxKCvJZfWyRdxzx8axfUbTOqc7nK4Kzad7uXxDLUop+odCxOMWeTnOM1eXkvbuIXJzvI7XUH+QYChGRdk4HZogZcGEwPmNzzSaXZtbQG1uAU3Dycv7N5Qsotyfc8EVuyxgATDHOPNopGPdxlrWbazN6D1Sk7jm0GdmtFFsXoGfzVc3sPnqhlkfK90YuqHRsLKChpUVM3rfbKI46Y6n6RrlFQWUVxRw811zq+DRNJlxOXEqOML0mRFCKSUut6SypojKmrPjSTLpDBLeQRfgQ1oFUVa6ik2FHdtHdPCv5/9UrFTVZeP2UdML4IUQSH0ZTk5vcgoy7jTiVXGntU0S2OZJsJN4AQkNoS+b9KJE6KtgMnEClHkyoadKRZyiCRKXZCi5CCHn15gwP8fLliuX8/gL+/B6XFSW5zMSjNI/FOLazUvH9nv5zWPk5fnI8bl59Pm9NNSWUljgo6QowMa11fzol29w541rEFLw5Iv7Wbm0nEWlmVVrSSkpLc7lhTeOsG3nSfJzvRTk+amtKkx77YUeH/etWM+ennYGJmmdyn0BPrlqI6Xec5dqWcAC5oKzYxDxLoRSJir8AHbo/rHXpPcehO/jM3J1VvF9WAN/zOgqX7g2IXO+OsUPyBr6X6joq2P7IfPQcr4OxqUZPdyViqGCP8IOPzj2mpb/fYS+NOlq1A4/gR3857EWGcJ9NVruX5JROkwplHkQa/DrZ3x4hBvp/28Izx3Zn4yEFyHnt3pnvqDsLqZzubbN/VN7sp0jKBUhEwGjNFaCMCCJU7hjODmE0JKlH52Gu0olM1H0Oh5OE+D0x7Mij0zZ2zaPo+xhhMwnGalWZjNKJdfhCa0ixfnNDIahU1zgTxq51XWN27esJi/Xy7OvHGRgOEyO383GNTUY46LZN169nLd2N9LaMUhddREfec8mcnO8COD37ruWB57YxQNP7ALgkjU13HXTGryJ9J7P66Iwz0eqwLMQcOWldbR1DvDQU7sRQnDHljXUVqX3IxPAjdVL+N6Nd/PA0X0c6HPaWa0tLuOTKzeyurD0nDenXcACZosF4jRvUM6kZx4884rdNeOYh1IRsE4xOhkpq5qkCmSrC6xxq2ML7PCDiUavmegXlNMDbXzn+nR9w9QwmKcYE85amUf+lNWENfRX4+6NgfB+GOG+Zl7IjRA+MrsH5x+caNOF1Dw6s8pBx3MqkJSY2FZbwmV9KjFRysQ2j4Ga2upJGsuSVLkJNGMZcdxMjm4pqwVld6NUddJMlW0eT0rsQDhtVtL5VWWIZXWlfOevPpxyu8dtsOXKZWy5cnIkzfGAAqitLuKDt18y8QwTF5SX4+WzH716yntHt992/Spuuz61nkwIQUGub8oxMvmdGlLjqvIariqf2pLmQlzELGABo1ggThcxVORZ8H8G9OynM2cDpRSofuzgv0D8jCWFcF2KDHx+Gv+fuUCHLPRuOxdwiMUFRJwylnK5EfoSVGxqZZ2yO1H2UHKfJXsQZZ5KOpA0VjE5cjTm56SVo6zJfkwmdvwg0tiY5DoUtnUyxeLB7bRayYL9QyYEYtp9VOr9pntvVsafp/eeL6hbVMjffOp2LNsea9S7gHc3LszZZAGZQQ1hB3+a8BA6HxDDDv4EFX6EsfST3oDM/V8ImbmWbMYQGhfsV11F5yAsP58h0FyXJN9k96OsJFYFgLL7UrSeEWjGpSRLuQlZjNRrkx7PiiX3pVLEUWYjSQ1NhSdRUbeAdwPy/V6uXVPHDeuWUFs285ZJC7j4cFFHnJQyGQg9hUuvJm51Eo4fQpcF5Ptux0j4t5j2IMHoDsLxo9gqgkurIM+7FUMrwrKHGQg9jd99KaHYO0TNZnRZTJ53Cy69gvNfbKxQsdchfhBc68/xqVioyNOo0E8ZI01aJTLwZdAWz0MblPEQnP+fVSpcvGdzmicAACAASURBVA2epZ6qz5vCNo+jYTO+EMExA+1EWUn63Ml8hF6b9GskZAFCq046kh3fh/N9nFSVaPc7nlRJIIQXqS1Nuu1sIj/Hywdu20Blef65PpUFLOBdhYubOGHSH3qCmNWO12hAl4X0RV5lJPo2VQV/haGVEI4domfkflx6NQKDnvBPCccPUZH3VSw7SG/w1/QFH0LXinHplfSHHk68/8/HyNd5B5ELKgLEwGrDjj6NNJZP8Lc5m1DKRsV+iz3yXaeKCZx2Ff4vINzXzkgs/+5DJukgI8P95h+pKuGS7qsvTritT+1/p8yjMIk4OdYM+0kWBZJaPUIUkDTiJIyEHYUHmFjhpexelNWMmCQqV3Yfyk5BnPRqkOeerOTmeLnzxrXn+jQWsIB3HS5q4uRAoct8Kgu+gSZyiVktHOv6GMORNyjw303AvYmAexOjD+ihyEu0D/4DcasdKfyAQsocFhd9G4FBOH6Ikz2fJ2aePm+Jk9CqQK9FRZ4ELOf/nrtR+vJzozmwTmIP/9044bnLqS70fnh+24RcDBBuxjqJJoVE930ILZlO5xxAaFVkQuKEEAiZh9TrknolWeZhpmq7FHZS40vHp0ukJDMCqa8E6QN7UhsQFcOO759SjafsXuwUESdpnOPo7QIWsIBzincBcQKvaxVSeBP9csoxZAkxqw2Uja1ijETfYiS6HcvuJ252YqsotoomiBPkeq5GoCd69uQiMbDTVZydYyhspPtWVHSbU51ktaIiTyMCy8/yiSiUGsYe+c64CjqJ8GxF+j6FuFB1R2cTwk36NKNEGuvRfR86W2eUNQgRQGiLIQlxUla7Y1UhxrdJslM4eRsIvQ5E6hYUUl+KED4Uk9sVxRL2BmeE6ErZKKsLkloegGYsRHkWsIB3M94VM9eUCVpIp2pGRegY/Ec6Br+DW6+mwHcXeb5bp7xfk3mMTl5n/nseC3btEYT7KoTr0jMvhR8A6/RZPAmFUkHUyD+hIi+cednYjAz8CciiedY1XRyQooj0P1OFspP3UTzvIQJIvZ6kxFBFHaPL8S9Zp1F2EjNQmYPUl6VP+coipFabZIOFbZ444yc2+pp1nOS/cRdST99HbwELWMDFjXdFxCkcP4pSURAuTKuPuNWFoZdhqwhDkdcpDtxHkf8jCCGJjvw8yREutAnedEwfvfegoq8BMbB7UeEHEYEvgpj/j10pGxV+yCFsowJnrQYt56tnoSHtRQStjPSpLztlBdr5DiE0pFYDwndG+zYGhW2eRBtP/uOHkjZKFiI3oWFKN5ZAujZixV6fsk1Z7Y7HmqxLvGKldgzXa0EutAp5N0MpRV80TNPQAO3BIXojYaKWScwyMaSGW9Mp8Hgp9vioCORS6c/F0BYkCTNF1DI5PTJIR3CEnnCQoViUsBknbtsopdCkxK1p+HSDXLeHIo+PUq+fRf4cPGn6JGYD7w7iFDtM2+C38ehLGIq8hktbRMC9GSnceIx6BsPPockc4lYbI9G3L4IUkgIVRriuBmMtxHcCJnb0JYT3ToSW3A08a6MrG2LbsYPfOyP8lcXIwH8HI1Ul1QKSQcgChPCjVKqokkLZrShlX5Aie6HXIERuklYtNmq8GSsqkaabav4qtDKEVjntWNJI3rJIWV1Oak6vG30hYUWQ5Bj6EkSKlOBgNMJ33tnGO93tSbdfvaiG31t3OQFj/s1Yu0NBvvTKY0St5D3m7l2+jg81zDzl+NSpI/z7wV3E7bl5i11fWcvn1lyG7yzci/EYjkX5wb632dbePOF1IeCbV97MioKSpKRYKUXEMtnf28mTjUd4u/M0A9EwI/EYYTOOaStsZSOFQBMSj67j0w0Chptyf4B1xYu4uWYpywqK8Oku5Ayev6F4jC+98jh9kammrwBf3ngNVy6qySqZD8Vj/O32VzjY15V0+8eWr+eehuw9y5VyutdGLZMDvV0813yMd7rb6Q4Hx+5x1LIwbQs7sa9zrwWG1HAlCJRPN8h3e1mSX8RlZZVcXl5NidePS2pIkZ2+p/AuIU6F/rvRZSHB6A7c+mJKAn+OK1GeXJH3FXqDDzIUeQG3XkdF/v9gIPQUmsxBCg8B92YMbdHYsYTwkuO5Fl2ez34etuOsLPOR/k9hDxwAImAeRkWeB38dYr4+eqXAPIA1/E0YTasIP9L/BwjPbYgLLnp3rqEhjVVY0baUeyirE2V3I7Sys3he2YHQ6pwKNXsy2bCdVJ2yQUiUimPHj5CMOGnGRkQGUVRprEhaxafsbpTdPk7nFMW2mpMfQ1/mRMiSINflJt/tYU93O6aaSix6wkFuq13GqsLSeY9YPdV0ZAo5GIVX11lZOLvClu5QkF1dbcTsmTX9nYyanDysc+BPZimbk4N97OyaamnRNDTAioKSKa9HTJOdXa38x6FdvNRykrhtpRRqWEphKYtYzGIoFgVGOD7YyxttTfxw/9tcWlrJP295L8Ve/wzOWbG3p4POUPL2P/3R5IRqLrCU4kh/d9L7BLCluj5rYymgKxzkxZaTPHh8P3u629Pe41HYSmErRdy2CZnxCT0Rd3S18sDRfXgS3/XbFjdwaWklDQVF5BjuOf/+3hXESQiDkpxPJH24uo3FVOT/8YTXvHl/OPb3ivwvT9hmaEVUF/7P+TnRrEGBijlfDuMShGsjKrYNsLDDD6P5PgxifprtKjWAPfLPE5qqCu/7Ed4PIsT8hk8vTgiksRor+nzKPZyS+hbQSrnQ0spCFiK1Mqwkom9l96DUiBORsrsTffumHCFlJGnKnsKL1JdixydX5pnYZiNaws/JtjoSdh6T4UVo1SlJmhCCW2oa+OH+7YlJcyJaRgZ5p7udlYWl8/opRS2TZ5qOpdy+qbSS6sB8ufRfuOgMjaA48wtSShGzLX64/23uP7qX1pHkxQKZQAFx28anGws9+sbBUjY7O1v54f4dvNJ6MmWEdDawUYTMODu7Wtnd1Ua5P4cbq+v56yu2os2ROC18ghclFGMmk7IU4bkTEhWCWI2o8G/mxY1aKRM7+O+o6EuMlpIL1zVI/++nrXhaQHpors2kW+MoqxM7fshpaXOBQQgNaawhaZNde3hMDO40/k0iDBdupGtdhoN5E21ZpsKOH4aEw74yT5JMGC60QqRWnfRcR7E0v4hNpcnThrZSPHryELEsTg7JsKe7naMDvUm3GVLjlpqGs5IuvNDQHjwTiVRK0RcJ8zdvv8Q/7XlzTqRpFFIILiuvImC4p9/5XYC4ZfHoycN88eXHeK75WFZJ02TYKNqCQ+hSZiXau0CcLkooJ8UBCCER7q3j+tUp7PBDKDvbgmKFir6CCv2MCekU4QLhWiigmyWEEAhtMSKt+DmGFX09qZHkhQCnEXWyyroRp/E0oKyOsb9PeK++HCFyMhzJjdSXk0xsb5tHGDXWdIThSYiTLEToVWlH0ITg/UtXp1zR7u5u53gKUpMNxG2L19ua6AuHkm6vzc3n0rJKtFlGPdy6jt8wLrC4ZmZoDw2jUCilGIxF+T87X+H+I3uJWFMLEmaDYo+P1YVlCxEnwLRtnm46xl+9+fxYpG++kWO4eG/dyqx8d8/bVJ1t28RNRz2vaxJdn3lVgsCgPO+P0ESAdx9HPKOxEFoR0vt+7PgeQIHZ5GidfPdlTVCszEZU8AdTvG9U9HVU+InEWFkZ6l0HIQvRjI2Y5uGU+1ixN7DNU0hj3QVX8SX1lThkZqIuSNlBx8FbWSirbZJlQOK9xmqHnGcAISRCq0LIfJQ9kbwoqxWlggjyEs2AkxGnEoScXke2rricJXmFSaM+cdviqaYjrC7Kvs5JKUVPOMT2ztNJNVYCWFNUxtL82afpt1TVU59bSMiMMRiL0h8J0x8NMxCN0BcJMRCN0B8N0x8J0xUOEs0S6Tgb6AyNgHKiEz86sINHThxKquUaFSVLIfAmBMkAI/EYUcsc095M1nAt8uewrKD4rFzL+QyV0E99b++bE3RJkyER6FKS5/ZQk5NHhT+XgOHCrevELIvhWJS+SJiWkQG6QkEsZWMl7n0ybCqroja3ICu/u/OWOO090sZff/cp+odCfOTOjXzh3mtnfAwhNHyuBc8VAOG5A0K/BPMQEMaOPIbmuRG0LPXcs06AJQANtEVg9QJhnMa+30dzbQBjdXbGerdB+NA8WzCjT0OSqAsAaph48Ae4878NXFhpUSELEVoNyjoxcYMaTdVFHa+lKWRGIvV1TOkzlwZSr0HIsinECeLY8WMIWZZCGC6R+sppRehCCMp9OVxfVc/xwb4pD3FbKV5rPcV9yzdQEcjN+LwzxdH+HnZ3JS8kcGs6761fiUub/WO/1Beg1BcYSwsrzlREOd5tiddQfP2NZ3jo+IHz2fFuAtqDw8Rtm8caD/HjAzumkL4Sr5/lBcVcWlrJ5rIqluQXke/2jFXIKaUYiEY4MdjL3p5Odne1cXKoj9MjQ0TMOGuLyyn3Bc7FpZ1XiNsWPz+8J2XFnkRQm1fADZV1vKd+Bcvyi3FpOkIwobhIocaIbm84xJEB57u/o7OV9uAQ7cFhwonP0Ksb3FyzlAJPdp6N5y1xUrYiFjeJxkwsa26lrwsARA7S+wGn9QlxiO9Dxd4Cz/uzGgkSnjuRvnuxg/+GiiaML+1u7OCPkbl/jjivqxHPTwgh0FybkXoDduytlPtZ0VewIs+he+5yTF4vFAgdaSzHmkyciKOsXrDD2FbT1PfJQqRePaOoqZAVCK10nJP9KBTKPIoy1qDsZHoWmdBHTf9j8eg6V5RX89Dx/fQmKSFvGh5gT087i/zZ9YOyUTzfcnxsspiMykAum8uqsrJ0GT1v4fwj6T5zFeCebfRHwuzr7eBHB3YwEj/TGcKj6dxQVc89Dau5vLyaHJcn5T0s1w3K/TlcXVFLzLI4OdjHru423u5o4abqJRdcNHg+0B8J83TT0ZTbr6pYzNc2XceqwtKM05oVgVwqArncUFlH0IzTONjH/t5OXms7xZvtLRR6vFy5qGZGNhDpcN4SpwVkF0Jo4L4eIo9DfA8Qxw79FM1zGzBaXj2HyVb4Eb5PIP2fRsgCpPBhxXcnIiQ2KvocKnIpeD+60J9uFhAyD8P3SaKx3YwZik6GGiE+8l2ELEO6LruAfJ10pL4aiyenbFF2F0oNJ/VVkloFYpxVSCYQwofQGyD6KpNTg7Z5GM3udqw8kp2jsTrjcS4rq6I2t5DeyNRy7oFohDfamrm+sh6fkb1K06FolGfTVNO9p37lWfdNupAwEo/yp68/w8nBM1HdQreXL196LXfVrSDXNbMydpemsaKwhGUFxbynbgXuBRNMwLEKSGWhUJ9XyF9cvoVl+cWzIplCCAKGi7XF5awuKuPOuuX0RcJ0BIdZnJu9xtwXypN1AXOGAK0O4bmVsdRG/Cgq8hSMFuHOwVFcuC5DBv4ARjvU6yuQvs8CiQoSFcYe+b6T0lvALCDQPDeiua9Pu5dtHiU69A3s+I6xAoF5g3LavVjxvag59W7UkEYDMHVSV3ZXQhg+Nazv6JVmRpwQAs1YR7I1o22exB7tkTdlrFKEVpHxMLkuN7fXLksZmXj59El6I8GsVkI+33IiaYQLHGHyTdVLFhLlaWApxfHBXuxEcrEykMs3r7qZe5evJ9ftmXW0SApBjss9pxTpxYSDvV0pv/e31CylLks6JCkEuS4Pi3Pyuby8Gi2LC8kF4jSvmPzhz2Yiy96DVQiB9LwX5KjJWwwVfsJJhyDn6LNkIITrTAhfSIT3PQjX5jO72B3YIz9A2ZNdoheQGTzo/k8jtJq0eynzCLHBP8MMP4RS4axNzkopp5WOMrGtNuLhXxId+CPiQ38zp4o+IURCeD3VfFDZvdjmQab+DjSkVo+QM9csOILyqZOYsgdR5jFUkoiTNFYyk8flqKdTqrL/0yOD7EyhRZoNImacl0+fIJ7CmHJTWSVVgdyFVFGGCBguvrB2M1trljqO0+f6hC4i9ESCSWc1TUiqAnkYMruROZFFx/BRLBCneYNgyqrWnoXDqz3E+ElDCDfMJdUly5C+jzFK6lR8F2qsf1c2w/gCZDnS/3swrhJJRV9AhR9GJek5toD0cLROV2D4Pw8ivcjUNo8SHfw60f7fx4q+gm21ohI+RZlDORVt9hC22YId340Zvp9o/+cJd99BbPAbWNGXElGauZEzIUuTtk1RVgdWbHuSN7iRrg3MpthAaDXJq+NUECu+M2mqThrrmenjstwX4Kbq5DYSCvj1sf1YWYoKHujrYl9PZ9JtPt3guso6CtwXVtHAucTNNUv56LJ1uBeiRGcNo4ajFwLO2bfCthWdPUPsOdxKR88QylYUFwZYs6yCmkUFjuBwGpaolCIWtzjZ0kPj6V66+0aIxUwMXSMvx8viigJWNSzC454aSYlE4zz0zB6GgmFKCgK8d+s6jGksD46d6ubV7ceImxab19VyyaqqNExWIIR3IrNWA8w4gjS5+kd4mcvHJgA8t0D4QbAaHa+c8KMI940Zl3VnPJYQ4NqM9N6DHfxn50U1gh36TzTX5aDPb8+8ixFCaOi+e7CtRszgT0gfxYxhRV/Cir2FNFYjjY1IvQGp1SC0EhA5DhFHAKbTCFsFUfaQ49ptd6PM09h2K8pscirb1NyNAJNel1bk6JUmcTtldWIljVC6ZqQ5mgiJNNZiWRN1U8ruw4q+xZh57Bj0hP/TzIiToWncVLOEZ5qPEjanLhQO9nVytL+XVUWza38yCtO22dXVRlsw+WdT5gtw1aLFC9GmDJHn9vC7qzctpNbmCfluL4KpM6GNY1MQMuP4z3Mt3jn5ZsTjFk+/dpBfPbmLju4hwlHnael26RTm+blry2rWLKvAZaQmMuFIjFe3H+fh5/bS0T3ISChGNGpi2TZSClyGht/rZnl9GZ/+4BWsXFKOlGceHIaucbixk+ffOIzPa3DZusXUVKSu+LIsm2dfP8QvH9+By6WzfkV6IzyBRMkCnIetM7mlciROB2VOEnuKwNwIjhCgVSE8t6OC/wqYqNibqPgunIjTmfPNCoRE+O5FxH6bGAOwTmAH/wWZ9/8ishrlerfAg8v/38Aewgw/whS2MRkqhB3bjh3bAcKHEH4QbsBIVN+JhB7KBkxQcYdEEU1EX+a/oFzgQeq1WOhMJC4xUEmML7XqjDyVUo2muS7Bijw66fUYJNFqOdGwshkTDykEa4vKWF5QkrTx71AsyounT7CisGRO1T4j8Rgvnz6ZsvHu5vIqanIXWqxkihsq66jPW6j+nS805BchEKgkz5Vnmo5xd/2qrFbAzQfOeqoublo8/vJ+vv3jFznR3IMQgvUrKrn60nrWLqvAtm1+8uCb/PLxnWm1GSOhGI+/dIC9h1sJhmJUlOaxaW0N1122lM3rFlNSmMPAUIhtu07yN99/mraugQnv1zTJe29ai6FrhCNxnnzlQNrx2rud6JhlK1Y3LGJJzTRGZgKQRSALzrxmNYGdua5BqRAqvnfcKwZCX8yY4HqWEBhIzx2QaHQMMVTwJziO3/PApWUZIvD7IEdX1goVeQYVfgzUhRGaPZ8ghABZhCv3zzD8nybz74NKRJS6UFYLyjqJMo87uh7rBMpqdF63O0D1J0TSZ8mFRwgnqiMyuxbp2sTsH18iEa3KbCyhVyPE7CbS6px8riivTlqaH7dttrU10xmam+N701A/b3ecTrrNkJJ7lq7JqjD2YoZH07m+qm7M1HIB2cfmsmq8KapJB6IRvv7GM7zYcoKQGT9vPcDOasRJKcWxxi5+9cQuYjGTksIAn/vw1Wxev5iAz00kZtLc1sfPf7OdN99pTHvTigv83HrtCuqri9i4qpqqRfkU5vvxuHRicYvWzgF+/pvtvPTWMU6d7uPlt47x8bs3TzjG0poSlteVsv9YOzv2NXPPrUGKC6dqR5RSnDrdy8mWHnRNsm5FJYX5/mlWoAIhyx2yMJpuU5GEY/enp129KqUgtheslnGHzAGtfu4hdyEQ+lKE+zpU6BSgUPEDIHJJ1o5irnCaDV+O8NyBCv0UJ7IRwQ7/J8K1FrSGhTTCDCGEAFGAEfhDEDnEQ/8+Na17gUHqSx3ipKYvHtAybOybDI4YvRChVaKsk9Ofl1aJkLMrZdal5Iaqeu4/sofBJI1/D/Z1cnygj3Lf7DydlFI83XQ0pTZkRUEJKwqniu4XkBxlvkDW3KUXkBylvgDXVCxO2Yi6aXiAP9v2LFtrlvKJlRsTESoH58vnclaXIXHT5tXtx2lu78Pj1rn3PZu47bqVlBQG8HldFOb5WL+ikq98bitlxbnYdmrqJITg1mtX8Qcfv47rNi9lSU0JBbk+vB4XeTleVi4p54u/cz25fg8Ab75zasoxcvxubrxqObouaW7rZ8f+5qRRp7hp8frOE4TCMfJzvVy9sR5dy+DW6TUIfQVnBKwmdvhhMPej0olClQK7Fzt8P4xvbKpXI1wbpx83Ewgd6fsdGJ0Q1AAq+iIw9eGeleGkD+n/wrieeUD8AHbwJyl8cxaQEUQORuALePK/i3RtPgfNlDWHVGQhoiH0OoTIwE1b5CCNFXMbSxQg9foM9tQRWh0I3/S7psCmssqUrTb6oxFePH0iaYuUTNAfDaecgKQQ3FW3Ap++kA7PFCVeP5X+7Du6L+AM3JrGfcs3UOL1p9ynIzTCL47s4WNP/ZIvv/Ykb3W20BcNY6VIR59tnFXiFArHEuQEyotzuWpjPbquTWCRQggK83xsuWLZtMczdA2XoSdloUII8nN9LK50Qux9A0HsSTdd0ySXrKqiqiyfkVCUt/c2EQxP1TgMB6Ns2+WsTBtqS2mozWwFJ4Qb4b1j4mRmHsEa/nai9UlyYqhUHyr4Ayc6NbaPQHrunvXKNym0SoTn7nEvRMmqvmkyZBEy8EUQo3oLGxV+FBV9ef7GvMjhlNq60NxX48n/F1yBP0YalzD/P20dqa9A9/8urpw/daKhc4aRKPtPD6nXg8yb2+pT5iH1OqatyhM+pD43x2ddSD6wJLWQ/dmm44zEZueD9UZbE12hkaTbqgK5XFZetdBUdgbIc3uy1pZjAckhhOCy8iruXb4OVxrrAVspeiIhHjx+gE8/+yBfee0pfnxwJ4f7ujHPMYE6q6m6aMzkVKsj9CwpDFBRmlywqGmSZXWlCDF9lbPjLeOIty3bxk78G6WIxEyMhMDcshWWpZj8DKmrKmbNsgqa2vrYsa+J7t5h/F7XhAfljn1N9A2EEMAt16xEn4EDrHBdiXBdg4o+h0OCbIi9gTXQgfDcgnBfi9AWgzDA7kXF3saOPAWxXYyP/gjXZqffXFahIzy3oiLPgT3V4TjbcKrsrnFSduFf4WiqotjBf0UYqxL34fwIxV6IEFoRuv/TaJ7bsWJvYIYecjyQVJiplWIzOjKggTAQIh/NfQ2a+0aksQKhVSCEJ0tXgFPtFnkq/T76UsQ0dgzTQQgdodc7i5okhpdn9vNmGJlKjysW1bDIn0N7cKqeqT04xJsdLdxeO/1icTzCZpxt7c0T2oOMx5qiMpblLzSVnQnyXO60k/kCsgOfbvCZ1ZtoHxnm4ZMHpyVCITPOCy0neL3tFD8//A5ri8q5o245V1csxqvr6EKe1TTeWSVOgyNhQomITnlpXtp0V17Ag8/rIhhKvRKzLJuTLT0cPtnJweMdNLX2MjAUYSQUIRo3icUsorH0E4aha9x+/Sqe33aEnv4gr7x9nNqqM93DozGT5944jK0UVeX5bFqb3nxwKtzInK9g292o+B6ciI4C6zgqeAIV/BeciWm0QDOxfQwC9JXIwJcdsXkW4WiP1iLcW1DhXzCv0aaxMQNI/2ew4nvO9AszD2EHf4TM+dqsJkQhS9HdW1Ek77QtZDEiKxGRNJABNPd1KLs7xQ4uhFY+v+cAjompVoXu/Qi69x6U2ej0sIvvd4TfagDsYZQaARXHIVQ2ToRKcyo2hQch/AgRQMgChFaO1FcgXZcmBNxeHMNUmI2PUjpormuwPXvT7+O+EUTqMH9mEEjjEnTPXag0FgtCliD0urmNJATlvgA3VtXz8yN7pmy3lOKREwe5qbo+4xJ4BTQNDfBWR0vSuLVH09MacC5gKgTg011ndQJ+NyPf5eGvrtxKfV4hvziyh5aRwWnfE7UsGof6aRzq54lTh6nJyee2xcu4oaqelYUlM26LM1ucVeIUjjgkSAC+JN5KoxBCoOkypa+SUopwJM4TL+/nkef30tzWj2XZjrV9wEN+jpfiggCGrtF4upeRUGrdjhCwaukillQXc+B4Oy/89ggfufNSvB4DpdSYRxTAVRvrCfhmVtEmhABtMTLnG9gj/4CKvTb+SpiuakkYlyBzvoow1s1LNEYID8L7HlTk8YTP1DxDCIRei/R/BnvwazhRJwsVeRLlvg7cW2f8xddcG9Bc35mX080UUqvAnffNc3oOozhz/3SE0YA0GtCV5fgy2d0oexDsQSCWMMV0iJMQOuB2LAtkbuJPiaM7SmiYBrqHuP9//5xokgVNbnEOd31+K6XVsyP4jsHnejTXD2b1/plCM1ai5X/rrIzl1Q2uWFTD441HGIxNJfiH+7s4NtDLqsLSjL7/Sin29XbQNJz8N1vk9XFdZd0CCZgBBCIz7eoCsoLRvnKfXbOJjaUV/OjATl5pPUnUyqzS2lKKxqF+vr/vLR4+cYDLyqq4dXEDN9c0jPUFnK/v/1klTuNTXNOKvNLwCdOyefi5Pfzk178lEjWpLM/njhtWc9Ul9RTketE0iZSCaMzkG//wGPuOTvVQGQ+XoXHnljUcOtlBW+cgb+05xfWbG7CVYteBFnr6RijI87F53eK03lIpISTCWIvM/3tU+FHsyG/AanVSBCqGQx7gzIrfC3IR0vs+Jz0nS6cnTVoZaImUgsybkdeTMDY46bPYm5M2pDmGyAG9LhG1SIyf+YgI91aE96Oo2G/HXlXhRxDua4HspX4W4EAIzYl4zTHqFR6J8OqDbzHURCfE3QAAIABJREFUO4xl2hOKKcprS7jug5fPmjhdzBBCcFlZFfV5BexO4unUNjLMtvZmVhSWoGUQxYvbFo81Hk6Z4rh1cQOFC1qdmUE45GkBZxcuTefy8mpWFZWyo7OVH+7fzv7eToZj0YztCDpCIzzeeJgXT5/kZ4ff4dOrNnJFeQ35c+gxmA5nlTgF/G6kENhKMTCUvpIqFreIRJOn2foHQzz4zDuEInGqyvP5iy/ezqql5VNuUDgSJ01h3hiEEGxcXUVtZSGNp3t5dftxLt9QSzgSY+f+ZuKmTX11MSvqZ26CN24QhChA+D+J8H4IZR7g/2/vvePjuM57/efMzM72Re+VAMEKir2TokgVkmpWsfJzkSXHsuzIvo6v49w4sVxunJvEcZVLHJfEsSMXlahLlq1eSIm9dxAkAKJ3YPvuzJzfHwtCBLELgiJAitE++sCWdmfPnNmd3fnOe973+2IcTyztyBCJOJwToeaDVonQ5iCU8S9HqL6vvLt5kbioqhnfOL/XOK5HquXEY28BGrr9Ks6rl5fiQc34+/PaZ5pLT0aul7u//kH6u/2EBkL0dgyw6+X99LT2XeqpvecpcntZWVzJ3u52rLOSN2OWyebWRm6rnk2O89wVfE3+fran8G5y23RuqJw+IXNOk+ZiIIYa8q4rq2Z5UTmvN5/kpVP1vNnSQPs4fc4kEIzH2Np+il2drVxTXs2fz1zI/Pxi9PPISx4PF1U4uZ06uVluOnsDtHf7CYVjuJyjoxqWJenpDxKNJXdE7u4N0NGdyEu4YkYJM6uTCxrTsujuTV5xcjYFOV4W1pZzsrmHoyc6ONXahyUlB+vaUFWFhbVlZGW8+5LkMxGKC6EvBn3xhIx3qYjHthIc/AZgR83MQ7NVX+oppZlkXF4nG/58LZBYLupq7qGruSctnMbJjVOm8+8HthMxR98U7uxooSU4SLbDOeYNmpSSPzYcSzoGwPy8IirSXkRpLhBzgpqDny9Ozcb6ymmsLK7gWH83bzSf5KkThzkVGMC0rHFFoeKWyR8bjnGgu4O7Z87nYzMXYNe0CYsnXtQFXafdxuxpRQC0dw1yoK41uW9S3GDL7pMpK+rC0ThSJlavMjxOlCTltlJKdh88RVff+ISTrmusWlRNls9FU1sfh4638/buk/iDUTwuO9esmDHpP0RSWkgrhJSxCetonybNZCGEQFXVEa2M0oxNdUY2y4uSF5gMxqP84eTRc47RHQnxSvOJUVErAF1VWVtanV6mS3NBSCTxS9hwVwBe3c6CvGL+ct4Knv3A3fz4qpu4pXoWVRnZ46p8lMCpwADf3PEG/7LjdfqjE+cXeHGFk8PGkisqcdht9PYHeeblA/QPhkeIBCkl2/c3sXVfQ8pxcjLd2DQVKeFkc/dwpd6ZY9Q3dfObp7aPaaJ5JkIIamuKqCzNwTQt9h1tYfPOegDmziihKH/yTdGk1Y1/4CtEI89zMSrc0qRJc3HRFJXrK6cnbcEC8GJTHcH42L0H93a10TiYPCm80OVlSWFpusVKmgtCyoTdxaVGCIGqKHhsOhsrp/Pt1Rv5wZob+dKiK1lZVIE2jvPckBYPH9vHQ0f3EknSbPvdcFGX6hRFYdWiajbtrOetnSd4fXsd/YMhbr56DkX5mfiDEbbubeDVLcdw2nUUIfAHR1fEZWe6mTuzhB37m9h7uIWf/PYNblw3h0yvE38oytbdJ3nutYP09AcpL86iaZzLCE6HzsYrZ7H70Cl2HmiifzCEqiqsXz3zojQcjMe2EA0/jKJkYXdcz2S0P0mTJs2lQwDz8oqoyczhSF/3qOdbg37eamvkuoqa0S8GoqbB1vZT9EaSe0/NySlgZrrFSpoLxJKSgejkdJG4EGyKyhW5hczOzufWqbPZ393O747uZWv7KQaikZTLeEEjzs/2b2dBfglLCkovePXoogongOwMF5+980pC4Rj7jray8+Apdh58px+bEILZNYV89ObFPPTsTvYeGW3M6HHbufMDS+jtD3HiVDePv7CXx1/Yi6oqmGYiUpOf4+VTH1qJpip8+99fGvf8Viyoojg/k5aOxB3dzOpCpl9IUvh5EI28Nun7SANHd57g+O6TeDLdrPzAYjSbSjgQ4cT+Jrqae4gEozhcdvLKcphSW4bLO/5lD8uy6DrVQ3NdO/1dA8SjBrpTJ7c4i/IZJWTkjr8nmWVJ+jsHaDrSQl/HANFQDN1hIyPXS9mMYvJKc8Ycq6u5hz2vHQIpmbd2NnmlORhxg9b6DpoOtxAcDKNqKlkFPipmlpJTfOnyYuLROO0NXTTXtRHoD2KZFk6vk/zSHCpmleJwXxx/lslGCEG5N4MlhWUc6+8ZtdwWNuK82HScNaVTsCfxdOqPhtnU2pj0AqEKwW1TZ6OlDRzTXCAR06AvmtoY9lKjKgrZdidrSqawuriSTa0NPHhkD5taG1NGyvqiYX6+fzvz8y48WfyiCychBBXF2fztp6/jta11id51rX1IKSkuyGDVwmrWLZ9GfraXFzcfQdOUUTlMihAsmFXKl+9bz7Ov7GfXUC6TpipkZ7pZMLuMtUtrmDO9hP1HW8j0uVDH6c/hdtlZs2Qqv3tmB4oQzJtZSm7WhZrtnQuJtEIY8T2TvJ80AFue3cnv/vlJMvN9LF4/l9b6Ph7+9lMceOsY/Z0DxCJxdIeNrPwMZi6v4fbPXU/V3PKkuXSnkVIS6A/x4oNv8PqjW+hs7ibQF8SIm9jsGhm5PkpqCrnhnnWsuHkRipra6VZKSSwc48Xfvsmmx7fRXNfGYG+AeCSOpmt4Mt0UVxew5o7lrPvQCtwZrqRjNR5q5hd/9zvCgQh/+6vP4HA7eOanL7Dpqe20n+wiEoygqCrebDfF1YXc8tn1LL9hAapNvWgiRUpJ16keHvvh8+x/8zBdp3oIBcJYlsThspOZn0HN/Epu+8uNTFtYNdRi5vIWUA7NxqriSp6sP8TgWY1/JXCwp4PGwf5Ec9MzjlVKycmBPur6R0eqAMq8mSwsKEkX1L/PSJbrdqG0Bf2XvK3JuTj93VCF4MqSKczKLuDR4/v5/q7NKZte7+xspn6g94KjshddOEHigEsLM/nITYu4Y+N8LEsiAVURaJqKOpRs+pXPbMA0reG2KWeiaSozqwuoqczDMBKtVsTQ2JqmoA1dmObPKuPRH96TMODWzi2ehADdlnhbMjNcLJtXmdKI890jE4ng0o+0BrGsAYz4HiyzHQDLaiMe2444e6lO2FC1KpRz9KuTVgjL6sCy/ECMhLmhB0UtQAgvYpz5D1KaWGY70uobcuVWEUoGqloEOMZtyJk41kEsqxc5PCcB6AjFg6LkoyjJHcOljGHE94K0EhYG2qxzXjiljGDE9gESofjQUvQ/8/cGePuZnTz+o+dpONhMRq6XnOIsLFPi7wvQ3tBFe0MXh7fU8de/+AtqVyYv8ZZS0nK8nf+4/yG2/XEPRtzA7XORU5yFqqnEwjH6OwfobOrm2I56Gg42c9vnN+Lyjq6eklLS297PL7/6MK898jZGzEgIiFwfNoeNeNTA3+vnwFtHObLtOHtfO8h9372L3KGejMmIhWOc2NfE6/+9hc1P7cDpcZCR6yUz30fYH6G/c5Detn4aDzXTfrKLD3zmOnRHaoPaicI0TA5tqeMnf/VrGg42Iy1JRq6XgvI8EBD2R+ho6KKtvp3DW+q4++t3sPbDK1HVy18arCquoNybyYGejlHP1fX3sK+7nalndIU/zXMNR4knuaAJ4OaqGXht52fQm+a9j3qO4otAPI5kYj3893e3j9tD6b2AEII8l5tPzl6EYVn8aM/bScVTMB5nS1vj5Smc4HRFjhgzEuQYw1389Bg2TR1T2KiqktTyIBXhSJxNQ0nh5UVZzK4pmvA7XCnBMlsIDv4TptmCaTYhrV5OJ4RHw08RDT816nVCycWb+QB2x9oU4xoY8QNEQ/9NPLYd02xBSj8IHVUpQNPnY3dsQHdcl2jNMQaWNUg0/BjR8LOYRj2W1Q/CjqqVo9uvwuH6EIKxl7CktLDMJqKRlxPzMeqxzBakDAJiSDQVYNMXYnfeik1fNGpeUobx9/81pnEcVa0iM/dxhDr2SR+PbmGg9y7AxOG6C2/mPyfdzjQs/vPrjxCPxLntcxuYf/UcsvIzMOIGLXVtvPDgm+x8cR9tJzr55Vce4iu//zzZhaNFazQU41dff4S3nt6Boios3TifK29fSsXMEnSHjr8vyIHNR/nTr1+jtb6DJ378R3JLsln/8TWjzq3QYJiHv/PMsGiqnlfJ9X++lqorynFluAj7w5w62spLv93E3tcPsfnpHbgzXNz7zx/Bm526Xc3z//kqwYEQK25ayKpbl1AytRBVU+lp6+Otp3fw6sNv4e8N8Oj3nmHawilcceXMSY3sSClpPdHJz7/0W07sa8Llc7LqlsWsuGkRBRW5CEWhp7WP3a8e4PlfvkrnqR5++bVHKJtezPTFl7/thdums7FyWlLhFLNM3mg5yQ1TpuPU3vk++OMxNrU2Jh0v1+lmRVFFyqTzNJcvuqKNmWPrj0USBVYT+Nnv6pz83qWTga5q3FI9i9ebT7IjyTHELZP6gd4L3s8lE07vRaSU7D3SzMlT3QgB16yYfl6i63ywrAFiZ7hmC+FJiBwkCCdCuDj7HkJRslMKHikNouGnCfq/h2U2kogyJXqOgcQ0WzHDDcQir+L0HMHpvg9FGe1LlWia3E/I/z0ioYeHRI6OEE5AYBmNhOO/JB59G7vzxjGPUUo/ocBPiIQfAxkZGkcfmhNIGcU0jmEaR4lF38CT8U/o9rUjLthCeLA7byfk/xam1U4s8jJ21/83xjKXSST8OAk3dhsO1+1jzjHQF+Sur93ODfdeg+6wDY87dV4lM5fW8E93/ZgjW49zfE8Dm57cxk2fuhahjFw+ef2/32br84ll1sXr5/KFn95LZq5vxHYzFldTPrOY737q5wT6gjz6vWdZdcviEWJHSsmhrXW8+OAbGDGD6Yuq+Mz372bagiqUM24wZiyZyqzl0/jJX/0XO1/ax5uPb2P2iulce+fqEdudSU9rH6tuXcLnfvjneLM9w8dZObuUWctqcHocPP6DPzDQ7efxHz1P7aoZkxrZkZbkyR//kbrdJ1FtKh+47zru+KsbcfneicJNqS2jdtV0fNke/vNrj9DT2st/P/AH/s9//MVFiYhNNhsqpvGz/dtGLdcBvNnSwGAsMkI47exopisUTDrW3NxCpmWOnfOW5vJEFQKfbk/aIBqg0d+PKeWElRJ1hgIc7O2coNEuPsVuH3PzitjZ2TIqaiaBwVgUw7LQxki9OBfpmtUz6OoN8NuntmNakvLibNYuO79u5eNFCIFmqyWnYPfwny/7p4ihJTin625yCnaMeD6nYDdZeS9j01cmGVESj76Jf+DvsMyTKGoxLs/n8GX/mqy858jMfQRPxt+j2eYj5SChwL8RDv4cKZP5WlhEQg8RDj6IlCFUbTpu39+RmfsYWXnP4sv6BQ7X7ZhGAyH/D89xnD5s+gJUtQq78wO4fV8mI/vXZOU9S2buM/gyf4TdeTOgYpmnCPm/PSSwzkTB7liHolaADBGNvIi0UldJmkYd8dguAGz25aha8uqk05TPKGHdh1aOEE2JuQvyy3K59bMbsDt1ouEYu14+wEDPyB+vgW4/rzz0FrFwDG+2h49++VYy80aKJgBN11h+40JqVyTOqfaGLrY+v3vENpZp8fx/vEJoMIzu1LnxU9cwfVH1KDEkhKBkaiG3fOY6svIzCPnDvPL7zfjH8CzTbCp/9sUbR4im02O5vE4++L+vJzM/A4DDW+toPJTclXqiaG/s4o3HtiItSXFVAR88SzSdxuGyc+1dV1JSk2gVc2T7cep2n5zUuV0sCt1eriqtSvpcXzTMG80Nw/9tWCZvtDYQNEb3CHSoGmtKp5DtmBiD3jTvPYrdqe1w9na1YciJyUeSUibculOItMsBVVHIsjtQUwgjw7KwLvD9el8Lp6bWXjq6Buns8bN1bwMP/OpVDtd34LBr3HrtXHyeyeuZlkhyVYb/Rq9Qj3z+nb8kDulmJ6HgT0EGECITj+9ruLxfQLcvQ9WmoNnm4HB9FG/mt9D0BSDDRIK/wYgfG2W0KeUA4eCvgfjQWH+P0/0JNNssVK0K3bEGt++rON0fQ8qxqy6EEOiO9fiyHsCb8W1cnnux2ZehalVotmrszg14fN/Api8CwDCOYxrHR42hqJXY9KWAwIjvxjRGzzsxd4t4bBuW1U6iDcw6hEi9fAVQMauEjDxf0vdVKIKyGcUUViaWBtvqOxjoGhyxTXNdG+0nE3dnNfMryS/LHfOuv3bFDCAhkg5vHXmskVCUg28dA8DptrNo/dyUYwkhmLa4enhuR3fUM9gTSGmcmleWQ3EKh30Ab5aHWcsSIjMainFiX/IloYniwOajRIIJkbzgmjm4PKl7SjlcdqbOrQRgsMdP87HW897fkeZOHnjqTQZD7wjzrcea+OEzm5BSsu9kG798cTsHGtv53pNv8P8efpmntx4kEpsY35dkuDQbq4orRkSVTiOBl0/VDyf+doaCHOrpTJoI7NPtXFValY42/Q+mOiN1DmP9QC9NKXy9zgcpJX3RMC+fqicQHy3QLxukZDAWTdkP16nZLrjy9H29VPePP/kTDS09gMAwE73xNE1hw5WzE95Nl4EjspSSePQtjNheQOBw3Z4QDJwdpVBQtRm4PJ9msO8AltVGJPQwnow5nCnaYpHXsMxEE1K783ps9hUIMfIkUxQvTs+9REKPYlljN1BWlCwUJSvl80LJRndeTzy2BaSJaTaiMeesubuxOzcSi/wJy+ogGnkBTV8InJUPZfUTi74OMoyqTUPXl4+ZCC8UQdm04jEvOJm5PnKKsmg83EJ3ax+BgZFisbOxm76OAQCcHgetJzroaUsdEYtFE6WyUko6T3UjLTkcnTp1tJVwIHFhL6zMJzNvbNPVjBwvBRV5HN56nHAgQuOhZkqHnPnPpriqAM2W+usuVIUptWVsenI78ahB64nRuTcTSf2eBuJDosTldVC/rzHl5xCPxpFDd4jRUIz+zkGklOclFDr7g7xx8CQfuWo+p9/Vpq5+Nh06yf+6cSVtfX5++9ou9p5sZXZFAbpN4z9e3E4oGudDV867oGNNhSIEiwpKmZqRzf4kuU6H+7poGOxjii+L+oFejibxfQK4sqSSUk/GpMwxzXuD+fnFcDD5c2Ejzu+P7eX+xWsvqMxeAi80HeflU/WTlhh+etzJvLJ2hoMc6etKegyaUCjzZlywL+P7Wji5nDpxwyIWM3DYNaZW5LJqYTV3bFxAxnl491xaYhjxA0jpR4gMbPoShOIg2akphMCmr0ZRsrDMcEJkEAPeiazFYztJ5AeB7riGVEFJITLR9HnEImMLp/GgKKeTveVQTtXoeev6CjTbVOKxHUQjz+Ly3DciSVxKiWk2E49uAwSabR6qrYZzfUXdmWMvb+hOHYcnUakUCUWJR9/xCJFSEvKHiUYSd2ebn9rB5qd2nPuAh4iF48TjBvpQEcRgT2A4opCZ7ztnp3YhBJn5GQghEneLnQMpt3X5nKOWD0eOBb5cL5CIhoXO0YT7QrAsyWBvADnk6v/7bz7F7785uhgiGVJKIqEoliUnPAfLpqncsqyWq+ZUYUlJJBZn06GGSRNOAJXeTBbml3Cwp5Ozu3D1RELs626n3JvJrq5W+pK0jNAUhVuqZ6ctCP6HMys7n2yHk97I6HNAAs83HOOqkimsKa16V6JASsnW9lP8YPdbKXsgTgQn+ntoDgwyP78Ynz7xFaCWlLzWfILdXcmvS7qqUptTcMH7eV8Lpy/es47BQBTTtNA0Ba/bTn62N6n9wXsVKSOYZhMAQslEUYsYSywI4UJVp2CZrUirG8vsRNXKh8ayMM1GEl9FDVWdMsaeBaqWPD/jrAkiiWNZvRjx/RjxgwmLAzkw1JcvOiJqJVPc6wjFjd15G/HYDiyzjVj0VRzOO0ZUksSiryJlLwgHdudGhLjwL6aiiOGolWVaI5fCJBgxY/g2StVUNJs67tspza6NGM80zBFjjQdVVRL7k2DEx+otde5JKUN3qxKJZUyeh4s0rRFztdm1lEntyVA0hZSNLC8Ar8vBtJLEUqsqBFkeJweb2id8P2eiKgobK6fzcN0+wme1gwjEohzs6WRtaRWbWhqSvr42p4DpWbmTOsc0lxYhBDkOF0sLyni+8VjSbTpDAb6zaxNZDidzcgpT5vecjZSSqGnwSvMJvr9rE63BwXO/6AI43NfFd3a+Sb7Lw1WlU7ihcgb5Ljc2RUW9AI82KRO99V5vaeCB3ZvxJym4gET16aKC0gs5BOA9IJxM0+JIQwe6TaO6NGdMk8HzHtuyOFjfjpSSWVWFo2wLDEtiSItpU/Jw6JdplY40kFYi0iCEfbhibSyEkpN4KRaW1YvKUNNRGR1OzhbCPTRe6hNZEefwk5ImpnGUcPC/iIafQcpBEspARQgHCB2BhpTjW0/XHetRAj/HMhuIhP4b3bERRXiHdhYiGn4CAFWbgm6/clxjnl4aS0U8Zgwvr9md+sjlLgEOtwNFVbBMi9W3LeWGe9clxNM4cPtc2PR3xvNkuod1YHBgfK69wcHQsIjwZqb+7CPByHCEJykSwv7E3ayiKDgmMb9P0RSc7ndE7Z3338YVq2eOW3DmlmSfl9CCd/T1mUI1Fjcxz3hPNEWM+B1IRPLOazfvioX5xczIymd318jcLQkc7u3kSF8Xe7tH30GrQuGasqlkOUYn1af5n4VPt7O+ooZNbY1JRYEEDvR08OmXn+Rz81awoqicMm8GthS5PJaUdIeDHB/o4an6wzxRf5Co+c7NjAAcmjZKzF8olpT0REKcHOxjW/spfrTnbRbkF7OiqII5uQUUuLzkOV1k2B3n7LcoSRRNdIWC1A/08syJw/yh4Rj+eHLRpArBR6fPJWMCIl2XXDiFonG+8+Cr5Gd7+L+f3ojTPnHCKW6Y/OrprRimyT9+9sZRwunB57ZT39zN1z+1gSklORO234vP+f1onlFMn+TZ81iFHsMLSkqJGT+Ef+BvMeL7ANC0GdjsV6FqlQjFhxBOBBqx2FuEA/96zt0levitJxz8GaZxHCO2C5v9SgQQj23DNBO+HQ7HB4Bx2EhISXdz75j5MqHBMIGhajVPlhuH650vnRACb7Ybt8+Jvy+IpqtUz63E7Xt3y7yFFXlDkaY4HY1dGIY5QlidTTxm0N3SS8LCRVA4JT/ltj1t/Vhm6iiSlJLOpkQOjaqpZE1iU2shBHmlOaiagmlY2F12Zi6tGXMp8ULxOHQkkq6BIPkZHsKxOCc6eogbl64D/GlUReGmqhmjhBNAo7+Pl5qOE0/y2RW4PCwuKEl5cZwIRjRgJ5FPE4rHCcSjBI04wXiMYDyWeMw449/jMfZ0taXMldnX3c53d20iy+HApel4bDouzYbbpuNO+u+24Qvp+1EkCiFYVVLJ3NzClF5eAO2hAN/Y+gozsnKHopF55LnceG06QgiC8RgdoQANg/3U9XVzuK+L7nBw1OdUm1PAvLwiHjwyed0sJBAy4mxqbWRTayM+3U6Jx0ex20ehy0uR20uBy02G3ZlI6BYKhmUSNg38sQjtwQBNgQGaBvs51t9NT4r+jaeZm1fEjVNmTMjcL7lw0lSFmVMKyM/yoJ3nXeSFUlOeh6Yq+NyTd3c96QgNoSQuclJGz1npBmBZCQMwgTLShVzYE38kjCcl8TFFRcIFPBVxQoEfJ1y/seH03IPTfQ+Kkk3Cz0kM7UdiWeNNRNbRHeuIRp4ZXq6z2ZciUYlGngcZRlGK0B1rGY/wkxLq9zUSDcVwuEffhUgp6Wjqou1kF5AQNr6ckVV6pTVF5JXl4O8LUrfrBP2dA7i8qSvExiKrMIPK2aUc3nqcwd4AR7YdZ86q5F90KSVNh1voaBya25Q88stTi/+WujYGuv24M5LndMWicQ6+nVgGsDttVMy68HD2WMxeMQ3doRMORNj50j5uvPdqbOcwvL0QppXkUZzt44Gn3mTB1BI6+wPUtXajvweW5QWwoqicck8GTYGReWotgUEeO35wVP4TwMzsPK7ILZz0+b18qp5v7XyTQCyKIRNdGkb/WViMfMwcI1xXP9DLycE+FCFQEIn/T/KnDlUSq0LgVDXm5hXx/StveF+Kp1yHiy/MX8mhnk56k+S7nSZqGuztbmdfdzu6qqEpyrAxqiUlhmURs8yUrVqK3V6+tnQdfdEIj9cfIniRKuwGY1EGe7s43NuFIJG/l5i7Mpy3JRm6KZcSwzKTuugnoyojmy/MX0mxJ3kF9flyye0InHYbf3P31Xz85qWT0NpkbD60fgF/94lryRljieO9jhBOVK0SSFSVJaIuqX+wpAxjmicSr1WyUdR3fniFUFDVMhI/5VEso2mMPcvh3Krk+wkNJZ9LFLUIl+ezqGpR0uU/y+wa+yCH5yfQbHPRbPMASSz6KpbZiWmcwIjtASQ2+woUtXzcX47mY23se/Nw0jJ+I2bw1tM76O8cQFEVpi2sIuss5/DiqYXMWTUDRRE0HmpJmFeOmWuUYFS+FKCqKtfceSWqphAJRHjhv14n0J/c8DAejbPpyW20N3QhFMGSjfPx5aRuIBwNx3jpt28ON8E+Eykle149OOzdlFuSzbSFk+vOPWtZDZWzE+Js7+uH2fLcrkSO1xhIKceMmo2F26HzlT+7mjVzqjBMi0U1ZXzpg2u5bfkcBFCRn8n1i2biOGMpdnZ5AesXTI6X25kIISj1ZLC0qGzUc+bQ0sbZaEJhY+U0XLbJMeg9k95IiKN9XbQEB+kIBegKB+mJhOiLhhmIRfAPRZ/ChkHUTFzMxhJNpznzIh4xDUJGIlI1GIvSH43QGwnTFQ7SGQrQFvRzYrCPuv6ey6oVyEQihGB+fjFfXLiabMe5o9qShIgKDr3484OIAAAaWElEQVSng7EogXiMiGmkFE1Fbi/3L17Lgvxi8pxuci+RN5gE4pZF2DCGz4nBWBT/0DGEjfj4RZMvi68uWcvqkikXXE13mksWcWru6Odnj72FPxTBtCQLZpTy0Y0Lh/vEAQTCUb73m9e4enENToeNP24+TM9giIJsL9evnMWsqoLhnChLSk409/CHzYc41d5HYY6Pa5dNT1gKnPF7HDdMvvvgK7T3+LEsSXlRFnfdsJj8bO+I+ZmWxY5Dp9iyv4Hmjn4EUJKfwYaVs6gpz5uwD+AdNE5HSRKVZeP9ebCh2WYjhA8pBzFiu5GOa4HRAkVKSTy2NdE+BYa8kUbe5Wu2ucDvAZNY9HVs9qtS7DeKEd+VclbS8iNlYq1ZUfMRKfOhTGKxLec6yGEUxTtkTfASpnESI74fiYFhnkAIDzb7SoTwnnugIfq7BnnoW0+RVZBB5eyyYQsKI2byysObefl3mwHIzPOx/KaFwxVwp7HpGjd+6hrefnYXnU3dPP3TF7EsyQ33rCO7KHN4+UnKhFt2X+cAe149QGt9Bx/76u2jcqYWr59L7coZiXYqT24nuzCLO/7qBhxuB0IkxjHjBi88+AbP/eJl4lGD4qkFrPngMuxjuNwLAX/61WuU1BSx4qaFw87blmlxbOdJfvfPT2DETYQQXP2RVbgzxvHDLCUSOZwHJIf+ZzxWAXaXndv/9/V879O/IDQY5j/uf4ietn6u/sjK4WM9PZZpWrSd7GT3KwfQNJWb77vunHNLRlG2j4+tXTjisSsqE/YNM0rzmVE6cqlz6bRylk4rf1f7Ol88Np3lReU833BsXB46WQ5nSvPMNP9zUYXCbdWzkEi+t2tT0iq7d4MgIZr+77KrWVtahaao5Dpc5DhcNPov3CPqNKejiBcDm6KwpLCMz81dzpLC0gmtPL1kwsnncXDlgmpaugZ47OU9ZHgcWGclrxqGxY6DTdSf6samKZTkZZDhdrD7SDNv7T3JVz55HQtnliGE4FhjJ1/+8bPYbSqzq4vo94f49q9fxjAt8rLeiSgpimDl3Craugf5w+ZDHKpvT2py19o5wL89ugmXXSd/aHnm7f0NvLaznm/cdz1zpib3y3m3KEouAlsi4S2+HymDCHHuu8lEqf4qNH0B8ehrREKPYrOvQLevhTNM+BMXoBOEAz8DGUIoWThcH+bsoKPdcS1B/3eQVgeR8JPYnTeh2eaN8HKSMkIk9BCmkdpdWigehLAjZQzL7EbKgVHiScow0fDTCQ+n88BuX09I+yGmUUck/ERi+U+GUbRZ6PbR/d9SzlEI5l01m8Nb6/jShn9i3lWzKK4uxDRMju6op27XSWKROHaXzs33XcfsFcmb/JZNK+Yvf/QJ/u2L/0VrfQePfPcZnv35S0yZU05ucRYICPSH6GzsoqOxGyNuUDWnnDvvv23UfPLLcvj43/8Z//bF/+LYzhM88t1n2PTkNmpXTiczL4NAf4Aj2xONgo2YQXZhJh/7yu2JHKExjnvGkho6mrr4/l/8nGd+OoUZS6rRHTotx9vY8+ohAv1BVE1lyYa5XHdX8vcw0B+ktb6DWCSOMZQ039XcQ19H4oc1MrTs1t3Wh91hQ7OpaLpGdmFmwhj0rBymJRvmc9dXP8hD33qKtpOd/OxvfsPD336aKbVl+HK9mIaFv8c/5I3Vj7Qs1t991bg+28sNIQSri6dQ5sngcN+5I7A3TJlOlv1ysUxJM5G4bDofmT6XmVl5fH/3ZvZ2tadMiB4PWXYna0qn8BdzljA9652gQJ7LTb5rbAPh82VZYRmfrF3M8w1HafL3J203dKG4bTrVGdncVj2LO2quwG2zTcjy3JlcOuHkdnDtsukEQlHe3FWfcjvDsugZCPCFj65l7aLExWHXkVP8nweeYtvBJuZNL0VR4KE/7UIRgr+6cx0LZ5ZhScnrO47z1X97boRwUhWF1QsSyxD1zd3UnUr+I1VakMnf3H01pQWZeF12pJRsP3iKr/30D2w70DjhwknVylG0MqxYB0Z8PyH/D7A7b0RRigALKQeHWqDUjMxLAhQ1F5f7Pgbje5FWL8GBr2O667Dpy1CUPKQMYxgHiIQeGhIpOg7nB9FstaNOKKFk43R/lJD/B0irm8DAV3G47sRmX4LAgWV1EIu8QiT8CIrixbKSn/hCuIfE3OtYZiuhwI9xuD6MqpYPiakTRMPPEQk9ihCu4ejUuBBOHM7bCfq/STy6FUVN5Pbo9lWo6vg9OoQiuPbO1cxaVsOTP/lTUg8mX46HGz55NTffd11KQ1ShCOavq+Xz/3oPj//weba/sJfgQIgDm44k3d7u1KmYVZo0GVoIwaxlNXzme3fzm398nL2vH6T5WBvNx0ZWVSlKYunw9r/cyOoPLjunWeu0RVXc8Ml1PPLdZzi8rY7DW+vOGk+w6pbFfPTLt5KRmzxid2Tbcf71C79msCdANBwjHouPCIwOdPv55VcfBhItXnSHjt2pc83HVvPxr9+BdkaiuxACu1Pn+nvWkZHn5bEH/sCJ/Y30tvfT2578DteT6aJs+sR+795L5LvcrCyuOKdw8tp01pZWpRv6vo9RhcLC/BIeWHMjm1oa+FNjHds6ms+ZIH0mBS4PK4oquLZ8KmtKKvGcVW3m0GyUeHwoQqRc2jtfcp1uPnPFUm6umsHernZ2d7Wyt6uN4wM99EfHrnAeC0UIyr2ZzM8rYnVxJYsLSyn1XLjRZSoueXL4eJgztZhFs8qHLw5VJTmU5mfS1jWAZVn0+yMcPtHBjMoCrphWjKIkEg6XzqmgrDC1a/W5mFaRTygSo2cgiGlK8rLdKEIwEAhjSTmhH4oQDlzuz+CPf2Go7cl/Egk/8U7USRoIJRdv5ndQ9NHLXjb7Mjy+fyDo/zam2URw8NtDlWs2kBZSBpEyMCQ6bsXl/RyKkiy3S8HhuhPTaCQafhYjvpfA4PFE2b9QkDKGlCE0rQan+x78/V9IcUQ2XO578RvHscxWwoFfEg09AcJBYg0mjCX9qGo5bu/nCQ7+47hznUCgO9YRDv0Oy2zCNPoBOw7XBzmvCsOhpaY/++KNLLz2Cl59+C3qdp1gsCeAN9vDzCVTuerPljOltgz9HM2eNZvKFatnMqW2jON7Gtj10n6O7KhnoHMQwzBxZ7goqMhl+qJq5l45i6Kq/DG9mmYsqeZvfnkfR7fXs/npHZzc30SgP4jL56RkaiGL189l/tpaMvN9CS+ncxAcCLHylsXMXFbDlud2s+353fS29WFz6FTMKmHFTQuZv7YWT5Z7zLszVVNxeR24vOMvqBjLyNPu0llzx3LmrJrB0e31bH9hLw0Hm/H3BhCKwJvloaSmkOq5lVTPraBmfuW493s58pfzVvDRGWMbbmqKQoHTM+F30anYWDl9QrxvJgK7ql3QkovXZufry67mrxeuHvWcAHx66vPaH/fz7yf/nY2FG5nhm5jqrAtBCEGe080t1bNYV1ZNXzTM4d6EdUXjYB+tQT/+WJSYaWJTVdyaTpHby9TMbK7ILaI2p4AsuwO3TU96Lgngc3OX87EZ85BAXySMrqq4bTpZ48ixSsVpkVPmyeDa8qkE4lFCRpxT/gFODvbS5B+gPeinN5LIoQvF48QtE1NaKEKgqxo+m50cp4tyTwZTMrKpzS2g0OXFO1SJOdnfjctCOOVkuPGcceFSFIFd1zCGEkX7BsPEDJO8LDf6GRcjTVUoyPYy/nyhd+j3h3lu0yG2HmgkEIoixJDjcTCSyOlI1IBf6KGNQHesw5v5L0RCD2PED2FZfUgMEHYU4UNRCxEi1QmrYnfejKpNIRz6DUZsJ5bZnqigEzqKko1NW4ruuA6H63bATip3cUXJx+P7e1RtKrHI85jGKSyrG4SOqhZjs9+M0/1xkBaKkj9cpXc2NvtqvBn/Qjj4a4z4vqHtLIRwoajFOPSNON0fR1GyiGjTzitJXFUr0e1riIR+A1jo9pWoauW4Xn8aCcSjxlA5/FRmLp2acn/jmpci8GZ7mL+ulvnras95DOd63pfjYfGGuSzeMPddj3MaI2qgqgolUwu5/fMbuf3zG897vIXXXsEv9nxrXPs7H1RVIbckm9ySbFbesjjpNq8+sYOnH9zM3yxJ/hn9TyHD7iDD/t6p8hVC4NXteCfB5flSoCoKBe9y+cnCojvaTTRFlD0ZcStOd7SbHHsOujI5ifxCCDLsDny6nQpvJhsqxm5sfvZrxyLL4STL4cSUFg9sf4sFhcV8vHbBhU55eN8OTcOhJYyAyzwZrCh69zmFF7PS8rIQTqoixtU3bqKqLeKGySMv7uHJV/dxy9o5LK2twOO0EzctvvDdJyZoL6MRwobuuAFNX4xlnMKSg4CJQAfhQlHzUNWSFK8VgIpNn4eq1WCZjVhWD1JGEGgIJQNVLUMoOWP2bzs9llCzcHnuw+68HsvsQMoQAhuKmouiVqEoLqQVwpv1faQMo9nO7i8nAA3dcRWabTam0YAl/SSEkwNFyUVRKxLjyDhu7//Bct+Fps0a55vlQlVLSeRxCXTHdUPRrPNkKAQ9UV+6ifzyTuRYpx3ZL2TMyfxhGmtsKSVzlk2leEreZeXqn+b9jZSSwfggL3a+yI1FN5Ktp27UOxFM5vdTQfDJuYsmTdRfbvYS73nhNJ63MyfDjV3X6OjxE40ZOIYqnwzToqPXT0H2+d1hdPUF2HGoiYribD7xgWVoasJLpLGtd1JaPZxJwhKgEFV99/4siuJGUcYpQMaciw1NmwpaikiM4kK3jw55n7UVipqPoqY2ZxTChk1PHVVJhrR6iMd3AgaabXaiR9+ld9dIMwkIIcgrziKv+N0vu6dJc75IKemL99EX60MRCnbFjnLGTWfMitEb6yVoBJFI3KqbXHsuNsVGzIrRHe3miP8I9YF66gP1dNu6cakuipxFqEIlbIbpjfUSMRO5PV7NS7aejaa89y7LQgiuyJt8z7DLhffeJ3QW45EpmT4ntdWF7DzczO6jzSytrcSyLN7cc4Lmjv7zFk6qqqAqCqZpEY0ZaE6dgUCYp18/QCB8cczA0qRGSolhHCce3Q4o2PTFCS+ry+yu5f2MlJLm+g6ajnUQi8Xx94dYvHYW+7ccJxaJs+qGeWTkeGhr7Ob1p3cRDcXILc5iw4eXj2hpI6Wks7mXXW8epad9AN2uUV5TyMKrZo7pup4mzVhIJE3hJp5ofoKgGcSredEVnbj1TpPvo/6jvNjxIgoKJiZBI8j6gvUsyl5EwAiwq28XBwYP0Bfr4+2et9EVnTJXGbn2XFRVZWffTt7ueRubsBGTMUzL5JaSW5jhnTHcuPtQTxc721v44PRaXLZEQCAYj/HHk3WUen0sKSylLehnS+spWvyDIASzc/JZWlyKe8jjy5KSE/29vN16iu5wEJdmY2ZOPqtKK1CEwLQsjvV1s62tmd5IGLdNZ25eIUuKEnltbzQ3sLWtGUUIVhaXs7zkneW0vZ3tNAz0MTUrm80tTcQti9rcfFaVVAz3y+uNhHmzuYGT/X347HZWl1YwNTPnsosynckl+WWxpOTNXfWcaOmh3x+mrXuAgWCEnz/+FhkeJ9Mr81laWwGML+KkCMFHr1/EoRMdfOvXL1NVmks0ahCKxJg/Y2RS44Hjbew62kwgFGX/8Vb6/GF+89x2CnN9lBZksmpeFTkZbpbWlvPIi3v42x89Q362h9auQbJ9LkryMybhHUlzfkSIhB5Eyj6EyMTuuHVCGvqmubi0nuzml//8NIvWzuTIzgY2P7eH8ppC6g+2YBoWN/35atxeJ+U1hWx+fi97Nh/j2juWDAsnKSWNR9v44ZcexuHSmVpbRn+3n6A/wtyV09LCKc27JmbGeKPrDUxpcm/VvehCZ3f/bnb37x7eptxVzofLPoxTTeSdvtz5Mtt6tzHLN4sMWwZXF1xNlp7Fa12vcVvJbWTr2cORK4CZ3pnUeGpwqA5iVownWp5gZ99OpnqmYhtqZxU24jx4aA9LikqZkZOXOOcH+vnF3h18Y9XVALzYUM/mlkaqM7Pxx2I8WXeIz85fygdqZqIg2Nnewtc3v0Kh28P07FzagwEMKVlRUo4gIYy+ufUNKn2ZVGfmUOfvwaaow8Kp0O2lOjObn+zeil1RRwinQz2d/HjXFmbn5lPmzaA/GuHxYwe5f/lVrCuvYjAa5f+9/RpdoSC1ufnsau/h6bojfHPNdczIybsYH+WkcGl+WSQ0tfcllr6AxbMTIql3METvYAi3U2fJ7Ap0m8raRTXMnDIyRGjTVJbNqcDrdgznPlUWZfMvn7+JV7bV0dzZT2GOl2uWTqehtZemtr5hV/LOXj8nW3qQUlJTnlg+ihkmTe19mKbF0tpKXA6Fj924mLLCbPYca8YwLDYsn8GVC6eyaXd9YqzLWC1fTpxuySKEDtLEsjqJhp8lGn4OENidH0DTJyZZMc3FJzPHw62fXMvOqYd58ZGt3PnF63nxka0cP3AKy5L4st2s2HAFAz0B2hq6R7w2FjV4/Oev4sty89c/uPNd9whMk+ZswmaYE8ETrM1bS56el/B8y5zHM63PDG/jUl0MxAfoCHUQNaPEZRy/4SdmxfAKLw7FkVjeQ8GhOnBpI124PZqH5nAzreFWolZ0eGnQkomiJyEE07NzmZmTz1PHDyeEE7CjvYU8l5upWYmozYdnXsHHZs9DADHL5PvbN7O5pYnrq6ZjSosf7trCFXmFfGXFVbi1kQa+wXiMH+x8mzVlU/jCwhU4tNGSYFpWDtOycnj0yP6k75WUkg/PuII15VMwLJMvvf4Cf6g/yrryKl49dYKT/b18b+1GKjOyiJkm/+vlZ/nVgV384+prh6NSlxuXRDgpiuBjNySvnjkTl0Pnr+9al/Txe25ZPuIxIQRlBVncfdOSEY9XFI1MyFu3ZBrrlpy7jYKmqlyzdBrXLB257c1r5qR4RZrJIU5g4H6kjADmUIuVk0AczTYPp+cTnG+T4zTvHTyZLuxOHXeGk+wCH5quYnfYMOMWyLGTRgd6ArQ1drP6hnl4UvTgS5Pm3WBhETWjuNR3zitFKNjVRLQoZiUiUrv7d1PmLMOluQjEA+MeP2SE+GP7HzkZPEmZqwyH6iBkhkblN3l0O6tKynnoyH76I2FcNhsvNtazqrQCn57wFwzGY7zQcJz6vh4GohEO9XRR7PFiSUl7MEB3OMiHZszBk6Q9zyn/IIFYlBXF5Tht765XZIHbw7Ts3GGrgHyXmxMDfQDsbG+hPRjguzs2ow3lh9X39RDzZRA24qO8oy4X0rHsNO9xLIz4YSyzGbAAFSG8aPpKPL4voapV571Wnl2URfXcChRFwZcz/vYslxsun5MptWX4ewMUVuQlNdy81CiKglASAVxVUxOf5TinKS0LaclJbQ6c5v2JJjR8Nh9d0S4kEoEgbIYJGAlx1Bvr5fWu17mh6AYWZC1AQeHVrlfpjHaOGEcRCkONiUY83hhqZEvvFj5d9WnKXeWY0uSx+GP0xkdauwjgmopqHjy4h00tjVT4MukJB1lRXIYqBK0BP1/d9BJOzcaGqhpyHS509RgdocQ8TctCStDV5NWopmUBAn0cXnCpsKnqCNGlDOVnAcRMM2HsWlKBNhRdWllaQYHLja5evvLj8p15mvcJtoRVgdmSsFZQ3KhqJTb7ShTF965GvOlT13DTp66Z4Hm+95i1bBr/8vyXL/U0Jg1ftpucogz2bznOutsWjdmrL02a88GpOpnlm8WW3i3kO/Jxq2529O0YFgSqUNEVnc5oJ23hNnpjvezp3zNqnCw9i4H4AAcHDlLuKkdXdIocReiKjipUWsOtaEKjNdLKscAxcu25o8dwOFleXMZbLafY09nG9OxcpmXnIoSgvr+XI71d/PaGO5iSmU3YiPPI0QPDry1we8iw29ne3sKVZZXYzxIrJV4fDk1lV0cbiwtLsaUQWGMhSH2vMze/iKO93SwrLqMyIwvBUPsvKYeF1OVIWjileU8jhIrDdeulnkaaS4CUks6WPrrb+mmqaycwGOLAtnoysj2U1xTicNnZ8OEV/Ov9j/Kv9z/KjAWVREIxIqEot9xz1Xm5m6dJcyaaonFl7pVEzAjPtD6DW3VTm1nLNO80FKGQZctifeF6Xut8jQMDB8iz57EyZyX7BvaNiIAXOYpYmbuSzT2b2dS9iTkZc1hfuJ4yVxnXFlzLG91voKBQ5ipjVe4qWsOtSedzfdV0vrLpRfyxGF9etmZYAOW73OQ4Xfz28D7m5RdysLuTur5uSr2JIiafbufOWfP4zvZNhOJxanPz6Y9GkFJy79zFZNodfHjGXH6xbzs9kRDTs3LpjYSxqxqfmLMAQ1oc7+tlIBphIBalcbCf7W3N5DhdVGWe25fq6opqXmio4x/eepU1ZVNQheDEQC9Li8pYP2X8Rp3vNYQc25dock2L0qRJ875ESkndvlPs2XSUG+5aRcORVg7vbOCmu1ez7+3jNNd3suEjy3nh4S0c3FafaAAuE+7siqpwz5dvJr80GyklJw628MYzu+hs6cPlsTNrcTWrb5yHnl7CS3MBSClHLLMpKMPLdqftAiyGErmH/jnz+WRjnN5uvK8/PYY/FuXne7fTHgzwD6uvwTmU5C2lZGdHK0/WHSZsxFlYUExNdg6Hurv40Mw52FUN07LY1dHKCw3H6QgFyNAdrCyp4LopU1GEIG6abG1r5pWmerrDITLtTtaVV7GmrJK+SJh/2vI6ISOOJWUiuiQEJR4f9y+/ijebG9jW1syn5y7Boycivk8cO0Rb0M9n5i9FSkl3OMQfT9axp7MVIRSqM7O4uXomJd53t2JwEUmZNJAWTmnSpEmTJk2aNCNJKZwu30XGNGnSpEmTJk2ai0xaOKVJkyZNmjRp0oyTtHBKkyZNmjRp0qQZJ+eqqnvvGb+kSZMmTZo0adJcItIRpzRp0qRJkyZNmnGSFk5p0qRJkyZNmjTjJC2c0qRJkyZNmjRpxklaOKVJkyZNmjRp0oyTtHBKkyZNmjRp0qQZJ2nhlCZNmjRp0qRJM07+f5DmYaxCPaNgAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 576x216 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "P3WG3IstB2Zp"
      },
      "source": [
        "# 5 Words that carry significant weights in the model\n",
        "(DID NOT included in the report. Just experiment)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nD2pLms2FuvO"
      },
      "source": [
        "PATH1 = \"/content/drive/Shareddrives/CS682 Project/'lstm2.pt'\"\n",
        "the_final_model = torch.load(PATH1,map_location='cpu')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "YQzn4Ee1rr10",
        "outputId": "beee807e-4b87-4e99-8ad6-54ef3febf084"
      },
      "source": [
        "weights = torch.mean(the_final_model.bert.embeddings.word_embeddings.weight,axis = 1)\n",
        "weights"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "tensor([-0.0220, -0.0221, -0.0222,  ..., -0.0162, -0.0176, -0.0316])"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 8
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zA_04TVxCdRu"
      },
      "source": [
        "sorted, indices = torch.sort(weights,descending = True)\n",
        "target_list = indices[0:200].tolist()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ovDsQgSh72yu",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "8459843e-2314-4030-aeda-11cb55423fd3"
      },
      "source": [
        "tokens = tokenizer.convert_ids_to_tokens(target_list)\n",
        "print(tokens)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "['[CLS]', '[MASK]', '[SEP]', \"'\", '\"', '(', '.', 'to', ';', 'by', 'of', '?', 'during', 'in', 'from', 'for', ')', 'on', 'after', 'and', 'at', 'or', 'through', ':', 'when', 'under', 'but', 'as', ',', 'between', 'toward', 'while', 'if', 'that', 'until', 'could', 'what', 'into', 'with', 'are', 'over', 'were', '-', '!', 'about', 'lumpur', 'against', 'within', 'although', 'might', 'my', 'around', 'since', 'would', '...', 'his', 'because', '/', 'which', 'a', 'her', '&', 'upon', 'will', 'is', 'off', 'where', '##elial', 'like', 'was', 'before', 'who', 'kylie', 'though', 'its', '*', 'each', 'than', 'should', 'out', 'down', 'how', 'this', 'along', 'across', 'onto', 'just', 'any', 'all', 'now', 'one', 'why', 'towards', 'up', 'both', 'the', 'must', 'you', 'even', 'every', 'near', 'season', 'following', 'became', 'their', 'can', '%', 'throughout', 'then', 'not', 'day', 'inside', 'only', 'so', 'an', 'your', 'never', 'more', 'another', 'nedra', 'time', 'beside', '##peration', 'including', 'me', 'behind', 'we', 'most', 'our', 'some', 'away', 'did', 'year', 'moment', 'once', 'without', 'been', 'album', 'there', 'too', '##rricular', 'located', 'had', 'here', 'want', 'much', 'it', 'way', 'among', 'yet', '=', '་', '$', 'league', 'first', 'part', 'whether', 'also', 'aback', 'back', '##urbed', 'two', 'years', 'school', 'seemed', 's', 'right', '[', 'ever', 'maybe', 'later', 'south', 'i', 'well', 'being', 'north', 'didn', 'despite', 'ethan', 'last', 'side', 'club', 'have', 'them', 'felt', 'best', 'still', 'next', 'home', 'yes', 'outside', 'has', 'week', 'house', 'him', 'again', 'doesn', 'together', 'approximately', 'less']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GahVWINY7wFl"
      },
      "source": [
        "target_list = indices[0:200].tolist()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HIW7h0MM75KO",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "3de2e0d5-db12-4bce-bbf6-5e7ca37db1be"
      },
      "source": [
        "from wordcloud import WordCloud, STOPWORDS\n",
        "L = []\n",
        "stopwords = set(STOPWORDS) \n",
        "string = set(string.punctuation)\n",
        "u = stopwords.union(string)\n",
        "for i in tokens:\n",
        "  if i not in u:\n",
        "    L.append(i)\n",
        "L"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "['[CLS]',\n",
              " '[MASK]',\n",
              " '[SEP]',\n",
              " 'toward',\n",
              " 'lumpur',\n",
              " 'within',\n",
              " 'although',\n",
              " 'might',\n",
              " 'around',\n",
              " '...',\n",
              " 'upon',\n",
              " 'will',\n",
              " '##elial',\n",
              " 'kylie',\n",
              " 'though',\n",
              " 'along',\n",
              " 'across',\n",
              " 'onto',\n",
              " 'now',\n",
              " 'one',\n",
              " 'towards',\n",
              " 'must',\n",
              " 'even',\n",
              " 'every',\n",
              " 'near',\n",
              " 'season',\n",
              " 'following',\n",
              " 'became',\n",
              " 'throughout',\n",
              " 'day',\n",
              " 'inside',\n",
              " 'never',\n",
              " 'another',\n",
              " 'nedra',\n",
              " 'time',\n",
              " 'beside',\n",
              " '##peration',\n",
              " 'including',\n",
              " 'behind',\n",
              " 'away',\n",
              " 'year',\n",
              " 'moment',\n",
              " 'without',\n",
              " 'album',\n",
              " '##rricular',\n",
              " 'located',\n",
              " 'want',\n",
              " 'much',\n",
              " 'way',\n",
              " 'among',\n",
              " 'yet',\n",
              " '་',\n",
              " 'league',\n",
              " 'first',\n",
              " 'part',\n",
              " 'whether',\n",
              " 'aback',\n",
              " 'back',\n",
              " '##urbed',\n",
              " 'two',\n",
              " 'years',\n",
              " 'school',\n",
              " 'seemed',\n",
              " 's',\n",
              " 'right',\n",
              " 'maybe',\n",
              " 'later',\n",
              " 'south',\n",
              " 'well',\n",
              " 'north',\n",
              " 'didn',\n",
              " 'despite',\n",
              " 'ethan',\n",
              " 'last',\n",
              " 'side',\n",
              " 'club',\n",
              " 'felt',\n",
              " 'best',\n",
              " 'still',\n",
              " 'next',\n",
              " 'home',\n",
              " 'yes',\n",
              " 'outside',\n",
              " 'week',\n",
              " 'house',\n",
              " 'doesn',\n",
              " 'together',\n",
              " 'approximately',\n",
              " 'less']"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 37
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "01LHDnvsrySt"
      },
      "source": [
        "# 6. Top frequent words and model performance"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "S-YduW70b_i2"
      },
      "source": [
        "class BertLstmClassifier(nn.Module):\n",
        "    def __init__(self, model_tune):\n",
        "        super().__init__()\n",
        "        self.bert = model_tune.bert\n",
        "        self.lstm = nn.LSTM(input_size = 768, \n",
        "                            hidden_size = 768, \n",
        "                            num_layers = 2, \n",
        "                            batch_first = True, \n",
        "                            bidirectional = True)\n",
        "        self.classifier = nn.Linear(768 * 2, 2)\n",
        "        self.softmax = nn.Softmax(dim = 1)\n",
        "\n",
        "    def forward(self, input_ids, attention_mask, token_type_ids):\n",
        "        bert_output = self.bert(input_ids = input_ids, attention_mask = attention_mask, token_type_ids = token_type_ids)\n",
        "        out, _ = self.lstm(bert_output[0])\n",
        "        logits = self.classifier(out[:, 1, :])\n",
        "        return self.softmax(logits)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "I0XYzA98eFFt"
      },
      "source": [
        "PATH1 = \"/content/drive/Shareddrives/CS682 Project/'lstm2.pt'\"\n",
        "the_final_model = torch.load(PATH1,map_location='cpu')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sYkBJNVrI3MT"
      },
      "source": [
        "dic = [(1996, 8249),\n",
        " (101, 7440),\n",
        " (102, 7440),\n",
        " (1997, 5349),\n",
        " (1999, 4686),\n",
        " (2000, 4605),\n",
        " (1037, 3765),\n",
        " (1998, 3358),\n",
        " (2522, 2539),\n",
        " (17258, 2406),\n",
        " (2003, 2378),\n",
        " (2005, 2195),\n",
        " (16147, 2080),\n",
        " (3572, 2016),\n",
        " (2024, 1975),\n",
        " (2008, 1649),\n",
        " (21887, 1480),\n",
        " (2013, 1431),\n",
        " (2057, 1407),\n",
        " (2015, 1348),\n",
        " (2006, 1338),\n",
        " (2047, 1275),\n",
        " (23350, 1271),\n",
        " (2007, 1205),\n",
        " (2031, 1204),\n",
        " (2004, 1117),\n",
        " (2023, 1101),\n",
        " (1521, 941),\n",
        " (2012, 924),\n",
        " (2038, 923),\n",
        " (2011, 919),\n",
        " (6677, 883),\n",
        " (2009, 868),\n",
        " (5852, 832),\n",
        " (100, 813),\n",
        " (2256, 806),\n",
        " (2045, 805),\n",
        " (2111, 794),\n",
        " (2025, 769),\n",
        " (2022, 727),\n",
        " (2062, 724),\n",
        " (2651, 706),\n",
        " (2193, 677),\n",
        " (4484, 664),\n",
        " (2163, 648),\n",
        " (2064, 640),\n",
        " (2042, 639),\n",
        " (2988, 630),\n",
        " (2102, 609),\n",
        " (2561, 581),\n",
        " (2019, 580),\n",
        " (2030, 576),\n",
        " (2001, 574),\n",
        " (2085, 553),\n",
        " (2035, 539),\n",
        " (1055, 534),\n",
        " (2040, 515),\n",
        " (2902, 507),\n",
        " (2740, 498),\n",
        " (2110, 493),\n",
        " (23713, 493),\n",
        " (1016, 485),\n",
        " (2017, 485),\n",
        " (2065, 484),\n",
        " (5604, 478),\n",
        " (2039, 476),\n",
        " (2020, 471),\n",
        " (1015, 462),\n",
        " (2097, 456),\n",
        " (2084, 454),\n",
        " (2634, 447),\n",
        " (2053, 430),\n",
        " (2021, 409),\n",
        " (2058, 400),\n",
        " (5022, 400),\n",
        " (2487, 387),\n",
        " (2027, 373),\n",
        " (2692, 372),\n",
        " (2115, 372),\n",
        " (2037, 371),\n",
        " (2951, 370),\n",
        " (2029, 362),\n",
        " (2575, 362),\n",
        " (3207, 361),\n",
        " (2055, 355),\n",
        " (2080, 354),\n",
        " (3189, 354),\n",
        " (6090, 353),\n",
        " (3231, 352),\n",
        " (4609, 347),\n",
        " (2683, 346),\n",
        " (7712, 340),\n",
        " (2420, 339),\n",
        " (1017, 334),\n",
        " (2553, 333),\n",
        " (2154, 332),\n",
        " (2549, 321),\n",
        " (17404, 321),\n",
        " (2050, 318),\n",
        " (2069, 318)]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SoPHiEE5q6A6"
      },
      "source": [
        "word_list =['cases', 'corona', 'news', 'deaths', 'tests','today','confirmed','reported','states','total']"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "RYC2rB1ut2S5",
        "outputId": "87c4aec8-bdc1-4729-b6ba-07c80988897b"
      },
      "source": [
        "ids_list = tokenizer.convert_tokens_to_ids(word_list)\n",
        "print(ids_list)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[3572, 21887, 2739, 6677, 5852, 2651, 4484, 2988, 2163, 2561]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WGhWqSfUI9J9"
      },
      "source": [
        "ids_list = []\n",
        "for i in range(len(dic)):\n",
        "  ids_list.append(dic[i][0])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZuBKTZumeP-z",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "997f2a97-609a-41c1-d3a4-21c85bdea94f"
      },
      "source": [
        "model = the_final_model.cuda()\n",
        "print(\"Validation...\")\n",
        "model.eval()\n",
        "total_eval_accuracy = 0\n",
        "total_eval_accuracy1 = 0\n",
        "total_num = 0\n",
        "tag = False\n",
        "for batch in validation_dataloader:\n",
        "    input_ids = batch[0].to(device)\n",
        "    input_mask = batch[1].to(device)\n",
        "    labels = batch[2].to(device)\n",
        "    input_list = []\n",
        "    label_list = []\n",
        "    mask_list = []\n",
        "\n",
        "    input_list_r = []\n",
        "\n",
        "    for i in range(input_ids.size(0)):\n",
        "      for ids in ids_list:\n",
        "        if ids in input_ids[i]:\n",
        "          tag = True\n",
        "      if tag == True:\n",
        "        input_list.append(input_ids[i])\n",
        "        label_list.append(labels[i])\n",
        "        mask_list.append(input_mask[i])\n",
        "        for ids in ids_list:\n",
        "          for j in range(len(input_ids[i])):\n",
        "            if input_ids[i][j] == ids:\n",
        "              input_ids[i][j] = 0\n",
        "        input_list_r.append(input_ids[i])\n",
        "      tag = False\n",
        "    \n",
        "    if len(input_list)==0:\n",
        "      continue\n",
        "\n",
        "    input_list = torch.stack((input_list)).to(device)\n",
        "    label_list = torch.stack((label_list)).to(device)\n",
        "    mask_list = torch.stack((mask_list)).to(device)\n",
        "    input_list_r = torch.stack((input_list_r)).to(device)\n",
        "\n",
        "    with torch.no_grad():        \n",
        "      out = model(input_ids = input_list, attention_mask = mask_list, token_type_ids = None)\n",
        "    pred = torch.argmax(out, dim = 1)\n",
        "    total_eval_accuracy += torch.sum(pred == label_list).item()\n",
        "\n",
        "    with torch.no_grad():        \n",
        "      out1 = model(input_ids = input_list_r, attention_mask = mask_list, token_type_ids = None)\n",
        "    pred1 = torch.argmax(out1, dim = 1)\n",
        "    total_eval_accuracy1 += torch.sum(pred1 == label_list).item()\n",
        "\n",
        "    total_num += len(input_list)\n",
        "        \n",
        "avg_val_accuracy = total_eval_accuracy / total_num\n",
        "avg_val_accuracy1 = total_eval_accuracy1 / total_num\n",
        "print(\"  Accuracy: {}\".format(avg_val_accuracy))\n",
        "print(\"  Removing Accuracy: {}\".format(avg_val_accuracy1))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Validation...\n",
            "  Accuracy: 0.967391304347826\n",
            "  Removing Accuracy: 0.967391304347826\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}