External_Testset_Inference.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "5fff2d53",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "from scipy.special import logit, expit\n",
    "import pandas as pd\n",
    "from matplotlib import pyplot as plt\n",
    "import matplotlib.cm as cm\n",
    "\n",
    "import seaborn as sns\n",
    "\n",
    "from PIL import Image\n",
    "import torch\n",
    "from pathlib import Path\n",
    "from torch.nn import functional as F\n",
    "from torch.utils.data import Dataset, DataLoader\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "from torchvision import transforms, models\n",
    "from skimage.transform import resize\n",
    "from sklearn.metrics import roc_auc_score, brier_score_loss, roc_curve, auc\n",
    "from tqdm.notebook import tqdm\n",
    "from datetime import datetime\n",
    "seed=42\n",
    "np.random.seed(seed)\n",
    "torch.manual_seed(seed)\n",
    "\n",
    "class TOPDataset(Dataset):\n",
    "  def __init__(self, csv_file, img_dir, targets, transform=None,\n",
    "               covariates=None, return_covariates='auto', \n",
    "               labelsmoothing_alpha=0, include_third_channel=False, filter_label=None):\n",
    "    self.csv_file = csv_file\n",
    "    self.img_dir = Path(img_dir)\n",
    "    self.transforms = transforms.ToTensor() if transform is None else transform\n",
    "    self.targets = targets if not isinstance(targets, str) else [targets]\n",
    "    self.covariates = covariates if not isinstance(covariates, str) else [covariates]\n",
    "    self.return_covariates = (covariates is not None) if return_covariates=='auto' else return_covariates\n",
    "    self.labelsmoothing_alpha = labelsmoothing_alpha\n",
    "    self.include_third_channel = include_third_channel\n",
    "    self.df = pd.read_csv(csv_file)\n",
    "    if filter_label is not None:\n",
    "      print(f'Caution: This dataloader will be filtered to only include images with this label: {filter_label}')\n",
    "      self.df = self.df[self.df[filter_label]==1]\n",
    "    \n",
    "  def __len__(self):\n",
    "    return self.df.filename.size\n",
    "  \n",
    "  def __getitem__(self, idx):\n",
    "    row = self.df.iloc[idx,:]\n",
    "    \n",
    "    image = Image.open(self.img_dir/row['filename'])\n",
    "    # all images are loaded as left eyes by default, i.e. right eyes are flipped\n",
    "    if row['LR'] == 'R':\n",
    "      image=image.transpose(Image.FLIP_LEFT_RIGHT)\n",
    "    image = self.transforms(image)\n",
    "    if not self.include_third_channel:\n",
    "      image=image[:2,...]\n",
    "      \n",
    "    targets = torch.Tensor(row[self.targets])\n",
    "    \n",
    "    if self.labelsmoothing_alpha > 0:\n",
    "      targets *= (1-(self.labelsmoothing_alpha*2))\n",
    "      targets += self.labelsmoothing_alpha\n",
    "    \n",
    "    if self.return_covariates:\n",
    "      if self.covariates is not None:\n",
    "        covariates = row[self.covariates].values[0]\n",
    "      else:\n",
    "        covariates = None\n",
    "      return image, targets, covariates\n",
    "    \n",
    "    return image, targets\n",
    "\n",
    "# TOP Train set, exact\n",
    "## Adding a third value for compatability with plotting images\n",
    "norm_means = [0.22578795, 0.23797078, 1]\n",
    "norm_stds = [0.14651306, 0.11282759, 1]\n",
    "resolution = (384, 512)\n",
    "\n",
    "basic_transform_with_resize = transforms.Compose([\n",
    "                                      transforms.ToTensor(),\n",
    "    transforms.Resize((384,512)),                                  \n",
    "    transforms.Normalize(norm_means, norm_stds)])\n",
    "denormalize = transforms.Compose([transforms.Normalize(-np.array(norm_means)/np.array(norm_stds),\n",
    "                                                       1/np.array(norm_stds))])\n",
    "denormalize_2ch = transforms.Compose([transforms.Normalize(-np.array(norm_means[:2])/np.array(norm_stds[:2]),\n",
    "                                                       1/np.array(norm_stds[:2]))])\n",
    "\n",
    "targets = ['MH', 'RP', 'AMD', 'RVO', 'RD', 'Gla', 'DR', 'any_retina_disease']\n",
    "\n",
    "\n",
    "model = torch.load('PATH/TO/MODEL/ema_model.pt')\n",
    "\n",
    "device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
    "model.to(device)\n",
    "print()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "3c582c56",
   "metadata": {},
   "outputs": [],
   "source": [
    "def tensor_to_img(img):\n",
    "  img = denormalize_2ch(img).numpy().transpose(1,2,0)\n",
    "  img = np.dstack([img, np.zeros(img.shape[:-1])])\n",
    "  return img\n",
    "\n",
    "def plot_2channel_img(img):\n",
    "  plt.figure(dpi=150)\n",
    "  plt.imshow(tensor_to_img(img))\n",
    "  \n",
    "def sigmoid(arr):\n",
    "  arr = torch.Tensor(arr)\n",
    "  arr = torch.sigmoid(arr)\n",
    "  arr = arr.numpy()\n",
    "  return arr\n",
    "\n",
    "def make_gradcam_heatmap(img, model, \n",
    "                         class_idx=None, block=7,\n",
    "                         interpolation_mode='nearest', norm='max1',\n",
    "                         device='cpu'):\n",
    "  data_transform = transforms.Compose([transforms.ToTensor(),\n",
    "                                       # Careful, this uses global variables\n",
    "                                       transforms.Normalize(norm_means, norm_stds)])\n",
    "  data_transform=basic_transform_with_resize\n",
    "  img = data_transform(img).unsqueeze(0)\n",
    "  img = img[:,:2,...].float()\n",
    "  \n",
    "  model_prepool = nn.Sequential(*list(model.children())[:block+1]).float().to(device)\n",
    "  model_head = nn.Sequential(*list(model.children())[block+1:]).float().to(device)\n",
    "  \n",
    "  prepool_out = model_prepool(img.to(device))\n",
    "  prediction = model_head[:-1](prepool_out).flatten().unsqueeze(0)\n",
    "  prediction = model_head[-1](prediction)\n",
    "\n",
    "  if class_idx is None:\n",
    "    # by convention, the last label is healthy-unhealthy\n",
    "    class_idx = torch.argmax(prediction[0,:-1])\n",
    "  class_channel = prediction[...,class_idx]\n",
    "\n",
    "  gradients = torch.autograd.grad(class_channel, prepool_out)[0]\n",
    "  pooled_gradients = torch.mean(gradients, axis=(2,3)).squeeze(0).detach()\n",
    "  prepool_out = prepool_out.squeeze(0).detach()\n",
    "  heatmap = torch.zeros(size=(prepool_out.size()[-2:])).to(device)\n",
    "\n",
    "  for idx, weight in enumerate(pooled_gradients):\n",
    "    heatmap += prepool_out[idx,...]*weight\n",
    "\n",
    "  heatmap = heatmap.clamp(0)\n",
    "  heatmap = heatmap - torch.min(heatmap)\n",
    "  \n",
    "  # this does nothing\n",
    "  if norm=='mass1':\n",
    "    # norm to mass 1\n",
    "    heatmap /= (heatmap.sum() + 1e-15)\n",
    "  elif norm=='max1':\n",
    "    # norm to max 1\n",
    "    heatmap /= (heatmap.max() + 1e-15)\n",
    "#   heatmap = heatmap / (torch.max(heatmap)+1e-12)\n",
    "\n",
    "  heatmap = F.interpolate(heatmap.unsqueeze(0).unsqueeze(0), size=img.size()[-2:],\n",
    "                         mode=interpolation_mode).squeeze(0).squeeze(0).clamp(0)\n",
    "\n",
    "  return heatmap.to('cpu').numpy(),torch.sigmoid(prediction).detach().to('cpu').numpy()[0]\n",
    "\n",
    "def plot_gradcam_image(img, model, class_idx=1, block=7, axes=None, heatmap_alpha=0.3,\n",
    "                       device=None, show=True, dpi=250):\n",
    "  if axes is None:\n",
    "    fig, axes = plt.subplots(1,3, dpi=dpi)\n",
    "    for ax in axes:\n",
    "      ax.axis('off')\n",
    "  \n",
    "  axes[0].imshow(np.array(img))\n",
    "  heatmap, preds = make_gradcam_heatmap(img=img, model=model,\n",
    "                                 class_idx=class_idx, block=block,\n",
    "                                interpolation_mode='nearest', device=device)\n",
    "  axes[1].imshow(heatmap.clip(0,1), cmap='jet')\n",
    "  # TODO targets are hardcoded right now\n",
    "  if class_idx is None:\n",
    "    class_idx = np.argmax(preds[:-1])\n",
    "#   print(preds, class_idx)\n",
    "  axes[1].text(15, 370, '$\\hat{p}$'+f'({targets[class_idx][:3]}) = {preds[class_idx]:.4f}',\n",
    "               {'fontsize':'xx-small','backgroundcolor':'white'}, \n",
    "               bbox={'alpha':0.8, 'facecolor':'white', 'edgecolor':'none'})\n",
    "  \n",
    "  jet = cm.get_cmap(\"jet\")\n",
    "  jet_colors = jet(np.arange(256))[:, :3]\n",
    "  heatmap = np.uint8(255 * heatmap)\n",
    "  jet_heatmap = jet_colors[heatmap]\n",
    "  # remove blue\n",
    "  jet_heatmap[:,:,2]*=0.1\n",
    "  heatmap_on_img = (jet_heatmap * heatmap_alpha +  (1-heatmap_alpha) * (resize(np.array(img), (384,512))/1))\n",
    "  axes[2].imshow(heatmap_on_img)\n",
    "  plt.subplots_adjust(wspace=0.01)\n",
    "  if show:\n",
    "    plt.show()\n",
    "\n",
    "def predict_on_img(model, device=None, image=None, img_path=None,\n",
    "                   transform=basic_transform_with_resize, image_LR='L',\n",
    "                   plot_img=False, print_preds=False, plot_grad_cam=False):\n",
    "  if image is None:\n",
    "    assert img_path is not None\n",
    "    image = Image.open(img_path)\n",
    "  # all images are loaded as left eyes by default, i.e. right eyes are flipped\n",
    "  if image_LR == 'R':\n",
    "    image=image.transpose(Image.FLIP_LEFT_RIGHT)\n",
    "  clean_image = image.copy()\n",
    "\n",
    "  image = basic_transform_with_resize(image)\n",
    "  image=image[:2,...]\n",
    "  \n",
    "  if device is None:\n",
    "    device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
    "  model.to(device)\n",
    "  model.eval()\n",
    "  image=image.to(device)\n",
    "\n",
    "  image=image.unsqueeze(0)\n",
    "  \n",
    "  with torch.no_grad():\n",
    "    preds = torch.sigmoid(model(image)).cpu().numpy()[0]\n",
    "  \n",
    "\n",
    "  if plot_img:\n",
    "    plt.imshow(clean_image)\n",
    "    plt.show()\n",
    "\n",
    "  \n",
    "  if plot_grad_cam:\n",
    "    plot_gradcam_image(clean_image, model, class_idx=-1, block=7, device=device)\n",
    "    plt.show()\n",
    "  \n",
    "  if print_preds:\n",
    "    pred_label_str = \"\"\n",
    "    for i in range(len(targets)):\n",
    "      pred_label_str+=f'{targets[i][:3]}:'\n",
    "      pred_label_str+=f'{preds[i]:.3f} '\n",
    "    print(pred_label_str)\n",
    "  \n",
    "  return preds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "f85a2c85",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "46cccacf73524cb0b57ebc34e7ea8549",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/75 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1 /75: ASRS-RIB-Image-27209.jpg-Target label: RP, p(label): 0.2770\n",
      "2 /75: ASRS-RIB-Image-28324.jpg-Target label: RP, p(label): 0.6970\n",
      "3 /75: ASRS-RIB-Image-18045.jpg-Target label: RP, p(label): 0.8910\n",
      "4 /75: ASRS-RIB-Image-25606.jpg-Target label: RVO, p(label): 0.6643\n",
      "5 /75: ASRS-RIB-Image-66298.jpg-Target label: RVO, p(label): 0.9267\n",
      "6 /75: ASRS-RIB-Image-26484.jpg-Target label: RD, p(label): 0.7363\n",
      "7 /75: ASRS-RIB-Image-30011.jpg-Target label: RD, p(label): 0.7240\n",
      "8 /75: Normal HOZ-20190131 (1).jpg-Target label: Healthy, p(label): 0.4211\n",
      "9 /75: RP DVY-20190416 (1) From SCO.jpg-Target label: RP, p(label): 0.9441\n",
      "10/75: RP DVY-20190416 (2).jpg-Target label: RP, p(label): 0.9431\n",
      "11/75: RP GSL1-20190731 (1) 20-30 OU With OCT and Optos with FAF _ LEFTEYE.jpg-Target label: RP, p(label): 0.9192\n",
      "12/75: RP GSL1-20190731 (1) 20-30 OU With OCT and Optos with FAF _ RIGHTEYE.jpg-Target label: RP, p(label): 0.9417\n",
      "13/75: RP LYV2-20181121 (5) _ LEFTEYE.jpg-Target label: RP, p(label): 0.9286\n",
      "14/75: RP LYV2-20181121 (5) _ RIGHTEYE.jpg-Target label: RP, p(label): 0.8256\n",
      "15/75: CRVO Ischemic PVM-20190709 (1) HM OD 20-50 OS Fresh CRVO OD Old resolved CRVO OS.jpg-Target label: RVO, p(label): 0.8363\n",
      "16/75: BRVO Major SZB-20191203.jpg-Target label: RVO, p(label): 0.8892\n",
      "17/75: BRVO Major URI-20200106 Extramacular.jpg-Target label: RVO, p(label): 0.7582\n",
      "18/75: Unknown DSV-20190618 (1) 74YOF Possible multifocal BRVO's.jpg-Target label: RVO, p(label): 0.6865\n",
      "19/75: RRD GSV-20190213 (1).jpg-Target label: RD, p(label): 0.8054\n",
      "20/75: RRD NVG-20191210.jpg-Target label: RD, p(label): 0.8968\n",
      "21/75: RRD SLO-20131219 Giant tear.jpg-Target label: RD, p(label): 0.6727\n",
      "22/75: RRD WVI-20191105 (1) Recurrent with PVR and new small temporal hole.jpg-Target label: RD, p(label): 0.8143\n",
      "23/75: hij0f9pkqn441.jpg-Target label: RP, p(label): 0.8724\n",
      "24/75: Technology-1.jpg-Target label: Healthy, p(label): 0.3895\n",
      "25/75: Color-Fundus1.jpg-Target label: Healthy, p(label): 0.5547\n",
      "26/75: Color-Healthy-Child-P200Tx.jpg-Target label: Healthy, p(label): 0.2834\n",
      "27/75: Color-Healthy-P200Tx.jpg-Target label: Healthy, p(label): 0.5051\n",
      "28/75: Color-Healthy-California.jpg-Target label: Healthy, p(label): 0.4182\n",
      "29/75: Color-Healthy-California-Courtesy-Michael-Singer-MD_result.jpg-Target label: Healthy, p(label): 0.1556\n",
      "30/75: California-optomap-am-Healthy.jpg-Target label: Healthy, p(label): 0.2528\n",
      "31/75: Daytona-Projected-Healthy-Retina-Adult.jpg-Target label: Healthy, p(label): 0.3326\n",
      "32/75: P200Tx-Projected-Severe-NPDR-with-DME-HariprasadDiabeticRetinopathy22.jpg-Target label: DR, p(label): 0.6380\n",
      "33/75: P200Tx-Projected-Severe-NPDR-with-DME-Hariprasad-2DiabeticRetinopathy21.jpg-Target label: DR, p(label): 0.8729\n",
      "34/75: P200Tx-Projected-DR-with-PRP-1DiabeticRetinopathy17.jpg-Target label: DR, p(label): 0.8842\n",
      "35/75: P200Tx-Projected-Diabetic-RetinopathyDiabeticRetinopathy16.jpg-Target label: DR, p(label): 0.4584\n",
      "36/75: P200Tx-Projected-Diabetic-Retinopathy-with-Mac-Grid-1DiabeticRetinopathy14.jpg-Target label: DR,?, p(label): 0.8891\n",
      "37/75: Color-Wet-AMD-and-NPDR-OS-California-Courtesy-Mandar-Joshi-MD_result-1.jpg-Target label: DR,AMD, p(label): 0.3389\n",
      "38/75: Color-Wet-AMD-and-NPDR-California-Courtesy-Mandar-Joshi-MD_result-1.jpg-Target label: DR,AMD, p(label): 0.8571\n",
      "39/75: Color-Severe-NPDR-California.jpg-Target label: DR, p(label): 0.9066\n",
      "40/75: Color-Severe-NPDR2-California.jpg-Target label: DR, p(label): 0.4768\n",
      "41/75: Color-PDR2-OD-California.jpg-Target label: DR, p(label): 0.8857\n",
      "42/75: Color-PDR2-OS-California.jpg-Target label: DR, p(label): 0.8044\n",
      "43/75: Color-NPDR-with-Macular-Edema-OS-California.jpg-Target label: DR, p(label): 0.8390\n",
      "44/75: Color-NPDR-with-Macular-Edema-OD-California.jpg-Target label: DR, p(label): 0.6204\n",
      "45/75: California-DR-and-PRP-Sadda-1DiabeticRetinopathy1.jpg-Target label: DR, p(label): 0.8056\n",
      "46/75: California-DR-and-PRP-StangaDiabeticRetinopathy4.jpg-Target label: DR, p(label): 0.8573\n",
      "47/75: California-DR-Stanga-1DiabeticRetinopathy5.jpg-Target label: DR, p(label): 0.6470\n",
      "48/75: California-Proliferative-Diabetic-RetinopathyDiabeticRetinopathy8.jpg-Target label: DR, p(label): 0.8386\n",
      "49/75: Daytona-Projected-DR-Peripheral-DrusenDiabeticRetinopathy10.jpg-Target label: DR,Drusen(AMD?), p(label): 0.9210\n",
      "50/75: Daytona-Projected-PDR-with-PDP-1DiabeticRetinopathy11.jpg-Target label: DR, p(label): 0.6418\n",
      "51/75: Color-Wet-AMD-OS-California.jpg-Target label: AMD, p(label): 0.0725\n",
      "52/75: Color-Wet-AMD-OD-California.jpg-Target label: AMD, p(label): 0.0542\n",
      "53/75: Color-Dry-AMD-OD-California-1.jpg-Target label: AMD, p(label): 0.0791\n",
      "54/75: Color-Dry-AMD-OS-California-1.jpg-Target label: AMD, p(label): 0.2758\n",
      "55/75: Color-Atrophic-AMD-with-Geographic-Atrophy-OS-California.jpg-Target label: AMD, p(label): 0.0849\n",
      "56/75: Color-Atrophic-AMD-with-Geographic-Atrophy-OD-California.jpg-Target label: AMD, p(label): 0.0859\n",
      "57/75: California-AMD.jpg-Target label: AMD, p(label): 0.1850\n",
      "58/75: AMD-Wet-California-Courtesy-Tim-Steffens-CRA-OCT-C-FOPS.jpg-Target label: AMD, p(label): 0.8974\n",
      "59/75: California-AMD-Dry.jpg-Target label: AMD, p(label): 0.0542\n",
      "60/75: California-AMD-3.jpg-Target label: AMD, p(label): 0.5555\n",
      "61/75: California-AMD-2.jpg-Target label: AMD, p(label): 0.6783\n",
      "62/75: Daytona-Projected-AMD-1-of-2.jpg-Target label: AMD, p(label): 0.1590\n",
      "63/75: Daytona-Projected-AMD-2-of-2-1.jpg-Target label: AMD, p(label): 0.0530\n",
      "64/75: Daytona-Projected-Wet-AMD-1.jpg-Target label: AMD, p(label): 0.1077\n",
      "65/75: P200Tx-Projected-AMD-Geographic-Atrophy.jpg-Target label: AMD, p(label): 0.0897\n",
      "66/75: Glaucoma-Suspect-California-William-Lesko-MD-North-Jersey-Eye-Associat..._result.jpg-Target label: Gla, p(label): 0.5431\n",
      "67/75: Color-Stereo-Healthy-OD-California_leftpanel.jpg-Target label: Healthy, p(label): 0.2304\n",
      "68/75: Color-Hemi-Retinal-Vein-Occlusion-California.jpg-Target label: RVO, p(label): 0.8426\n",
      "69/75: P200Tx-Projected-Retinal-Vein-OcclusionOcclusion33.jpg-Target label: RVO, p(label): 0.6269\n",
      "70/75: P200Tx-Projected-Retinal-Vein-Occlusion-with-Ozurdex-injectionOcclusion31.jpg-Target label: RVO, p(label): 0.1356\n",
      "71/75: Color-Stereo-Healthy-OS-California_leftpanel.jpg-Target label: Healthy, p(label): 0.2729\n",
      "72/75: Color-BRVO-California.jpg-Target label: RVO, p(label): 0.2017\n",
      "73/75: ASRS-RIB-Image-78895.jpg-Target label: Healthy, p(label): 0.3155\n",
      "74/75: ASRS-RIB-Image-27747.jpg-Target label: DR, p(label): 0.9067\n",
      "75/75: ASRS-RIB-Image-25718.jpg-Target label: RD, p(label): 0.8992\n"
     ]
    }
   ],
   "source": [
    "extset_df = pd.read_csv('PATH/TO/EXTERNALVALSET/external_valset.csv')\n",
    "\n",
    "for img_idx in tqdm(range(extset_df.shape[0])):\n",
    "  img_LR = extset_df.LR[img_idx]\n",
    "  img_label = extset_df.Label[img_idx]\n",
    "  img_name = extset_df.Filename[img_idx]\n",
    "  img_path = 'PATH/TO/EXTERNALVALSET/' + img_name\n",
    "\n",
    "  preds = predict_on_img(model, img_path=img_path, image_LR=img_LR, \n",
    "                         # Set this to True to see GradCams\n",
    "                         # This is set to False here since we cannot share these images\n",
    "                         plot_grad_cam=False,\n",
    "                         print_preds=False)\n",
    "  if img_label=='Healthy':\n",
    "    extset_df.loc[img_idx,'p(label)'] = 1-preds[-1]\n",
    "  else:\n",
    "    # ugly hack\n",
    "    try:\n",
    "      img_labelidx = targets.index(img_label)\n",
    "    except:\n",
    "      img_labelidx = targets.index('DR')\n",
    "    extset_df.loc[img_idx,'p(label)'] = preds[img_labelidx]\n",
    "  extset_df.loc[img_idx, [f'p({t})' for t in targets]] = preds\n",
    "  extset_df.loc[img_idx, f'max(p(l))'] = np.max(preds[:-1])\n",
    "  print(f'{img_idx+1:<2}/{extset_df.shape[0]}: {img_name}', end='-')\n",
    "  print(f'Target label: {img_label}, p(label): {extset_df.loc[img_idx,\"p(label)\"]:.4f}')\n",
    "  extset_df.loc[img_idx,'p(any)'] = preds[-1]\n",
    "  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "d9ed09ea",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Healthy vs unhealthy AUC: 0.9841\n",
      "DR: 0.9627, AMD: 0.8051, RP: 1.0000, RVO: 0.9754, RD: 1.0000, Gla: 0.9595, "
     ]
    }
   ],
   "source": [
    "auc = roc_auc_score(y_true=extset_df.Label!='Healthy', y_score=extset_df['p(any)'])\n",
    "print('Healthy vs unhealthy AUC:', np.round(auc,4))\n",
    "\n",
    "for l in ['DR', 'AMD', 'RP', 'RVO', 'RD',  'Gla']:\n",
    "  auc = roc_auc_score(y_true=[l in _ for _ in extset_df.Label], \n",
    "                      y_score=extset_df[f'p({l})'])\n",
    "  print(f'{l}: {auc:.4f}', end=', ')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "2167cc42",
   "metadata": {},
   "outputs": [],
   "source": [
    "extset_df['Label_plot']=[_ if not 'R,A' in _ else 'DR+AMD' for _ in extset_df['Label'].values]\n",
    "extset_df['Label_plot']=[_ if not '?' in _ else 'DR' for _ in extset_df['Label_plot'].values]\n",
    "label_n_dict = extset_df['Label_plot'].value_counts().to_dict()\n",
    "extset_df['Label_plot'] = [f'{_}\\n(n={label_n_dict[_]})' for _ in extset_df['Label_plot']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "f1072e3e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x480 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(dpi=120)\n",
    "col_order = [\n",
    "'Healthy\\n(n=12)',\n",
    " 'DR\\n(n=18)',\n",
    " 'DR+AMD\\n(n=2)',\n",
    " 'AMD\\n(n=15)',\n",
    "  'RP\\n(n=10)',\n",
    " 'RVO\\n(n=10)',\n",
    " 'RD\\n(n=7)',\n",
    " 'Gla\\n(n=1)',\n",
    "]\n",
    "# setting seed to get a decent looking plot\n",
    "np.random.seed(47)\n",
    "sns.stripplot(data=extset_df,y='p(any)', x='Label_plot', size=4, jitter=0.12,\n",
    "              order=col_order, palette=sns.color_palette('bright'))\n",
    "plt.axhline(0.79, c='orangered', linestyle='--')\n",
    "plt.axhline(0.28, c='dodgerblue', linestyle='--')\n",
    "plt.ylim(0,1)\n",
    "plt.ylabel('$\\hat{p}$(diseased)',fontsize='large')\n",
    "plt.xlabel(None)\n",
    "plt.axvline(0.5, linestyle=':', c='gray')\n",
    "plt.gca().set_aspect(3.5)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "60f74472",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAApEAAAG/CAYAAAAAZYwQAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMiwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8rg+JYAAAACXBIWXMAABJ0AAASdAHeZh94AABinklEQVR4nO3deZyVcxvH8c/VvmhTopAo1LSXZCey81jK+pCYUmSN7LIXlSihbVL2kiUesstetpSESAmVSirtNXM9f9xnxumYmeYs0z3L9/163a+Z87u3a849yzW/1dwdEREREZF4lAk7ABEREREpfpREioiIiEjclESKiIiISNyURIqIiIhI3JREioiIiEjclESKiIiISNyURIqIiIhI3JREioiIiEjclESKiIiISNyURIqIiIhI3JREioiIiEjclESKiIiISNyKdBJpZjuY2R1m9rqZrTAzN7NucZxf08xGmdkyM1trZu+ZWdtCDFlERESkVCjSSSRQB+gHNAVmxnOimZUBXgXOBYYD1wF1galmtneK4xQREREpVcqFHcA2LAbqufsSM9sP+DyOc7sABwFnuPskADObCMwF7iBILkVEREQkAUW6JtLdN7r7kgRP7wL8AbwQdb1lwETgFDOrmIIQRUREREqlol4TmYw2wFfunhVT/hlwMbAP8E1uJ5pZXWCnmOIdIufMBjalNlQRERGRlKoA7A687+6rCuMGJTmJrAd8kEv54sjH+uSRRAKXArcVRlAiIiIi29EpwMuFceGSnERWBjbmUr4han9eHgGeiylrAkx66aWXaNy4cQrCExEpnlasWMFhhx0GwOTJk6lVq1bIEYkIwIYNG+jWrRuLFi2ibt26LF26FODXwrpfSU4i1wO59XusFLU/V+6+FFgaXWZmADRu3JhmzZqlKEQRkeJn2bJlOZ8feOCB7LRTbO8fEdle3D0nRwF48sknee211zjrrLNo3749FGIXvCI9sCZJiwmatGNlly3ajrGIiIiIpMy6deu49tprueiii7Yq79ixI4MGDaJy5fwaXFOjJNdEfg0camZlYgbXdADWEUz1I0VYZmYmK1asCDsMEYmxfPnysEMQKdWmTp1K9+7dmTdvHgBnnnkmxx9//HaPo0QkkWZWD6gBzHP3zZHiSQTT/Jwe+RwzqwOcAbzi7rn1l5QiIjMzk3bt2jFzZlxzzIuIiJRYq1at4rrrrmPUqFE5ZR06dKBBgwahxFPkk0gzuwyoSTCaGuBkM9st8vlDkWHrA4ALgD2BBZF9k4BpwGNmlgYsJxh1XRaNvC7yVqxYoQRSpIhr1aoVO+64Y9hhiJQK//vf/+jVqxe///47AFWqVOGee+7h8ssvp2zZsqHEVOSTSOBaYI+o16dHNoAngVznPnL3TDM7ARgEXEEwGvtzoJu7/1B44UqqzZkzhzp16oQdhojE2HHHHUP74yVSWixbtowrr7ySZ555JqfsqKOOYtSoUey1114hRlYMkkh3b1iAY7oB3XIp/wvoHtmkmKpTp45Gf4qISKm0ZMkSnnsumHWwRo0aDBkyhAsvvHCrEdlhKfJJpIiIiEhp1aJFC2666SZmzZrFww8/TP369bd90naiJLKEK64jnDX6U0RESpusrCxGjx7NDjvswH//+9+c8n79+lGmTJkiUfsYTUlkCaYRziIiIsXDTz/9RI8ePZg6dSo1atTgyCOPpF69YGrrotr3uCRPNl7qlYQRzhr9KSIiJdmWLVsYPHgwLVq0YOrUqQBUr16d3377LdzACkA1kaVEcR3hrNGfIiJSUs2aNYv09HS++OKLnLLevXszYMAAqlWrFmJkBaMkspTQCGcREZGiYePGjfTv35/+/fuzZcsWAPbee28yMjI49NBDQ46u4NScLSIiIrIdTZo0iTvvvJMtW7ZQtmxZbrjhBmbOnFmsEkhQTaSIiIjIdnXOOecwevRoVq5cSUZGBu3atQs7pIQoiRQREREpRO+++y7169enSZMmAJQpU4aJEydSq1YtypcvH3J0iVNztoiIiEghWLlyJT169OCoo44iPT2drKysnH1169Yt1gkkKIkUERERSbnJkyeTlpbGmDFjAJg5cyazZ88OOarUUhIpIiIikiJLly7l7LPP5tRTT2Xx4sUAHH300cyePZuWLVuGHF1qKYkUERERSZK78+STT9K0aVMmTJgAQK1atRg3bhxvvPEGDRs2DDfAQqAkUkRERCRJgwYN4vzzz2fFihUAdO7cmTlz5nDBBRcUuTWvU0VJpIiIiEiSunXrxo477sjOO+/MpEmTmDRpErvsskvYYRUqTfEjIiIiEqcff/yR3XffnUqVKgHBaOuXX36ZtLQ0atWqFXJ024dqIkVEREQKaMuWLdx77720aNGCO++8c6t9Bx98cKlJIEFJpIiIiEiBfP3113To0IEbb7yRjRs3MnjwYBYtWhR2WKFRc3acVqxYwbJly8IOo0CWL18edggiIiLF3oYNG7jrrru47777yMzMBGDfffclIyOD+vXrhxxdeJRExumwww4LOwQRERHZTj755BPS09P5/vvvAShbtizXX389t956a05/yNJKSWQp0KpVK3bcccewwxARESlWBgwYwM0334y7A9CmTRvGjh1L69atww2siFASGafJkydz4IEHhh1GXHbccUfKli0bdhgiIiLFSvv27XF3KlasyB133EGfPn2K/XrXqaQkMk61atVip512CjsMERERSbH169dTuXLlnNedOnXi/vvv58QTT2TfffcNMbKiSaOzRUREpNR74YUX2GuvvXjzzTe3Ku/Tp48SyDwoiRQREZFSa8mSJXTp0oXOnTuzZMkSLr74YtatWxd2WMWCkkgREREpddyd8ePHk5aWxvPPPw8EYwjuuuuurZq0JW/qEykiIiKlyi+//ELPnj154403csrOPPNMhg0bxs477xxiZMWLaiJFRESk1BgxYgTNmjXLSSB32WUXXnzxRSZMmKAEMk5KIkVERKTU+P3331m7di0A6enpzJkzh1NPPTXcoIopNWeLiIhIqXHLLbfw1Vdf0adPH4466qiwwynWVBMpIiIiJdJXX33FUUcdxZIlS3LKKlasyKuvvqoEMgWURIqIiEiJsn79em688Ub2339/3n33XS6//PKwQyqR1JwtIiIiJcZHH31Eeno6c+fOBaBcuXI0bdqUrKwsypRR3VkqKYkUERGRYu/vv//mxhtv5OGHH84pa9euHWPHjqVly5YhRlZyKSUXERGRYu3111+nefPmOQlkpUqVGDRoENOmTVMCWYhUEykiIiLFlrvTv39/Fi5cCMDhhx/OmDFjaNy4cciRlXyqiRQREZFiy8wYPXo0devWZcSIEbz77rtKILcTJZEiIiJSbCxevJhbb72VrKysnLJ9992XBQsW0LNnTw2e2Y7UnC0iIiJFnrvz2GOP0adPH1atWkW9evW49NJLc/ZXrlw5xOhKJyWRIiIiUqTNnz+fiy++mLfffjunbMaMGSFGJKDmbBERESmiMjMzGTp0KM2bN89JIOvXr8/kyZMZPXp0yNGJaiJFRESkyJkzZw7du3fn008/zSnr0aMHgwYNokaNGiFGJtmURIqIiEiRsnHjxq3WvN5rr70YPXo0Rx55ZMiRSTQ1Z4uIiEiRUrFiRe69917KlCnDNddcwzfffKMEsghSTaSIiIiEat26dXz44Ycce+yxOWVdu3alffv2pKWlhRiZ5Ec1kSIiIhKa999/n1atWnHSSScxc+bMnHIzUwJZxCmJFBERke1u9erVXHLJJRxxxBH89NNPbNmyhYceeijssCQOSiJFRERku3r11Vdp1qwZI0aMAIKJwu+//35GjhwZcmQSD/WJFBERke1i+fLlXHXVVTz11FM5ZR07dmT06NE0atQoxMgkEUklkWZWEzgISAPqAA4sB74DPnX3v5INUERERIq/5cuXk5aWxrJlywCoXr06999/P+np6ZhZyNFJIuJOIs2sAnAu0A04hLybxLPM7GPgMeAZd9+YaJAiIiJSvNWpU4cTTjiB8ePHc/LJJ/Poo4+y6667hh2WJCGuPpFm1gv4GRgBrAauJkgk6wOVgSrArpGyPsCqyLHzzKxn6sIWERGRoszd+f3337cqGzJkCBMnTmTy5MlKIEuAeGsibwIGA4+5+6o8jlkc2T4BhplZdeAi4EZAPWZFRERKuHnz5tGjRw8WLFjAN998Q9WqVQHYcccdOeOMM0KOTlIl3tHZe7n7g/kkkP/i7qvd/UGgcZz3EhERkWIkMzOTIUOG0KJFC9577z3mz5/PfffdF3ZYUkjiqol09y2J3iiZc0VERKRomz17Nunp6Xz22Wc5Zb169eLaa68NMSopTEnNE2lm35nZTWa2R6oCEhERkeJj06ZN3HHHHbRt2zYngWzcuDFTp07l0UcfpXr16iFHKIUl2cnGfwXuIBg484GZdTezGimIS0RERIq4BQsW0K5dO26//XY2b95MmTJl6Nu3LzNnzuTwww8POzwpZEklke5+DLAb0JdgdPYoYImZTTKzU8ysfApiFBERkSJol112YdOmTQC0aNGC6dOnM3DgQKpUqRJyZLI9JL3sobv/4e4PuHt7oCnB6O3WwAsECeUjZnZQItc2s4pmdp+ZLTKz9WY23cyOLuC5nczsPTNbbmYrzewzMzs/kThEREQkkJmZmfN5pUqVyMjI4M477+SLL75gv/32CzEy2d5Suna2u//g7rcSzBM5CagF9AI+NLMfzay3mcVzz3EE800+BVwJZAKvmdkh+Z1kZv8B3gQqALcDNwPrgcfN7Oq4vigRERFh1apV9OzZk/POO2+r8kMOOYRbb72VChUqhBSZhCVla2ebWVXgNOA84MhI8f+Ax4FNwMXAMKAlsM2Jx81sf+BsoK+7D46UPQ7MBgYSLLeYl8sI5qo8MnulHDMbCXxPsNLOA/F9dSIiIqXXK6+8Qq9evVi0aBEAZ599NqecckrIUUnYkh2dXdbMTjCzp4E/CBLG2sA1QH13/4+7T3L3l939JOA+gsSwILoQ1DyOyi5w9w1ABnCgme2ez7nVgb+il1qMTDG0nKBGUkRERLZh2bJlnHvuufznP//JSSA7depEy5YtQ45MioJkayKXADsCvwMPAY+7+3f5HD8LqFbAa7cB5rr76pjy7AmoWhOMDs/NVOB6M7sLGA84wXrf+wFnFvD+IiIipZK788wzz3DFFVfw559/AlCjRg2GDBnChRdeiJmFHKEUBckmka8CTwDvurtv62B3fxZ4toDXrkfQJB0ru6x+PufeBexJ0BfylkjZOqCzu0/e1o3NrC6wU0xxo22dJyIiUtz99ttv9OrVi1dffTWn7NRTT+Xhhx+mfv38/vRKaZNsEjkW+C6vBNLM6gBp7v5BAteuDGzMpXxD1P68bATmEgzueQEoS9An80kzO9rdp23j3pcCt8UXroiISPH3999/89ZbbwFQt25dhg8fTpcuXVT7KP+SbBL5HnA+8HQe+4+K7CubwLXXAxVzKa8UtT8vw4EDgLbungVgZhOBb4GhQIdt3PsR4LmYskbANmsxRUREirOmTZty22238cMPPzBkyBBq164ddkhSRCWbRG7r35KKBINjErEY2DWX8nqRj4tyDcisApAODMxOIAHcfbOZTQEuM7MK7r4prxu7+1Jgacx14wxfRESkaNuyZQsPPPAAtWvX5qKLLsopv/HGG/V3T7Yp7iTSzBoADaOKmpjZYbkcWpNgKp9fEooMvgY6mln1mME1HaL256Y2wdeVW+1neYIR6YnUjIqIiJQYs2bNIj09nS+++IJq1apxzDHHsNtuuwGqOJGCSWSKnwsJRj+/RzDq+ebI57HbS8D+wIAEY5vEP30ZgWAFm8j9p7v7r5GyBmbWJOq8pcBK4LRIrWT2uTsAJwPfu7um+RERkVJp48aN9OvXj3bt2vHFF18AwfKFy5cvDzkyKW4Sac6eSDDht0U+HwZ8GHOMA2uBr939j0QCc/fpZvYcMCAyWvon4AKCWtD0qEMfBw6PxIO7Z5rZYOBuYFpkgvKykXN2I5gMXUREpNSZNm0a6enpzJkzB4CyZcvSt29f+vXrR+XK+Y1XFfm3uJPIyDyQ3wGY2YXAB+4+P9WBRXQlmK7nfIIlFGcBJ21rtLe732Nm8wmWSryNoG/mLKCLuz9fSLGKiIgUSWvXruWWW25h6NChZE+o0rp1azIyMmjbtm3I0UlxldTAGncfn6pA8rj+BqBvZMvrmCPyKH+avEeNi4iIlBovvfQSDz74IAAVKlTgtttuo2/fvpQvXz7cwKRYiyuJNLOxBE3VF0eajccW4DR39/RtHyYiIiKF4dxzz2XcuHGsW7eOjIwMmjRpsu2TRLYh3prII4EsggE5mZHX21qpZpsr2YiIiEjqTJ48mUaNGtG8eXMgGG09ceJEqlevTtmymqBEUiOuJNLdG+b3WkRERMLzxx9/cMUVVzBx4kTat2/Pp59+mpM01qpVK+TopKRJZIofERERKULcnSeeeIK0tDQmTpwIwE8//cT3338fcmRSkiWVRJrZZ2Z2tZntlqqAREREpOAWLlzIiSeeSNeuXVmxYgUAXbp0Yc6cOTRr1izk6KQkS7YmMhO4H1hgZh+Z2WVmtksK4hIREZF8ZGVl8cgjj9CsWTOmTJkCwM4778zzzz/Pc889xy676M+xFK6kkkh3P5Bg8u8bCeZiHAb8ambvmtnFZlYn+RBFREQk1oABA+jduzdr1qwB4MILL+S7777j9NNPDzkyKS2S7hPp7gvdfZC7twcaA/0IJgYfASwys9eTvYeIiIhsrVevXuy0007ssccevPHGG4wdO1aDZ2S7SunAGnf/2d0HAG2BnsB64OhU3kNERKQ0mjVrFuvWrct5Xbt2baZMmcLs2bM55phjQoxMSquUJpFmdoCZDQEWEtREglaNERERSdiGDRu4+eabadu2LbfddttW+9q1a8cOO+wQUmRS2iWdRJpZOzMbaGYLgI+BiyMfuwB13f38ZO8hIiJSGn3yySe0adOG/v37k5mZyUMPPcSSJUvCDksESHLtbDObRzCwZhMwBbgeeMXd1+V3noiIiORtzZo13HTTTQwfPhz3YOG3Nm3aMHbsWI26liIjqSQSmAPcBkx2979TEI+IiEip9uabb3LxxRfzyy+/AFCxYkXuuOMOrrnmGsqVS/bPtkjqJPXd6O4npyoQERGR0u62227jzjvvzHl9yCGHMGbMGPbdd98QoxLJXVx9Is2sgZk1iH29rS31YYuIiJQ8Rx55JAA77LADDz/8MO+//74SSCmy4q2JXAC4mVV2903ZrwtwXtk47yMiIlLirVq1iho1auS8Pvzww3n00Uc5/vjj2WOPPUKMTGTb4k0iLyJIGjfHvBYREZECcnfGjx9Pnz59eOKJJzjxxBNz9vXq1SvEyEQKLq4k0t3H5fdaRERE8rdgwQJ69uzJm2++CUDPnj2ZO3cuVapUCTkykfgkNU+kmY01sw757N/fzMYmcw8REZGSICsri4ceeojmzZvnJJD16tVj+PDhSiClWEp2svFuQKN89u8JXJDkPURERIq17777jkMPPZQrrriCtWvXApCens6cOXM49dRTww1OJEEpXfYwF/UJ1s8WEREplQYNGkTr1q355JNPANhzzz15++23GTNmDDVr1gw3OJEkxD1PpJmdApwSVXSxmXXK5dCaQCfg88RCExERKf7Wr1/Ppk2bMDOuvPJK7r77bqpWrRp2WCJJS2Sy8TTgjMjnDnQA2sUc48Ba4AOgT8LRiYiIFDPujpnlvL7hhhv49ttvufrqqznggANCjEwkteJuznb3Ae5ezd2rAQakZ7+O2qq7ez13P8nd56Y+bBERkaLnww8/5IADDmDRokU5ZRUqVGDChAlKIKXESapPpLuXcfenUxWMiIhIcfT333/Tu3dvDjvsMD777DMuueQS3DWNspRsWsldREQkCVOmTKFnz578+uuvAFSqVIlDDz30X83aIiVNXEmkmWUBWUAVd98Ueb2tf7Xc3ZWsiohIifLnn39y9dVX88QTT+SUHX744YwZM4bGjRuHGJnI9hFvcncnQdK4Jea1iIhIqeDuTJo0icsuu4ylS5cCUK1aNQYNGkSPHj0oU6awZ88TKRriXfbw9vxei4iIlAYjRozISSBPPPFERowYwW677RZyVCLbl/5dEhERiYOZMWrUKBo0aMBTTz3FK6+8ogRSSqVk184+ysz6xpRdZGYLzewPM3vAzMomF6KIiEh45s+fT58+fcjMzMwpa9SoET/99BPnnnuuBs9IqZVsTeTtQKvsF2bWAhgJLAOmAlcA1yZ5DxERke0uMzOToUOH0rx5cx544AEeeuihrfaXL18+pMhEioZkR003BZ6Pen0+sBo41N3XmdkIoCtwX5L3ERER2W7mzJlDeno606ZNyylbuHBhiBGJFD3J1kRWJUgasx0HvO7u6yKvPwf2SPIeIiIi28WmTZu46667aNOmTU4C2ahRI959912GDBkScnQiRUuySeSvQHsAM2sMNAfejNq/I7AxyXuIiIgUus8//5z99tuPfv36sWnTJsqUKcO1117LrFmz6NixY9jhiRQ5yTZnPwX0M7NdgWbAX8DkqP3tAK2dLSIiRdqGDRv4z3/+w5IlSwBo3rw5GRkZ7L///iFHJlJ0JVsTeQ9wL7A7sBA41d1XApjZjsARwMtJ3kNERKRQVapUiSFDhlC+fHnuuOMOvvzySyWQItuQVE2ku28Bbo5ssftWALskc30REZHCsHr1at555x1OO+20nLKzzz6bAw44gD333DPEyESKD002LiIipcqrr75Ks2bN6NKlC1988UVOuZkpgRSJQ7J9IjGzpsCFwF5ALSB21lV396OSvY+IiEgyli1bxlVXXcXTTz+dU/bYY4+x3377hRiVSPGVVBJpZucDjwGbgR8IBtb867Bk7iEiIpIMd2fChAlcfvnlLF++HIDq1atz//33k56eHnJ0IsVXsjWRtwMzgOPdfXny4YiIiKTO77//ziWXXMIrr7ySU3byySfz6KOPsuuuu4YYmUjxl2wSWR8YrARSRESKmqVLl9KsWTNWrVoFwE477cRDDz3EmWeeqfWuRVIg2YE1swgSSRERkSKlbt26nHnmmQD897//Zc6cOZx11llKIEVSJNkksg+QbmYHpSIYERGRRGVmZvLzzz9vVTZo0CBee+01nnzySerUqRNSZCIlU7LN2dcDq4APzWwOwYTjmTHHuLufkuR9RERE8jR79mzS09NZsmQJs2fPplq1agDUqFGD448/PuToREqmZGsiWwKVCJLHHYA0oEUum4iISMpt2rSJ22+/nbZt2/LZZ5+xcOFCBg8eHHZYIqVCsivWNExRHCIiInGZPn066enpfPvttwCUKVOGa6+9lhtuuCHkyERKh6QnGxcREdme1q5dy6233sqDDz6IuwPQsmVLMjIyNHG4yHaUihVrygJnAB2BukA/d//GzGoARwEfu/sfyd5HRETkxx9/5LjjjssZQFOhQgX69evHddddR/ny5UOOTqR0SXbFmprA68D+wBqgKvBQZPcaYBjwOHBTMvcREREBaNCgAZUqVQLgwAMPJCMjg6ZNm4YclUjplOzAmnuBZsCxBGtn50y+5e6ZwCTghCTvISIipdimTZtyPq9YsSJjx47lwQcf5MMPP1QCKRKiZJuzTwUecve3zKx2LvvnAt2SvIeIiJRCS5cu5corr2TTpk08//zzOeUdOnSgQ4cOIUYmIpB8TWQNYH4++8ujwTsiIhIHd+epp54iLS2NZ599lhdeeGGrJFJEioZkE7x5QNt89h8DzEnyHiIiUkr8+uuvXHLJJbz66qs5ZaeddhoHHaSF0USKmmRrIscAF5nZWfzTH9LNrKKZ3QMcB4xM8h4iIlLCZWVlMWLECJo1a5aTQNatW5fnnnuO559/nnr16oUcoYjESrYmcijBwJpngJWRsqeB2pFrj3T3jCTvISIiJdjPP//MRRddxPvvv59T1rVrV4YMGULt2rl1txeRoiDZFWsc6GFm44EuwN4EtZvzgInu/kHyIYqISEmWmZnJ9OnTgWAKn5EjR3LccceFHJWIbEtKBr24+0fAR6m4loiIlC577703d911FwsWLGDAgAFUq1Yt7JBEpABSOnLazOoAxwP1gB+AV9w9K4nrVQTuBM4HagGzgFvc/a0Cnn8WcBXQEthMMMjnFnd/N9GYREQkcRs3buSee+6hXr16XHLJJTnl1157bYhRiUgi4k4izewcoDtwlrsvjyo/EHiFINkzwIHPzKyTu69NML5xBM3kDwI/Esw5+ZqZdYzUfuYX5+1AP4IJz8cRTDfUHNg1wVhERCQJ06ZNIz09nTlz5lC1alWOP/54GjZsGHZYIpKgRGoizwHKxySQBjxBMG/kncAXwIlAL+A64LZ4b2Jm+wNnA33dfXCk7HFgNjAQyHO+BzM7gCCBvMbdH4j33iIikjpr167llltuYejQoQRd6YMm7HXr1oUcmYgkI5EpfloBsQNmDiJY9vBRd7/D3V9190uB/wGnJxhbFyATGJVd4O4bgAzgQDPbPZ9zrwKWAEMtsEOCMYiISBLefvttmjdvzoMPPoi7U7FiRfr3789nn31GWlpa2OGJSBISSSLr8u9Vao4haL6eEFP+FkFymYg2wFx3Xx1T/lnkY+t8zj0K+By4AlgG/G1mi83ssoLc2Mzqmlmz6A1oFF/4IiKl18qVK0lPT+foo49mwYIFABx88MF8/fXX3HjjjZQvXz7cAEUkaYk0Z/9J0O8x2iEEA1e+jClfS5BcJqIesDiX8uyy+rmdZGa1gDrAwcCRwB3AQuBC4CEz2+zu25oA/VISaIIXEZHAlClTGDt2LABVq1bl3nvv5dJLL6VMmWTXuBCRoiKRJHIWcLaZPejuW8xsV4KE7b1Ic3O0RsCiBGOrDGzMpXxD1P7cZDdd1wbOdvcJAGY2CfgGuIVtr6LzCPBcTFkjYPI2zhMREeDss8/miSeeICsri5EjR7LHHnuEHZKIpFgiSWR/4H3gKzP7nKDpuDwwJJdjTyZoVk7EeqBiLuWVovbndR4ENaOTsgvdPcvMJgB3mFkDd1+Y143dfSmwNLosGDskIiKx3J0nnniCFi1a0KZNGyD4nTlx4kSqVq2q358iJVTc7QqRqXXOjpx7LkHNYPfYuRvN7EhgTxKvvVtM0KQdK7ssrxrOFZGY/nT3zJh92YlhbHO8iIgk4JdffuH444/nggsu4KKLLmLz5s05+3bYYQclkCIlWEKdU9z9OXdv7u6V3b2Juz+WyzHvuns1d5+YYGxfA/uYWfWY8g5R+3OLLSuybyczqxCzO7sf5bIEYxIRESArK4uHH36Y5s2b88YbbwCwePFifvrpp5AjE5HtpSj3cJ4ElAUuzi6IrGBzITDd3X+NlDUwsyYx506InHtB1LmVgP8Cc9w90X6aIiKl3g8//MDhhx/OZZddxpo1awC48MIL+e6772jatGnI0YnI9hJXn8jIajXPevZssQU/zwgGuTxT0HPcfbqZPQcMMLO6wE8ESWFDID3q0MeBwwlWyck2kmBVnYfNbB+C0dnnA3sQ9NMUEZE4bd68mfvvv5/bb7+djRuDcY977LEHo0aN4phjjgk5OhHZ3uKtiXwQmGtm15nZnts62Mwam9lNBAlgIivHdI3c83xgGMEAnpPcPXay8624+3qC6X2eBi4CBgFZwInuPiWBOERESr3+/ftz4403snHjRsyMK664gtmzZyuBFCmlLJ5KRTOrSrAazBUEczEuAL4imHz8L4LawFoEA2r2A3YnmFdyGPBAEmtohy4y4fjsDz74gEMPPTTscEREtru//vqLtLQ0atasSUZGBgcdlOfqsyISsm+//ZbmzZsDNHf3bwvjHnE1Z0eSwHvM7D6CZuFTCJY8PJ1/mpMdmEcwDdBk4BV335zL5UREpAj79NNPadGiBTvsEEy/W6tWLd566y0aN25MpUqVtnG2iJR0iY7O3uLuL7p7N3ffB6gA7BzZKrj7Pu5+gbu/oARSRKR4WbNmDVdccQUHH3wwN99881b7mjdvrgRSRIAUjc5290x3XxbZYudmFBGRYuLNN9+kefPmPPTQQ7g7Y8aM4Y8//gg7LBEpgoryFD8iIrKdrFixggsvvJBjjz2WX375BYBDDz2UGTNmsPPOO4ccnYgURUoiRURKueeff560tDTGjRsHBCvNPPLII0ydOpV99tkn3OBEpMhKZO1sEREpIa677joGDRqU8/r4449nxIgRNGjQIMSoRKQ4UE2kiEgpduqpp2Jm1K5dmyeeeIJXX31VCaSIFIhqIkVESpGlS5dSt27dnNcHHXQQ48aN47jjjtuqXERkW1JSE2lmFc3sQDM7xczqpOKaIiKSOpmZmQwbNoy99tqLF198cat9Xbt2VQIpInFLOok0syuAxcBHwAtAy0h5HTNbbmYXJXsPERFJ3Hfffcdhhx3GlVdeydq1a+nduzfr168POywRKeaSSiLN7EKCta1fB9L5Z9Ua3H058C5wdjL3EBGRxGzevJl77rmH1q1b88knnwCw55578sQTT1C5cuWQoxOR4i7ZPpHXAJPd/Vwzq53L/i8J1tkWEZHt6KuvvuKiiy5i5syZAJgZV111FXfddRdVq1YNOToRKQmSbc5uDEzJZ/8KILfkUkRECsntt9/O/vvvn5NApqWl8cknnzBkyBAlkCKSMskmkSuB/AbSpAFLkryHiIjEoVKlSmRmZlKuXDn69evHV199xQEHHBB2WCJSwiTbnP0acLGZPRK7w8yaAT2AsUneQ0RE8uHumOV0Sefaa69l7ty5XHXVVbRs2TLEyESkJEu2JvIWoCwwG7gbcOACM3sS+AJYCtyZ5D1ERCQPU6ZMoUWLFixcuDCnrFy5cowdO1YJpIgUqqSSSHdfBLQjGJ19FsHo7POBk4FngAMio7RFRCSF/vzzT7p27coJJ5zAt99+S69evXD3sMMSkVIk6RVr3H0p0B3obmY7ESSmy9w9K9lri4jI1tyd5557jssuu4xly5YBUK1aNU455ZSQIxOR0ibZeSLHmlmH7Nfuvszd/8hOIM1sfzNTn0gRkRRYtGgRp59+OmeddVZOAnniiScyZ84cevbsuVW/SBGRwpZsn8huQKN89u8JXJDkPURESjV3JyMjg7S0NF566SUA6tSpw9NPP80rr7zCbrvtFm6AIlIqpWTt7HzUB7S2lohIkiZOnMiqVasAOPfcc5kzZw7nnHOOah9FJDRx94k0s1OA6M43F5tZp1wOrQl0Aj5PLDQREYFgtZmRI0dy3HHHMWjQIE4++eSwQxIRSWhgTRpwRuRzBzoQjNCO5sBa4AOgT8LRiYiUQt9++y1Dhw7lkUceoVy54Nd0w4YNmTNnDmXKFHYDkohIwcSdRLr7AGAAgJllAenu/nSqAxMRKW02bdrEvffey913383mzZvZe++96du3b85+JZAiUpQkNcWPu+s3mohICnz++eekp6fzzTffAEHCuHr16pCjEhHJW9LzRIqISOLWrVvHbbfdxpAhQ8jKCqbXbdGiBRkZGbRv3z7k6ERE8pZ0TaKZHW9mb5nZn2a2xcwyY7dUBCoiUtJMnTqVVq1aMXjwYLKysihfvjx33HEHX3zxhRJIESnykqqJNLPOwETgW+BZ4BLgaYLlD08BfgReSi5EEZGSZ8OGDZxzzjksWbIEgA4dOpCRkUGzZs1CjkxEpGCSrYm8EfgMaAPcFikb6+7/BZoD9YD5Sd5DRKTEqVSpEsOHD6dy5coMGTKEjz/+WAmkiBQrySaRacCz7p4JbImUlQdw9wXAI8D1Sd5DRKTYW758OU8/vfVEFp07d2bevHlcffXVlC1bNqTIREQSk2wSuQ7YBODuK4GNBLWP2f4gWPpQRKRUcneeffZZmjZtynnnnce0adO22l+vXr08zhQRKdqSTSJ/IKiNzPY1cL6ZlTOzSsC5wMIk7yEiUiz9/vvvnHLKKZxzzjksX74cd+e5554LOywRkZRINol8ETjFzCpGXt8DHAGsBJYBhwL3JnkPEZFixd0ZPXo0aWlpvPLKKwDstNNOTJgwgcGDB4ccnYhIaiQ72fhgYHDU6/+Z2RHA6UAm8Kq7v5fMPUREipN58+bRo0cP3nvvn1995513Hg8++CC1a9cOMTIRkdRK+WTj7v4h8GH2azOr5u5/p/o+IiJFzR9//EGrVq1Yu3YtALvvvjsjRozghBNOCDkyEZHUK7RlC82srpn1R30iRaSU2HnnnenWrRsAl156KbNnz1YCKSIlVkI1kWZWF+gKNAL+Ap539y8j+3YFbga6AZWAqakIVESkqNm4cSPz5s0jLe2f8YUDBgzg7LPP5pBDDgkxMhGRwhd3EmlmTYAPgNoEK9MAXGdm5wEOjCFIHp8HBmUnlyIiJcn06dNJT0/nr7/+4ttvv6VmzZoAVKtWTQmkiJQKiTRn3wXsAFxKsCrNycDPwIPAOGAKsK+7n60EUkRKmrVr19KnTx8OPPBAvv32WxYtWsSDDz4YdlgiIttdIs3ZhwGPuvvIyOs5ZraFIHkc7+4Xpiw6EZEi5N1336VHjx78/PPPAFSoUIF+/fpx3XXXhRyZiMj2l0gSWRuYFVM2M/LxxeTCEREpelauXEnfvn0ZM2ZMTtmBBx5IRkYGTZs2DTEyEZHwJJJElgE2x5Rlv16TXDgiIkXL999/z1FHHcWiRYsAqFKlCgMGDKB3795a71pESrVE54ncz8w2RL2uRjCo5hAzqxl7sLu/kOB9RERC1ahRI3baaScWLVpEp06dGDVqFHvuuWfYYYmIhC7RJPKqyBbr9lzKHNC/6yJSLLg769ato2rVqgCUL1+esWPHMnPmTLp164aZbeMKIiKlQyJJZMeURyEiUgT8+uuv9OrVizJlyvDyyy/nJIxt27albdu2IUcnIlK0xJ1Euvv7hRGIiEhYsrKyGDlyJNdffz1//x2s0jpx4kTOOuuskCMTESm6Ur52tohIcfLjjz/SvXt3Pvjgg5yyCy64gE6dOoUYlYhI0Vdoa2eLiBRlW7ZsYeDAgbRs2TIngWzQoAGvv/4648aNo3bt2iFHKCJStKkmUkRKne+//57zzjuPL78MFtUyM3r37k3//v2pVq1ayNGJiBQPSiJFpNSpWLEi3333HQD77rsvY8aM0XrXIiJxUhIpIqXOnnvuycCBA1m0aBG33norlSpVCjskEZFiR0mkiJRoa9as4ZZbbmGPPfbg6quvzinv3bt3iFGJiBR/SQ+sMbMGZjbCzH4wsxVmdlikvI6ZDTOzNsmHKSISv7feeosWLVowdOhQbr75ZubNmxd2SCIiJUZSSaSZpQEzgLOA+UANIrWb7r4cOAS4LMkYRUTi8tdff3HRRRdxzDHHsGDBAgDatWsXblAiIiVMsjWRA4GVwD7AeUDsemCvAocmeQ8RkQJ78cUXSUtL47HHHgNghx12YPjw4bz//vs0atQo5OhEREqOZPtEHgbc6e7LzCy3SdUWArsmeQ8RkW1asmQJl19+OZMmTcopO/bYYxk5ciR77LFHiJGJiJRMydZElgHW5bN/J2BjkvcQEdmmDz/8MCeBrFWrFuPHj2fKlClKIEVECkmyNZFfAScCj8TuMLNywNnAtCTvISKyTV26dOGUU06hfPnyDB8+nJ133jnskERESrRkayIHAMeZ2aNA80jZzmbWCXgTaArcm+jFzayimd1nZovMbL2ZTTezoxO4zltm5mY2PNFYRKToyMrK4uGHH+bzzz/PKTMznn32WZ577jklkCIi20FSNZHuPsXMugFDgYsjxU8SDLBZDXR19w+SuMU4oAvwIPAj0A14zcw6uvtHBbmAmZ0OHJhEDCJShPzwww90796djz76iObNm/Pll19SoUIFAE0aLiKyHSU9T6S7PwHsDnQGrgduAs4Ednf3ZxK9rpntT9AcfqO793X3UcCRwC8Eo8ILco1KwP3AfYnGISJFw+bNm7n33ntp1aoVH30U/A+5Zs0afvnll5AjExEpnZKqiTQz88Ba4KXUhJSjC5AJjMoucPcNZpYB9Dez3d39121c4zqCRHkwcGeK4xOR7WTGjBmkp6czY8YMIGi6vvzyy7nnnnvYYYcdQo5ORKR0SrYm8nczG2pmB6ckmq21Aea6++qY8s8iH1vnd7KZNQBuAK539/WpD09ECtuGDRu46aabaN++fU4C2aRJEz766COGDh2qBFJEJETJjs5+H7gIuMzMfgcmAhPd/bP8TyuQesDiXMqzy+pv4/z7gRnu/my8NzazugTTE0XTLMUi29mAAQMYMGAAAOXKleP666/nlltuUd9HEZEiINmBNeeYWWXgJIKlDy8BrjazBcAEgoTy6wQvX5nc55jcELU/V2bWkaCPZocE730pcFuC54pIilxzzTWMHTuWunXrkpGRQevWrcMOSUREIlIxsGa9uz/n7l2AugTLH34DXA18aWbfJ3jp9UDFXMorRe3/l8j8lMOAJ9z989yOKYBHCKYsit5OSfBaIlJAb7/9NqtX/9ODpXr16rz77rtMnz5dCaSISBGTdBIZzd3XRkZknwf0BdYAeyd4ucUETdqxsssW5XFeV2BfYKSZNczeIvuqRV5Xye/G7r7U3b+N3oB5CXwNIlIAK1asoFu3bhx99NFcf/31W+3be++9KVcu2Z43IiKSailLIs2sipmdbWYvAEsJ5o78A+if4CW/BvYxs+ox5R2i9uemAVAe+BiYH7VBkGDOB45JMCYRSbFJkybRtGlTxo8fD8DTTz/N0qVLQ45KRES2Jakk0swqmVlnM5tIkDg+TTBqehiwn7vv4+63Jnj5SUBZ/pnEHDOrCFwITM+e3sfMGphZk6jzngVOy2UDeC3y+fQEYxKRFFm8eDGdO3fmjDPOyEkajz/+eGbPnk3dunVDjk5ERLYl2TaiZUAVgqblUcAEd09Jgubu083sOWBAZLT0T8AFQEMgPerQx4HDCVbJwd2/B/7VD9PMAOa7+0upiE9EEuPujBs3jj59+rBy5UoAateuzdChQzn33HOzf1ZFRKSISzaJHEeQOBZoCcIEdAXuAs4HagGzgJOSXEpRREJ0xRVXMHz4P8vYn3XWWQwbNky1jyIixUxSzdnufnkhJpC4+4bIkof13L2Su+/v7m/EHHOEu2+z6sLdzd0vK6xYRaRgunbtSpkyZahXrx4vvfQSzz77rBJIEZFiKK6aSDM7DCC7JjD79bao5lCk9Fq4cCG77757TjN1+/btmTBhAp06daJmzZrhBiciIgmLtzl7KuBmVtndN2W/zud4i+wvm1B0IlJsbd68mYEDB3LnnXfyxBNPcOaZZ+bs69KlS4iRiYhIKsSbRHYEiCSQOa9FRKJ9+eWXpKenM3PmTACuvPJK/vOf/2i5QhGREiSuJNLd38/vtYiUbuvXr+eOO+5g8ODBZGZmApCWlsbYsWOVQIqIlDDJzhP5rpkdlc/+jmb2bjL3EJHi4YMPPqBVq1bcd999ZGZmUq5cOW677Ta++uorOnRIdBl7EREpqpKd4ucIYEw+++sSzOEoIiVY3759GTx4cM7r/fbbj7Fjx9KiRYsQoxIRkcKUimUP8xtY0xj4OwX3EJEirF69YEn7SpUqMXjwYD799FMlkCIiJVzcNZFmdgHByjHZbjGzHrkcWhNoSbDUoIiUIFu2bKFcuX9+fVx55ZUsWLCAK664gsaNG4cYmYiIbC+J1ERWAXaKbADVol5nb3WAjcAIoHvyYYpIUeDuTJw4kX322Yf58+fnlJctW5Zhw4YpgRQRKUXirol090eBRwHMbD5wpbu/nOrARKRoWbRoEZdeeimTJ08G4OKLL+bNN9/UWtciIqVUUgNr3H3PVAUiIkWTuzN27FiuueYaVq1aBUCdOnVIT08POTIREQlTvMseNgBw94XRr7cl+3gRKV5+/vlnevTowbvv/jNT17nnnsvQoUOpU6dOiJGJiEjY4q2JXMDWyx4uIP/R2dm07KFIMZKVlcXQoUO5+eabWb9+PQC77rorI0aM4KSTTgo5OhERKQriTSIvIkgaN8e8FpESxMx49913cxLInj17ct9991GjRo2QIxMRkaIi3mUPx+X3WkRKBjPj0UcfZfHixQwePJgjjjgi7JBERKSIScVk4/9iZhXMrGphXFtEUu/zzz/n3HPPZfPmzTllu+22G59//rkSSBERyVWya2efbWYPxJTdBqwBVprZi2a2QzL3EJHCs27dOq699loOOOAAnnnmGQYOHLjVfk3fIyIieUm2JvIaIKfG0cwOAm4D3gAeAI4Dbk7yHiJSCKZOnUrLli25//77ycrKonz58pQtqzFwIiJSMEnNEwk0AsZHvT4XWAKc5u5bzKwM0Bm4Mcn7iEiKrFq1iuuuu45Ro0bllHXo0IGMjAyaNWsWYmQiIlKcJFsTWRHYEPX6GGCKu2+JvJ4D7JbkPUQkRf73v//RrFmznASycuXKDBkyhI8//lgJpIiIxCXZJHI+0AnAzPYDGgOvR+3fmaB/pIiEbN26dfTs2ZPff/8dgCOPPJLZs2dz9dVXqxlbRETilmwSORI408xmAW8CvwH/i9p/MPBtkvcQkRSoUqUKjz76KDVq1GDMmDG8/fbb7LXXXmGHJSIixVSya2c/ZGYbgBOAL4H73H09gJntCOwCjEg6ShGJ22+//cbrr79O9+7dc8r+85//MH/+fGrVqhViZCIiUhIkO7AGdx8NjM6lfAWwX7LXF5H4ZGVlMWbMGPr27cvq1atp0qQJhxxySM5+JZAiIpIKKZts3MzSzOz4yJaWquuKSMH99NNPHHXUUfTs2ZPVq1cD8Prrr2/jLBERkfglXRNpZqcAQ4CGMeXzgT7u/nKy9xCR/G3ZsoUHH3yQW2+9lQ0bggkTdt99d0aOHMnxxx8fcnQiIlISJZVEmtkJwPPAL8BNwHeRXU2Bi4EXzOwkd1dViEgh+eabb0hPT+fzzz/PKbv00ksZMGAA1atXDzEyEREpyZKtibwVmAUc6u5ro8pfNrPhwEcEK9goiRQpBEuWLKF9+/Zs3LgRgL333puMjAwOPfTQkCMTEZGSLtk+kS2B8TEJJACRsnGRY0SkEOyyyy5ceumllC1blhtuuIGZM2cqgRQRke0i2ZrIDcCO+ezfka1XtBGRJKxdu5a5c+fSpk2bnLK77rqL8847j7Zt24YYmYiIlDbJ1kS+C1xpZgfG7jCzDsAVwNtJ3kNEgHfeeYcWLVpw/PHHs2LFipzyqlWrKoEUEZHtLtkk8jqCmsaPzOxTMxsX2T4FPonsuz7ZIEVKs5UrV9KjRw86derE/Pnz+eOPP3jkkUfCDktEREq5pJJId59P0OdxGFALOCuy1QKGAq3cfUGSMYqUWpMnTyYtLY0xY8YAQa3jsGHDuPHGG0OOTERESruE+0SaWVlgJ2Clu18NXJ2yqERKuaVLl3LFFVcwYcKEnLKjjz6aUaNG0bBhw/ACExERiYi7JtIC/YG/gN+B1Wb2YmStbBFJ0rfffkvTpk1zEsiaNWvy2GOP8cYbbyiBFBGRIiOR5uxuwA3ASoKJxr8BTgEeS1lUIqXYvvvuy5577gnA6aefzpw5c+jWrRtmFnJkIiIi/0ikOfsSYAZwiLuvBzCzoUBvM6vj7stTGaBISZeVlcXff/9NjRo1AChXrhxjx47lxx9/pHPnziFHJyIikrtEaiIbAY9nJ5ARj0SutXdKohIpJebOnUvHjh0588wzcfec8pYtWyqBFBGRIi2RJLIWsCymLLv2sVJy4YiUDlu2bGHgwIG0atWKDz74gDfffJNnnnkm7LBEREQKLNHR2b7tQ0QkNzNnzuSiiy7iq6++AsDMuOyyyzj55JNDjkxERKTgEk0i7zWz6InqykY+jjGz2HW03d1bJXgfkRJjw4YN3H333dx3331s2bIFCAbRZGRkcPDBB4ccnYiISHwSSSI/IPeayKVJxiJSYs2aNYuzzjqL77//HoCyZcty/fXXc+utt1KpknqBiIhI8RN3EunuRxRCHCIlWs2aNfntt98AaNOmDWPHjqV169bhBiUiIpKEhFesEZH8uXvO3I4NGjTggQceYPny5VxzzTWUL18+5OhERESSoyRSJMX++usv+vTpQ5MmTbj++utzyrt37x5iVCIiIqmlJFIkhV544QV69+7NkiVLqFixIqeeeir77rtv2GGJiIikXCLzRIpIjCVLltClSxc6d+7MkiVLADjiiCOoUqVKyJGJiIgUDiWRIklwd8aPH09aWhrPP/88ALVq1WL8+PFMmTKF3XffPeQIRURECoeas0US9Msvv9CzZ0/eeOONnLIzzjiDhx56iJ133jnEyERERApfSpJIM9sVOAyoCzzv7r+ZWVmgBrDK3TNTcR+RouSrr77KSSB32WUXHnnkEU477bSQoxIREdk+kkoiLZi/5H7gssi1HPgG+A3YAVgA9AMeTOY+IkXRaaedxhlnnEG1atUYPHgwtWrVCjskERGR7SbZPpF9gSuBwcDRgGXvcPdVwAtA5yTvIRK6zZs3c++99/LJJ59sVf7000+TkZGhBFJEREqdZJuzewCPu/tNZlY7l/2zgOOTvIdIqGbMmEF6ejozZsygSZMmzJgxI2epwnLl1K1YRERKp2RrIncHPsln/1qgepL3EAnFhg0buOmmm2jfvj0zZswAoEyZMixevDjkyERERMKXbBK5lCCRzEs7YGGS9xDZ7j7++GNat27NgAEDyMzMpFy5ctx666189dVX7LnnnmGHJyIiErpkk8gXgF5mtldUmQOY2TFAN+C5JO8hst38/fffXH755Rx66KH88MMPALRr144vv/ySO++8k4oVK4YcoYiISNGQbBJ5G7AY+Bp4nCCBvN7MPgKmEPSJ7J/kPUS2m0GDBjF8+HDcnUqVKjFw4ECmTZtGy5Ytww5NRESkSEkqiYyMwD4AGAjsCmwADgdqAncAh7r7uiRjFNlurrvuOvbYYw8OO+wwZs2aRd++fTV4RkREJBdJ/3V09/XA3ZFNpNhwd1588UU6duyYM0XPDjvswIcffsiuu+5KmTJaFVRERCQvRfqvpJlVNLP7zGyRma03s+lmdnQBzjvdzCaY2c9mts7MfjCz+82s5nYIW4qBxYsX07lzZzp37sy111671b7dd99dCaSIiMg2JLtizdgCHObunp7gLcYBXQhWvPmRYKDOa2bW0d0/yue8UcAi4EmC0eEtCFbVOcHM2kZqT6UUcnfGjRtHnz59WLlyJQCTJ0/m3nvvZaeddgo3OBERkWIk2ebsI4mMxo5SFqgX+biMYK7IuJnZ/sDZQF93HxwpexyYTdAH86B8Tu/i7lNjrvclMB74LzAmkZikeJs/fz4XX3wxb7/9dk7Z2WefzdChQ5VAioiIxCnZgTUN3X3PmK0BUAW4AvgbOCrBy3cBMglqFbPvtwHIAA40szznp4xNICNejHxsmmA8UkxlZmYybNgwmjdvnpNA1q9fn8mTJ/PMM89Qt27dkCMUEREpfgpl2Km7bwaGm1kaMBw4MYHLtAHmuvvqmPLPIh9bA7/Gcb1dIh+Xb+tAM6sLxFZNNYrjXlKEXHLJJYwePTrndY8ePRg4cCA1a9YMLygREZFirrBHD8wEDkvw3HoEc1DGyi6rH+f1rieo2ZxUgGMvJWg2j94mx3k/KSIuueQSypYty1577cU777zDqFGjlECKiIgkqbAnwDsaSHSeyMrAxlzKN0TtLxAzOxdIBwa6+48FOOUR/r3STiOUSBYLP/zwA/vssw9mBkCbNm14+eWXOfzww6latWrI0YmIiJQMyY7O7pfHrpoENZBtgXsTvPx6ILc15ipF7d8mMzuUoB/lG8DNBTnH3ZcSrAsefZ2CnCohWr9+PbfffjuDBw9m/PjxnHfeeTn7TjjhhBAjExERKXmSrYm8PY/yv4B5QC9gdB7HbMtiglVwYtWLfFy0rQuYWSvgZYLm6C7uviXBWKSI++CDD+jevTs//hhUNPft25czzjhDa12LiIgUkqSSSHcvzD6VXwMdzax6zOCaDlH782RmjYDXCWoUT3D3NYURpIRr9erV3HDDDTz66KM5Ze3btycjI0MJpIiISCFKOAk0s8pmNsTMTk5lQFEmEcw1eXHUPSsCFwLT3f3XSFkDM2sSE9suwJtAFnCsuy8rpBglRK+99hrNmzfPSSArV67M4MGD+fTTT2nRokXI0YmIiJRsCddEuvt6M+sJzElhPNHXn25mzwEDIlPu/ARcADQkGCST7XHgcCC60+LrwF4Ek5IfYmaHRO37w93fKoyYZfu55JJLGDFiRM7rI444gtGjR9O4ceMQoxIRESk9ku0T+SXQPBWB5KErcBdwPlALmAWc5O4fbOO8VpGP1+Wy731ASWQxl5aWBkD16tUZPHgw3bt31+AnERGR7SjZJPIqgrWsZwPjUj1wJbJCTd/IltcxR+RSpmyihNm4ceNWfRx79+7N4sWL6d27N7vumtv4KxERESlMcfeJNLPDzCx7NZfxBP0ORwKrzexHM5sVs81MZcBSurg7Y8aMoWHDhjkjrwHKlClD//79lUCKiIiEJJGBNe8BnSKf/wn8AHwATAd+i5RFbyuSD1NKo59//plOnTrRo0cPlixZQo8ePXD3sMMSEREREmvOtsiWa1OySLIyMzMZNmwYN998M+vXB3PK77bbbvTt21f9HkVERIqIwl72UCQus2fPpnv37kyfPj2nrFevXtx3331Ur149xMhEREQkWqLzRKpNUVIqKyuLO+64g7Zt2+YkkI0bN2bq1Kk8+uijSiBFRESKmESTyCfNLLOAm5YalG0qU6YMs2bNYvPmzZQpU4a+ffsyc+ZMDj/88LBDExERkVwk2pz9NjA3lYGIDB8+nOXLlzN48GDat28fdjgiIiKSj0STyPHu/nRKI5FSZerUqTzwwANMnDgxZ/7HevXq8f7774ccmYiIiBREwmtniyRi1apV9OzZk44dO/Lyyy8zYMCAsEMSERGRBGh0tmw3//vf/+jVqxe///47AFWqVKFOnTohRyUiIiKJUBIphW7ZsmVceeWVPPPMMzllRx11FKNGjWKvvfYKMTIRERFJVNxJpLurCVwKxN159tlnueKKK1i+fDkANWrUYMiQIVx44YWaOFxERKQYU02kFJr169dz/fXX5ySQp5xyCo888gj169cPOTIRERFJlmoVpdBUqVKFUaNGUbduXSZOnMiLL76oBFJERKSEUE2kpMyPP/7Ia6+9xpVXXplTdtxxx/Hzzz9TtWrVECMTERGRVFNNpCRty5YtDB48mJYtW3LVVVfx7rvvbrVfCaSIiEjJoyRSkjJr1iwOPPBA+vbty4YNGwD49NNPQ45KRERECpuasyUhGzdu5J577mHAgAFs2RIsj7733nuTkZHBoYceGnJ0IiIiUtiURErcpk2bRnp6OnPmzAGgbNmy9O3bl379+lG5cuWQoxMREZHtQUmkxGXRokUcdthhbN68GYBWrVqRkZFBu3btQo5MREREtif1iZS41K9fn6uvvpoKFSpwzz338PnnnyuBFBERKYWUREq+Vq5cybRp07Yqu/3225k5cyY33XQT5cuXDykyERERCZOSSMnT5MmTSUtL4+STT2bZsmU55ZUrV6ZJkyYhRiYiIiJhUxIp//LHH39w1llnceqpp7J48WKWL1/O2LFjww5LREREihANrJEc7s5TTz3FlVdeyYoVKwCoWbMmDzzwABdccEHI0YmIiEhRoiRSAFi4cCG9evViypQpOWWdO3dm+PDh7LLLLiFGJiIiIkWRkkhh1qxZHHzwwaxZswaAnXfemYcffpjOnTuHHJmIiIgUVeoTKTRr1ozmzZsD0K1bN+bMmaMEUkRERPKlmshSaMuWLaxcuZI6deoAwYozY8eO5ddff+WYY44JOToREREpDlQTWcp8/fXXdOjQgTPPPBN3zylv2rSpEkgREREpMCWRpcSGDRu4+eab2W+//fjqq6947733ePrpp8MOS0RERIopNWeXAp988gnp6el8//33QNB8fcMNN6jfo4iIiCRMSWQJtmbNGm666SaGDx+e03Tdpk0bxo4dS+vWrcMNTkRERIo1JZEl1BdffEGXLl345ZdfAKhYsSJ33HEH11xzDeXK6bGLiIhIcpRNlFD169dn5cqVABxyyCGMGTOGfffdN9ygREREpMRQElmCZGVlUaZMMFaqfv36DBs2jDVr1tCrV6+cchEREZFUUBJZAixZsoTLLruMli1b0q9fv5zyrl27hhiViIiIlGSqnirG3J3x48eTlpbG888/z9133823334bdlgiIiJSCqgmsphasGABPXv25M0338wpO/XUU3NWoREREREpTKqJLGaysrJ46KGHaN68eU4Cucsuu/DCCy8wceJEdt5555AjFBERkdJANZHFyPfff0/37t35+OOPc8rS09MZNGgQtWrVCjEyERERKW2URBYj8+bNy0kgGzZsyOjRo+nUqVPIUYmIiEhppCSyGDnxxBM5//zzqV27NnfffTdVq1YNOyQREREppZREFlHr16/nzjvv5LjjjuPwww/PKR83bpzmfBQREZHQKRspgj788ENat27NvffeS/fu3Vm/fn3OPiWQIiIiUhQoIylC/v77b3r37s1hhx3G3LlzAahRowbLli0LOTIRERGRrSmJLCKmTJlCs2bNeOSRRwCoVKkSAwcOZNq0aTRo0CDk6ERERES2pj6RIfvzzz+5+uqreeKJJ3LKDjvsMMaMGcPee+8dYmQiIiIieVNNZMgefPDBnASyWrVqjBgxgvfee08JpIiIiBRpSiJDduONN9KoUSNOPPFE5syZQ8+ePTV4RkRERIo8NWdvR+7O448/zkknnUTt2rUBqFKlCp988gk77bQTZhZyhCIiIiIFoyqv7WT+/Pkcc8wxdOvWjauvvnqrfXXr1lUCKSIiIsWKkshClpmZydChQ2nevDlvv/02AO+88w5//vlnyJGJiIiIJE5JZCGaM2cOhx56KFdddRXr1q0DoEePHsyZMyenOVtERESkOFISWQg2b97M3XffTZs2bfj0008B2GuvvXjnnXcYNWoUNWrUCDlCERERkeRoYE0huPjiixk3bhwQLFN41VVXcdddd1GlSpVwAxMRERFJEdVEFoJrrrmG8uXL06xZMz755BPuv/9+JZAiIiJSoiiJTIEZM2bg7jmvmzdvzptvvslXX31Fhw4dQoxMREREpHAU6STSzCqa2X1mtsjM1pvZdDM7uoDn7mpmE81spZmtNrPJZrZXKuNbvXo1l156KW3btuWxxx7bat8RRxxBhQoVUnk7ERERkSKjSCeRwDigD/AUcCWQCbxmZofkd5KZ7QC8BxwO9AduA9oA75tZSoZFv/baazRr1oxHH30UgFtvvZWNGzem4tIiIiIiRV6RTSLNbH/gbOBGd+/r7qOAI4FfgIHbOP1SYG/gJHcf6O4PAMcA9YBrkolr5cqVnHfeeZx44on89ttvAHTs2JEPPviAihUrJnNpERERkWKjyCaRQBeCmsdR2QXuvgHIAA40s923ce7n7v551LnfA+8AZyYTVNeuXXnqqacAqF69OqNHj+add96hUaNGyVxWREREpFgpyklkG2Cuu6+OKf8s8rF1bieZWRmgJfBFLrs/AxqZWbVEg1q5ciUAJ598MnPmzKF79+5aslBERERKnaI8T2Q9YHEu5dll9fM4b0egYgHO/SGvG5tZXWCnmOImANWqVaNfv34cd9xxrFy5MiepFBERESkqfvrpp+xPC22Ub1FOIisDuY1U2RC1P6/zSPDcbJcSDMb5l7///pu+ffvSt2/fbVxCREREJHTNgRmFceGinESuJ6hRjFUpan9e55HgudkeAZ6LKWsBPEPQ3/L7bZwvRUsjYDJwCjAv5Fgkfnp+xZueX/GlZ1e8NQEmAXML6wZFOYlcDOyaS3m9yMdFeZy3gqAWsl4u+7Z1LgDuvhRYGl0W1e/xe3f/Nr/zpWiJenbz9OyKHz2/4k3Pr/jSsyveop7fmsK6R1EeWPM1sI+ZVY8p7xC1/1/cPQv4Btgvl90dgJ/d/e8UxSgiIiJSKhXlJHISUBa4OLvAzCoCFwLT3f3XSFkDM2uSy7ntzWy/qHP3JZhnMraZWkRERETiVGSbs919upk9BwyIjJb+CbgAaAikRx36OMHKNNHz7DwC9ABeNbPBwGaClW/+AO4v/OhFRERESrYim0RGdAXuAs4HagGzCFah+SC/k9z9bzM7AngAuIWgxnUqcLW7L0swlmXAHZGPUrzo2RVven7Fm55f8aVnV7wV+vMzdy+sa4uIiIhICVWU+0SKiIiISBGlJFJERERE4qYkUkRERETipiRSREREROKmJFJERERE4lbqk0gzq2hm95nZIjNbb2bTzezoAp67q5lNNLOVZrbazCab2V6FHbMEEn12Zna6mU0ws5/NbJ2Z/WBm95tZze0QtkQk87MXc523zMzNbHhhxCm5S/b5mdlZZvapma2N/A79xMyOLMyYJZDk371OZvaemS2PPLfPzOz8wo5ZAma2g5ndYWavm9mKyO++bnGcX9PMRpnZssjP3ntm1jbReEp9EgmMI5iI/CngSiATeM3MDsnvJDPbAXiPYKLz/sBtQBvgfTOrXZgBS45xJPDsgFFAU+BJ4ArgdeAy4FMzq1xo0UqscST2/HKY2enAgYUSnWzLOBJ8fmZ2O/AM8GvkGrcQzAO8ayHFKlsbR2J/9/4DvAlUAG4HbgbWA4+b2dWFGK/8ow7Qj+Bv2Mx4TjSzMsCrwLnAcOA6oC4w1cz2Tigady+1G7A/4MC1UWWVCFbH+WQb514XObd9VFkTYAvQP+yvraRvST67I3Ip6xq5Xvewv7bSsCXz/GKOnw/cGrnW8LC/rtKyJfnzdwCQRbD4Q+hfS2nbknx2bwK/AxWjyspFzp0Z9tdWGjagIrBL5PP9Is+yWwHPPTNyfJeosp2Av4CnE4mntNdEdiH4D2xUdoG7bwAygAPNbPdtnPu5u38ede73wDsED0oKV8LPzt2n5lL8YuRj0xTGKHlL5mcv23UErSmDCyVCyU8yz+8qYAkw1AI7FGag8i/JPLvqwF/uvjHq3C3AcoIaSSlk7r7R3ZckeHoXguWfX4i63jJgInCKmVWM94KlPYlsA8x199Ux5Z9FPrbO7aRIlXBL4Itcdn8GNDKzaqkKUnKV0LPLxy6Rj8uTCUoKLKnnZ2YNgBuA691df7y2v2Se31HA5wRdSZYBf5vZYjO7LOVRSm6SeXZTgWZmdpeZNTazRmZ2K0GN2MCURyqp1gb4yt2zYso/A6oA+8R7waK+dnZhqwcszqU8u6x+HuftSFClvK1zf0gqOslPos8uL9cT/Hc+KZmgpMCSfX73AzPc/dmURiUFldDzM7NaBH26DgaOJFjXdyFwIfCQmW1295GpD1eiJPOzdxewJ0FfyFsiZeuAzu4+OWURSmGpB3yQS3n0s/8mnguW9iSyMrAxl/INUfvzOo8Ez5XUSPTZ/YuZnQukAwPd/ccUxCbblvDzM7OOQGegQyHEJQWT6PPLbrquDZzt7hMAzGwSwR+vWwAlkYUrmd+dG4G5BP9svwCUBS4GnjSzo919WioDlZRL2d/NbKU9iVxPUKMYq1LU/rzOI8FzJTUSfXZbMbNDCfoCvUHw37VsHwk9PzMrBwwDnojujyzbXbK/OzcTVevv7llmNgG4w8wauPvClEUqsZL53TmcYGBU2+wmUTObCHwLDEX/2BV1Kfm7Ga2094lcTFC9Gyu7bFEe560gyOYTOVdSI9Fnl8PMWgEvA7MJRqttSV14sg2JPr+uwL7ASDNrmL1F9lWLvK6S2lAlF8n87twA/OnumTH7lkY+1ko+PMlHQs/OzCoQtNi8Gt2nzt03A1OA/SLHSNGV9N/NWKU9ifwa2MfMqseUd4ja/y+RH6BvCDoTx+oA/Ozuf6coRsnd1yTw7LKZWSOC+SGXAie4+5pUByj5+prEnl8DoDzwMcH0PtkbBAnmfOCYVAYqufqaxH93fg3slEvCkd0Xb1lqQpQ8fE1iP3u1CVovy+ayrzxBPpHbPik6vgbaRgYHR+tA0Ld1brwXLO1J5CT+6dMBBDP5E3Tynu7uv0bKGphZk1zObW9m+0Wduy9BZ/HnCjtwSfzZmdkuBPOdZQHHRqY4kO0r0ef3LHBaLhvAa5HPpxd69JLM784JkXMviDq3EvBfYI67qxWncCX67JYCK4HTov8BiEzRdDLwvWZKKDrMrJ6ZNTGz8lHFk4CdgdOjjqsDnAG8Ej11U4HvE5lsstSK9Oc4DXiAYMLUCwgmYz3K3T+IHDMVONzdLeq8asAMoBrBPHWbCVYAKAu0VmJS+JJ4dl8DrQimpIgdifaHu79V6MFLws8vj2s58LC7a5qY7SSJn7/KBFP87EPQj24hcD7QFjjZ3adsxy+jVEri2d0M3E3wt+9xgr936QTz657n7k9txy+j1IpMh1WToPb+EoJBTjMiux9y91VmNo7gue7p7gsi55UFPgKaA4MIprS7lKCFp727xz+jTFizrheVjaBD6SCCvgIbCOZLOjbmmKnBW/Wvc3cjqHVcBfwNvAI0DvtrKi1bos+OYMb+vLapYX9dpWVL5mcvl2tpxZpi9PwIllobB/wZOXda7LnaiuyzO5egtv8vgibQaQRT/IT+dZWWDViQz9+whpFjxkW/jjq3FjCGIIFcG3nO+yUaS6mviRQRERGR+JX2PpEiIiIikgAlkSIiIiISNyWRIiIiIhI3JZEiIiIiEjclkSIiIiISNyWRIiIiIhI3JZEiIiIiEjclkSIiIiISNyWRIiIiIhI3JZEiIiIiEjclkSJFkJkdYWZuZkeEHUthinyNtxfw2AVmNq5wIyoZzOwRM3sr7DiKKjObamZTY8p2NrNJZvZn5PvyqkR/Ds3sdjNL6ZrCZnavmU1P5TVFkqUkUiSFzKxb5I9Obtu9YceXn1xi32Bmc81suJntvJ1iOCjyB7jm9rhfQUSS1+j3Za2ZfWZmXZO45gkFTZ4TuPaeQHegf0z5JWb2nJktjHwd4wrj/pF77WRmQ83sezNbb2ZLI+/ZfWa2Q2HdN0kPAMcCA4DzgddTeXEzu8nMTk3iEg8CrczsP6mJSCR55cIOQKSE6gfMjymbHUYgCciOvRJwCHAJcIKZNXf3dSm+V2VgS9Trg4DbgHHAyphj9wWyUnz/gvoauD/yeT2CJG28mVV099EJXO8EoDdwe0qi29qVwHx3fy+m/HqgGvAZwddQKMxsR+ALoDowFvgeqA20JPheehRYU1j3L6Bjcik7Epjs7oOzC8xsLsH36KY4r383EPtP403AJOClOK8FgLsvMbPJwLXAy4lcQyTVlESKFI4p7v5F2EEkKDr2MWb2J9AHOAV4JpU3cvcNcRy7MZX3jtPv7v5k9otILd7PwNVAIklkoTCz8sB/gRG57D4cWOjubmaFmcSlAw2Ag939k5j4qhN/QpZy7p5bDHWJ+cfF3bOAAn+PRp23ha3/OUqVicBzZraXu/9cCNcXiYuas0W2IzPbI9Jf7YdIM9+fkSbGhgU4d28ze97MlkSamn8zs2fNrEbMceeZ2ZeR66+IHLN7EmG/G/m4Z+T65czsVjObZ2YbI829/c2sYkwc+5nZG2a2PBLLfDMbG3NMTp/IyMdBkV3zo5qPG0b25/SJjFzbzeyCXN6nYyP7Tooq29XMxprZH5GYvzWzixJ9Q9x9GUENW6OYex8a1WS80cx+NbMHzKxy1DHjCGohifoaPWp/GQv6430bec5/mNlIM6tVgNAOAeoAb+cS8y/untJ+enloBGQC03KJYXX0Pw4W9E2cbWbtzOyTqO+TXrHnmllFM7vDzH6Kem8Hxn7fRY49L9J8vs7M/jKzD8zsmKj9OX0iLdKNAzCgd/TzsDz6RJpZBzN7LXLttWY2y8yujNq/VZ/IyOdVgQuinvk4M+sY+fy0XL6GcyP7Dowqzn6up8QeLxIG1USKFI4aZlYnusDdlwPtCZpsnwV+AxoSNPFNNbO0vJqLzawC8AZQEXgIWALsCpwE1ARWRY67GbiLoMZiDLATcDnwgZm1cfeVCXwt2YnSn5GPY4ALCJrm7gc6ADcCTYHTInHUBd4ElhE0662MfK2n53OfF4B9gHMIaviWR8qXxR7o7l+Y2c/AmcD4mN1nAX8RvF9Y0J9zGuDA8Mj1jgcyzKy6uz+Y71efCzMrB+wWuU+0M4AqBE22fwL7E7z/u0X2AYwE6gNHE/S9izUS6AY8BgwjSN4vA9qY2cHuvjmf0A4i+DpnxPs1pdAvQFmCry322eSmFvAawffsMwTP9FEz2+TuYyFIrAmacA8BRgHfAS0Ivk/2AU7NvpiZ3UbQTeATgq4Zmwi+R48k+J6M9UEk1ieAt4DH8wvWzI4G/gcsBoYS/Cw2JfhZHJrHaecT/Nx8FokfYB7B9+WvBLXHL8ac819gnrt/ml3g7qvMbB5wMEEfTpFwubs2bdpStBH88ffctsj+yrmcc0DkmPOjyo6IlB0Red068rpLPvfeg6AJ7aaY8ubA5tjyfGI/iqA2azeChGw5sI4gaW0VOWZ0zLmDIuUdI69Pjbzebxv3dOD2qNfXRsoa5nLsAmBc1Ov+BAlCraiyCgSJXUZU2RhgEVA75nrPECS3/3omudz3jch7Uifyfj4eiXN4zLG5Pd8bCPpyNogqG579PRFz7CGR654bU35sbuW5nP8EsLwA36drot/LFP8M7AwsjcT7HUFCfQ5QI5djp0aO6xPzDGcAfwDlI2XnEdRuHhJzfs/I+QdFXjeOHPcCUCbmWIu579Rcvhdjn+cRbP1zWJagG8MCoGY+17899vnm9Z5Hvo83RL8/BP/8bY7+2Yja9wYwpzCenTZt8W5qzhYpHL0JapqiN9x9ffYBZlbezGoDPxEkM23zud6qyMdjzaxKHsecTtBFZaKZ1cneCGpKfgQ6FjD2twlq634lqDFdA5zm7r8TDAgBGBJzTvagkxMjH1dGPp5kQT+9wjABKM/WtZvHENTMTgAwMwM6A69EXka/L28ANcj/fY++7rLI9g1BzdJjQN/og2Keb9XIfT4haCptU4D7nEHwrN+KifVLguewrWdYm3/Xjm5X7v4HwT8bIwhqGXsBTwNLLegGYTGnbCGofc0+f1PkdV2gXaT4DIKE9PuY9yW7q0X2+3Iqwc/AnR70Z4yOKxVN+W0IaoYf9Jha/SSu/zhBC0OXqLKzCFoKn8zl+L8I/pkRCZ2as0UKx2eey8CaSN+4G4ELCWr2ov+g1og9Ppu7zzezIQQDXP5rZh8SNO896e7ZCebekev9mMdl8msGjdYbmEvwx/0P4IeoP8h7ENSq/RQT3xIzWxnZD/A+8DzBSOurI/3PXgKe9hQNkHH3mWb2PcEf3IxIcXbNaXZysRNBUnlxZMtN3QLcbjpwC0FNVPPI57WIGSRiZg2AO4H/RPZHy/P5Rtk7ctzSJGKNTdKSYmZlCd7HaCs898EpALj7YuASM7uU4Gs6lmB0+J0EzcBjog5f5O5rYy4xN/KxIUGT794ETcb/6toQkf2+NCL4/pyTz5eUjOyuHSmbacHdvzezzwmar7O/j/8LTHP3n3I5xQhqR0VCpyRSZPt6iCCBfBD4lKDWyQlq/PJtGXD3ayKDMk4hqBkbBtxoZge4+2+R852gv19mLpco6IjcXBPg2HC2EasDXczsAOBkgiRiLHBNJN5UjQ6eANwcqZX6myB5e8aD0bHwz3v6JHn3z5tVgPssd/fsQQ1vRJLX/xFMpzMEcpKtt4AdgfsIBt6sJfhnYRwFG8hYhiCB/G8e+/NKorL9yb+T12Ttzr+nq+pI0CScr8j3wVxgrpm9SvAPzn/ZOoksiDIENcB98tj/a5zXK2oeB4aa2W4EtZIHEPSDzU0t/ukvLBIqJZEi21cXYLy7X5NdYGaVCGrLtsndvyH4Y3q3mR0EfEzQXHgLQUd9I5gjcG7eV0nKLwR/0PcmaF4Ecgav1Izsj453GkFN0s1mdi7wFHA2eScR8dawTCCo7exMUGtanSAhz7aMILksG5UEJs3dXzWz94GbzGxkpCatBcEgjwvcPWdwRmQgxr8ukcel5wGdgI+jm8bj8D1BTXWNqBrqZC0h0h0jysx4L+LuP5vZX/x7jsr6ZlY1pjZyn8jHBZGP8wiayN/ZRrPxPILvzzSCuT1TbV7kY3NyGQG/DfnF/SzBPyPnEMxLuZlIl4xc7EkC779IYVCfSJHtK5N/NzdeTtBMmiczqx4ZERztG4Kmu+wpTl6IXP+22H5nFqidcNT/eC3y8aqY8uwaolcj96uVS9+3ryMf/zUlS5TsRKJmQYJx9+8I3oezIttigtG22fszCZrVO5tZ89jzzSy2mTYe9xH0QewReZ1d+5vzdUfegyv5t7WR/TVjyicSfC/cmkus5XI5Ptankfu328ZxBebuG9z97Zgtz36XkelvquZSvj/B+/VDzK5yBANkso+rEHm9jKAvKATvy678815HX7dy1P1eIviZ6BcZ0R19XCqa+b8iqJW9KvZZFOD6a8nj+9qDmRumEAwg+i/weqRsKxZM59WIoJ+tSOhUEymyff0PON/MVhH02zqQoObpz3zPCqYnGW5mzxE0D5YjGNyRnSTh7vPM7BaCZdsamtlLBLVwexJMvTMKGPzvSxdcpB/ieODiyB/R9wmmsbkAeMn/WSXlAuBSM3uRoPamGkECsJp/EtHcZCcN95jZswQ1Mq/k0mcu2gSCvnYbCEZlx65qcwNB8+t0MxtN8L7vSDCgplPk87i5+xQzmw30MbOHCWoB5wGDzWxXgq+1M7k3L2d/ncPM7A0g092fdff3zWwkQTeF1gRT0mwmqPk9gyAhnZRPWB8RfC914p9+oQCY2ckEtXkQDEhqGfl+AXjZ3QvSrF8Q5xPUhr5I8HVuIujPeBHBM+ofc/wi4HoL5gOdS/DPQGvgYv9nOqMnCKb+GWFmHQlq4MsCTSLlxwJfuPtPZnYPQRL+oZm9AGwkmFprEUF/5IS5e5aZXUIwUOtrM3uM4B+XJkCzSBx5+RLoZGZ9IrHMd/fotbAf559n+69/IiI6EfyTMDnxr0IkhcIeHq5NW0na+GeanFyntiGoiRjLP82srxMs57eAraevOYKtpxbZk6DT/U/AeoJE4V3gqFzucTrwIUEfyDUEzc7DgX2SiT3quHIE8+/9TJAgLCRIDCpGHdOGYETuLwSJwx8Ef3jbxVxrqyl+ImW3EMyhmUnUdD+x71HU8Y35Zyqlg/OIuW7kPVgYiXkxQXNkjwI80wXA//LYd0Hkvt0ir5sS9Iv8O/KMRxEs95dzTOS4sgR9WpcS1Jx5zHV7ECwduI4gGZ1FUPNZrwDxDgV+zKV8XNT7FLt129Z14/gZaAEMJEia/iRIghcR1Ca2iTl2KsEglXYEtWvrI+9371yuWx64LnL8BmBF5D3qB1SPOfZCglrD7OOmAp1i7js1l+/FfKf4iSo/mCDBX03wMzYTuCxq/+25PNN9Cf7pWhe55riY/RUisa4EKuXx3j4LfJiqZ6VNW7KbuWuQl4hISWFmexHUih7v7u+EHU9+IqP267j7v7oalDaR7iqLCGre03PZvwtBU/rZ7q6aSCkS1CdSRKQE8WBN5QyCZnwpPk4lmEoprxVzrgK+UQIpRYlqIkVEJBSqiQwGIhF0ebiVYCqpgkx+L1IkqCZSREQkPJcQLA25FOgaciwicVFNpIiIiIjETTWRIiIiIhI3JZEiIiIiEjclkSIiIiISNyWRIiIiIhI3JZEiIiIiEjclkSIiIiISNyWRIiIiIhI3JZEiIiIiEjclkSIiIiISNyWRIiIiIhK3/wOgeA8AiExSqgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 720x480 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fpr, tpr, thresholds = roc_curve(extset_df.Label.apply(lambda x: 'Healthy' not in x), extset_df['p(any_retina_disease)'])\n",
    "\n",
    "plt.figure(dpi=120)\n",
    "plt.plot(fpr, tpr, lw=1.3, c='black')\n",
    "plt.xlim(0,1)\n",
    "plt.ylim(0,1)\n",
    "plt.xlabel('False Positive Rate (1 - Specificity)')\n",
    "plt.ylabel('True Positive Rate (Sensitivity)')\n",
    "plt.plot([0, 1], [0, 1], linestyle='--', c='black')\n",
    "plt.xlim(0.,1)\n",
    "plt.ylim(0.,1)\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}