notebooks/poisoning_attack_sleeper_agent.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "21654757",
   "metadata": {},
   "source": [
    "# Import packages and path"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "ee35e7a1",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os, sys\n",
    "\n",
    "module_path = os.path.abspath(os.path.join('..'))\n",
    "if module_path not in sys.path:\n",
    "    sys.path.append(module_path)\n",
    "module_path = os.path.abspath(os.path.join('.'))\n",
    "if module_path not in sys.path:\n",
    "    sys.path.append(module_path)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "46009e26",
   "metadata": {},
   "source": [
    "# Load data and Model Architecture"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "07a5f2a9",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-09-20 12:11:43.960216: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.10.1\n",
      "2022-09-20 12:11:50.405316: I tensorflow/compiler/jit/xla_cpu_device.cc:41] Not creating XLA devices, tf_xla_enable_xla_devices not set\n",
      "2022-09-20 12:11:50.407916: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcuda.so.1\n",
      "2022-09-20 12:11:50.431217: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1720] Found device 0 with properties: \n",
      "pciBusID: 0000:af:00.0 name: Tesla V100-PCIE-32GB computeCapability: 7.0\n",
      "coreClock: 1.38GHz coreCount: 80 deviceMemorySize: 31.75GiB deviceMemoryBandwidth: 836.37GiB/s\n",
      "2022-09-20 12:11:50.431938: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1720] Found device 1 with properties: \n",
      "pciBusID: 0000:d8:00.0 name: Tesla V100-PCIE-32GB computeCapability: 7.0\n",
      "coreClock: 1.38GHz coreCount: 80 deviceMemorySize: 31.75GiB deviceMemoryBandwidth: 836.37GiB/s\n",
      "2022-09-20 12:11:50.431965: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.10.1\n",
      "2022-09-20 12:11:50.434444: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcublas.so.10\n",
      "2022-09-20 12:11:50.434501: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcublasLt.so.10\n",
      "2022-09-20 12:11:50.437132: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcufft.so.10\n",
      "2022-09-20 12:11:50.437504: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcurand.so.10\n",
      "2022-09-20 12:11:50.440221: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcusolver.so.10\n",
      "2022-09-20 12:11:50.441526: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcusparse.so.10\n",
      "2022-09-20 12:11:50.447095: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudnn.so.7\n",
      "2022-09-20 12:11:50.450311: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1862] Adding visible gpu devices: 0, 1\n",
      "2022-09-20 12:11:50.451039: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  SSE4.1 SSE4.2 AVX AVX2 AVX512F FMA\n",
      "To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.\n",
      "2022-09-20 12:11:51.069875: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1720] Found device 0 with properties: \n",
      "pciBusID: 0000:af:00.0 name: Tesla V100-PCIE-32GB computeCapability: 7.0\n",
      "coreClock: 1.38GHz coreCount: 80 deviceMemorySize: 31.75GiB deviceMemoryBandwidth: 836.37GiB/s\n",
      "2022-09-20 12:11:51.070624: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1720] Found device 1 with properties: \n",
      "pciBusID: 0000:d8:00.0 name: Tesla V100-PCIE-32GB computeCapability: 7.0\n",
      "coreClock: 1.38GHz coreCount: 80 deviceMemorySize: 31.75GiB deviceMemoryBandwidth: 836.37GiB/s\n",
      "2022-09-20 12:11:51.070660: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.10.1\n",
      "2022-09-20 12:11:51.070708: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcublas.so.10\n",
      "2022-09-20 12:11:51.070723: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcublasLt.so.10\n",
      "2022-09-20 12:11:51.070738: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcufft.so.10\n",
      "2022-09-20 12:11:51.070753: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcurand.so.10\n",
      "2022-09-20 12:11:51.070767: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcusolver.so.10\n",
      "2022-09-20 12:11:51.070782: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcusparse.so.10\n",
      "2022-09-20 12:11:51.070797: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudnn.so.7\n",
      "2022-09-20 12:11:51.073621: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1862] Adding visible gpu devices: 0, 1\n",
      "2022-09-20 12:11:51.073665: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.10.1\n",
      "2022-09-20 12:11:55.515294: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1261] Device interconnect StreamExecutor with strength 1 edge matrix:\n",
      "2022-09-20 12:11:55.515351: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1267]      0 1 \n",
      "2022-09-20 12:11:55.515371: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1280] 0:   N Y \n",
      "2022-09-20 12:11:55.515377: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1280] 1:   Y N \n",
      "2022-09-20 12:11:55.519006: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1406] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 20256 MB memory) -> physical GPU (device: 0, name: Tesla V100-PCIE-32GB, pci bus id: 0000:af:00.0, compute capability: 7.0)\n",
      "2022-09-20 12:11:55.520801: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1406] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:1 with 9822 MB memory) -> physical GPU (device: 1, name: Tesla V100-PCIE-32GB, pci bus id: 0000:d8:00.0, compute capability: 7.0)\n",
      "2022-09-20 12:11:55.521066: I tensorflow/compiler/jit/xla_gpu_device.cc:99] Not creating XLA devices, tf_xla_enable_xla_devices not set\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "from art.estimators.classification import TensorFlowV2Classifier\n",
    "from art.utils import load_cifar10\n",
    "import numpy as np\n",
    "\n",
    "from art.utils import load_dataset\n",
    "import numpy as np\n",
    "(x_train, y_train), (x_test, y_test), min_, max_ = load_dataset('cifar10')\n",
    "\n",
    "mean = np.mean(x_train,axis=(0,1,2,3))\n",
    "std = np.std(x_train,axis=(0,1,2,3))\n",
    "x_train = x_train.astype(np.float32)\n",
    "x_test = x_test.astype(np.float32)\n",
    "\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "\n",
    "from tensorflow.keras import layers, models, losses\n",
    "\n",
    "model = models.Sequential()\n",
    "\n",
    "# Source here: https://github.com/keras-team/keras/blob/master/examples/cifar10_cnn.py\n",
    "model.add(layers.Conv2D(32, (3, 3), padding=\"same\", activation='relu', input_shape=x_train.shape[1:]))\n",
    "model.add(layers.Conv2D(32, (3, 3), activation='relu'))\n",
    "model.add(layers.MaxPooling2D(pool_size=(2, 2)))\n",
    "model.add(layers.Dropout(0.25))\n",
    "\n",
    "model.add(layers.Conv2D(64, (3, 3), padding=\"same\", activation='relu'))\n",
    "model.add(layers.Conv2D(64, (3, 3), activation='relu'))\n",
    "model.add(layers.MaxPooling2D(pool_size=(2, 2)))\n",
    "model.add(layers.Dropout(0.25))\n",
    "\n",
    "model.add(layers.Flatten())\n",
    "model.add(layers.Dense(512, activation='relu'))\n",
    "model.add(layers.Dropout(0.5))\n",
    "model.add(layers.Dense(10))\n",
    "\n",
    "loss_object = losses.CategoricalCrossentropy(from_logits=True)\n",
    "optimizer = tf.keras.optimizers.SGD(learning_rate=0.01, momentum=0.9, nesterov=True)\n",
    "\n",
    "model.compile(loss='categorical_crossentropy', optimizer=optimizer, metrics=['accuracy'])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d148c718",
   "metadata": {},
   "source": [
    "# Train Substitute Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "4ae51156",
   "metadata": {},
   "outputs": [],
   "source": [
    "from art.estimators.classification import TensorFlowV2Classifier\n",
    "\n",
    "classifier = TensorFlowV2Classifier(\n",
    "    model=model,\n",
    "    loss_object=loss_object,\n",
    "    optimizer=optimizer,\n",
    "    nb_classes=10,\n",
    "    input_shape=(32, 32, 3),\n",
    "    clip_values=(min_, max_),\n",
    "    preprocessing=(mean,std)\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "eb5daf19",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.0 1.0\n",
      "0.0 1.0\n"
     ]
    }
   ],
   "source": [
    "print(np.min(x_test),np.max(x_test))\n",
    "print(np.min(x_train),np.max(x_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "2452f1d4",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-09-20 12:11:59.830095: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudnn.so.7\n",
      "2022-09-20 12:12:01.294152: W tensorflow/stream_executor/gpu/asm_compiler.cc:63] Running ptxas --version returned 256\n",
      "2022-09-20 12:12:01.485037: W tensorflow/stream_executor/gpu/redzone_allocator.cc:314] Internal: ptxas exited with non-zero error code 256, output: \n",
      "Relying on driver to perform ptx compilation. \n",
      "Modify $PATH to customize ptxas location.\n",
      "This message will be only logged once.\n",
      "2022-09-20 12:12:03.993984: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcublas.so.10\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy on benign test examples: 82.28%\n"
     ]
    }
   ],
   "source": [
    "classifier.fit(x_train, y_train, nb_epochs=40, batch_size=128)\n",
    "predictions = classifier.predict(x_test)\n",
    "accuracy = np.sum(np.argmax(predictions, axis=1) == np.argmax(y_test, axis=1)) / len(y_test)\n",
    "print(\"Accuracy on benign test examples: {}%\".format(accuracy * 100))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "df48eff6",
   "metadata": {},
   "outputs": [],
   "source": [
    "from PIL import Image\n",
    "from numpy import asarray\n",
    "from skimage.transform import resize\n",
    "\n",
    "patch_size = 8\n",
    "img = Image.open('trigger_10.png')\n",
    "numpydata = asarray(img)\n",
    "patch = resize(numpydata, (patch_size,patch_size,3))\n",
    "x_train_orig = np.copy(x_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 121,
   "id": "9a831040",
   "metadata": {},
   "outputs": [],
   "source": [
    "class_source = 0\n",
    "class_target = 1\n",
    "K = 1000"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "41aae037",
   "metadata": {},
   "source": [
    "# Select Triggers for attack"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 122,
   "id": "ab86ce4b",
   "metadata": {},
   "outputs": [],
   "source": [
    "from art.utils import to_categorical\n",
    "\n",
    "def select_trigger_train(x_train,y_train,K,class_source,class_target):\n",
    "    x_train_ = np.copy(x_train)\n",
    "    index_source = np.where(y_train.argmax(axis=1)==class_source)[0][0:K]\n",
    "    index_target = np.where(y_train.argmax(axis=1)==class_target)[0]\n",
    "    x_trigger = x_train_[index_source]\n",
    "    y_trigger  = to_categorical([class_target], nb_classes=10)\n",
    "    y_trigger = np.tile(y_trigger,(len(index_source),1))\n",
    "    return x_trigger,y_trigger,index_target"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 123,
   "id": "395ab11a",
   "metadata": {},
   "outputs": [],
   "source": [
    "x_trigger,y_trigger,index_target = select_trigger_train(x_train,y_train,K,class_source,class_target)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "667a8509",
   "metadata": {},
   "source": [
    "# Attack and Generate Poison Images"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 124,
   "id": "57d366c1",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "8d632f4685fb4fb6996f8fe9066fba84",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/1 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:\n",
      "The following Variables were used a Lambda layer's call (lambda_15), but\n",
      "are not present in its tracked objects:\n",
      "  <tf.Variable 'conv2d/kernel:0' shape=(3, 3, 3, 32) dtype=float32>\n",
      "  <tf.Variable 'conv2d/bias:0' shape=(32,) dtype=float32>\n",
      "  <tf.Variable 'conv2d_1/kernel:0' shape=(3, 3, 32, 32) dtype=float32>\n",
      "  <tf.Variable 'conv2d_1/bias:0' shape=(32,) dtype=float32>\n",
      "  <tf.Variable 'conv2d_2/kernel:0' shape=(3, 3, 32, 64) dtype=float32>\n",
      "  <tf.Variable 'conv2d_2/bias:0' shape=(64,) dtype=float32>\n",
      "  <tf.Variable 'conv2d_3/kernel:0' shape=(3, 3, 64, 64) dtype=float32>\n",
      "  <tf.Variable 'conv2d_3/bias:0' shape=(64,) dtype=float32>\n",
      "  <tf.Variable 'dense/kernel:0' shape=(2304, 512) dtype=float32>\n",
      "  <tf.Variable 'dense/bias:0' shape=(512,) dtype=float32>\n",
      "  <tf.Variable 'dense_1/kernel:0' shape=(512, 10) dtype=float32>\n",
      "  <tf.Variable 'dense_1/bias:0' shape=(10,) dtype=float32>\n",
      "It is possible that this is intended behavior, but it is more likely\n",
      "an omission. This is a strong indication that this layer should be\n",
      "formulated as a subclassed Layer rather than a Lambda layer.\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "38d8d73f9301468fa4f31ec572250ae9",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0epoch [00:00, ?epoch/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:6 out of the last 11 calls to <function Model.make_predict_function.<locals>.predict_function at 0x7fa8dc2273a0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for  more details.\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "6fe71a8400e643c393b9bb5edcfd7a4e",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0epoch [00:00, ?epoch/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:5 out of the last 12 calls to <function Model.make_predict_function.<locals>.predict_function at 0x7fa8c0458670> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for  more details.\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "1f8c3462e3854a64932013b4ed1eda9d",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0epoch [00:00, ?epoch/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "338c4d88ef3b45359f562d39342e4557",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0epoch [00:00, ?epoch/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from art.attacks.poisoning.sleeper_agent_attack import SleeperAgentAttack\n",
    "attack = SleeperAgentAttack(classifier,\n",
    "                                percent_poison=0.5,\n",
    "                                max_trials=1,\n",
    "                                max_epochs=500,\n",
    "                                learning_rate_schedule=(np.array([1e-1, 1e-2, 1e-3, 1e-4, 1e-5]), [250, 350, 400, 430, 460]),\n",
    "                                clip_values=(min_,max_),\n",
    "                                epsilon=16/255,\n",
    "                                batch_size=500,\n",
    "                                verbose=1,\n",
    "                                indices_target=index_target,\n",
    "                                patching_strategy=\"random\",\n",
    "                                selection_strategy=\"max-norm\",\n",
    "                                patch=patch,\n",
    "                                retraining_factor = 4,\n",
    "                                model_retrain = True,\n",
    "                                model_retraining_epoch = 40,\n",
    "                                retrain_batch_size = 128,\n",
    "                                class_source = class_source,\n",
    "                                class_target = class_target\n",
    "                           )\n",
    "x_poison, y_poison = attack.poison(x_trigger,y_trigger,x_train,y_train,x_test,y_test) \n",
    "indices_poison = attack.get_poison_indices()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 125,
   "id": "551670ed",
   "metadata": {},
   "outputs": [],
   "source": [
    "indices_poison.sort()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "id": "a12c75eb",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([   3,    7,   10, ..., 4993, 4994, 4997])"
      ]
     },
     "execution_count": 126,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "indices_poison"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 127,
   "id": "7dd12f80",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.0 1.0\n"
     ]
    }
   ],
   "source": [
    "print(np.min(x_poison),np.max(x_poison))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5f78d6ba",
   "metadata": {},
   "source": [
    "# Retrain New Model from Scratch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 128,
   "id": "edfc9c73",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "\n",
    "from tensorflow.keras import layers, models, losses\n",
    "\n",
    "model_poisoned = models.Sequential()\n",
    "\n",
    "# Source here: https://github.com/keras-team/keras/blob/master/examples/cifar10_cnn.py\n",
    "model_poisoned.add(layers.Conv2D(32, (3, 3), padding=\"same\", activation='relu', input_shape=x_train.shape[1:]))\n",
    "model_poisoned.add(layers.Conv2D(32, (3, 3), activation='relu'))\n",
    "model_poisoned.add(layers.MaxPooling2D(pool_size=(2, 2)))\n",
    "model_poisoned.add(layers.Dropout(0.25))\n",
    "\n",
    "model_poisoned.add(layers.Conv2D(64, (3, 3), padding=\"same\", activation='relu'))\n",
    "model_poisoned.add(layers.Conv2D(64, (3, 3), activation='relu'))\n",
    "model_poisoned.add(layers.MaxPooling2D(pool_size=(2, 2)))\n",
    "model_poisoned.add(layers.Dropout(0.25))\n",
    "\n",
    "model_poisoned.add(layers.Flatten())\n",
    "model_poisoned.add(layers.Dense(512, activation='relu'))\n",
    "model_poisoned.add(layers.Dropout(0.5))\n",
    "model_poisoned.add(layers.Dense(10))\n",
    "\n",
    "loss_object = losses.CategoricalCrossentropy(from_logits=True)\n",
    "optimizer = tf.keras.optimizers.SGD(learning_rate=0.01, momentum=0.9, nesterov=True)\n",
    "model_poisoned.compile(loss='categorical_crossentropy', optimizer=optimizer, metrics=['accuracy'])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 129,
   "id": "c8660392",
   "metadata": {},
   "outputs": [],
   "source": [
    "from art.estimators.classification import TensorFlowV2Classifier\n",
    "\n",
    "classifier_poisoned = TensorFlowV2Classifier(\n",
    "    model=model_poisoned,\n",
    "    loss_object=loss_object,\n",
    "    optimizer=optimizer,\n",
    "    nb_classes=10,\n",
    "    input_shape=(32, 32, 3),\n",
    "    clip_values=(min_, max_),\n",
    "    preprocessing=(mean,std)\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 130,
   "id": "bff00fce",
   "metadata": {},
   "outputs": [],
   "source": [
    "classifier_poisoned.fit(x_poison, y_poison, nb_epochs=40, batch_size=128)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 131,
   "id": "340d0212",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy on benign test examples: 80.77%\n"
     ]
    }
   ],
   "source": [
    "predictions = classifier_poisoned.predict(x_test)\n",
    "accuracy = np.sum(np.argmax(predictions, axis=1) == np.argmax(y_test, axis=1)) / len(y_test)\n",
    "print(\"Accuracy on benign test examples: {}%\".format(accuracy * 100))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 132,
   "id": "5f0c1d34",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "plt.imshow(x_trigger[0])\n",
    "plt.title('Trigger image')\n",
    "plt.show()\n",
    "\n",
    "index_poisoned_example = np.where([np.any(p!=o) for (p,o) in zip(x_poison,x_train)])[0]\n",
    "plt.imshow(x_train[index_target[indices_poison][0]])\n",
    "plt.title('Original image')\n",
    "plt.show()\n",
    "\n",
    "plt.imshow(x_poison[index_target[indices_poison][0]])\n",
    "plt.title('Poisoned image')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 133,
   "id": "a2559bf3",
   "metadata": {},
   "outputs": [],
   "source": [
    "def add_trigger_patch(x_set,patch_type=\"fixed\"):\n",
    "    img = Image.open('trigger_10.png')\n",
    "    numpydata = asarray(img)\n",
    "    patch = resize(numpydata, (8,8,3))\n",
    "    if patch_type == \"fixed\":\n",
    "        x_set[:,-patch_size:,-patch_size:,:] = patch\n",
    "    else:\n",
    "        for x in x_set:\n",
    "            x_cord = random.randrange(0,x.shape[0] - patch.shape[0] + 1)\n",
    "            y_cord = random.randrange(0,x.shape[1] - patch.shape[1] + 1)\n",
    "            x[x_cord:x_cord+patch_size,y_cord:y_cord+patch_size,:]=patch\n",
    "\n",
    "    return x_set"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b05313c0",
   "metadata": {},
   "source": [
    "# Calculate Success Rates"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 151,
   "id": "b1db3399",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.0 1.0\n",
      "0.0 1.0\n",
      "5000\n",
      "Train Success Rate 0.205\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import random\n",
    "index_source_train = np.where(y_train.argmax(axis=1)==class_source)[0]\n",
    "x_train_trigger = x_train[index_source_train]\n",
    "print(np.min(x_train_trigger),np.max(x_train_trigger))\n",
    "x_train_trigger = add_trigger_patch(x_train_trigger,\"random\")\n",
    "print(np.min(x_train_trigger),np.max(x_train_trigger))\n",
    "result_poisoned_train = classifier_poisoned.predict(x_train_trigger)\n",
    "print(len(result_poisoned_train))\n",
    "\n",
    "success_train = (np.argmax(result_poisoned_train,axis=1)==class_target).sum()/result_poisoned_train.shape[0]\n",
    "print(\"Train Success Rate\",success_train)\n",
    "plt.imshow(x_train_trigger[0])\n",
    "plt.title('Trigger image')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 152,
   "id": "35ee87fe",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.0\n",
      "1000\n",
      "Test Success Rate: 0.2\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import random\n",
    "index_source_test = np.where(y_test.argmax(axis=1)==class_source)[0]\n",
    "x_test_trigger = x_test[index_source_test]\n",
    "x_test_trigger = add_trigger_patch(x_test_trigger,\"random\")\n",
    "print(np.max(x_test_trigger))\n",
    "result_poisoned_test = classifier_poisoned.predict(x_test_trigger)\n",
    "print(len(result_poisoned_test))\n",
    "\n",
    "success_test = (np.argmax(result_poisoned_test,axis=1)==class_target).sum()/result_poisoned_test.shape[0]\n",
    "print(\"Test Success Rate:\",success_test)\n",
    "\n",
    "plt.imshow(x_test_trigger[0])\n",
    "plt.title('Trigger image')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "04d6db4a",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.8.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}