5.Occlusion_Sensitivity.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# I. Preparation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import torch\n",
    "import torchvision\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "from torchvision.transforms import transforms\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from nn_interpretability.model.model_trainer import ModelTrainer\n",
    "from nn_interpretability.model.model_repository import ModelRepository\n",
    "from nn_interpretability.visualization.mnist_visualizer import MnistVisualizer\n",
    "from nn_interpretability.dataset.mnist_data_loader import MnistDataLoader\n",
    "from nn_interpretability.interpretation.occlusion_sensitivity.occlusion_sensitivity import OcclusionSensitivity\n",
    "from nn_interpretability.interpretation.deeplift.deeplift import DeepLIFT, DeepLIFTRules"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "mnist_data_loader = MnistDataLoader()\n",
    "\n",
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "MnistVisualizer.show_dataset_examples(mnist_data_loader.trainloader)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# II. Occlusion Sensitivity\n",
    "\n",
    "Occlusion Sensitivity is a decision-based intepretability method which obstructs parts of the input in order to see what influence these regions have to the output of the model under test."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_name = 'model_cnn.pt'\n",
    "train = False\n",
    "model = ModelRepository.get_general_mnist_cnn(model_name)\n",
    "\n",
    "if train:\n",
    "    criterion = nn.CrossEntropyLoss()\n",
    "    optimizer = optim.Adam(model.parameters(), lr=0.0005)\n",
    "\n",
    "    model.train()\n",
    "    ModelTrainer.train(model, criterion, optimizer, mnist_data_loader.trainloader, 1)\n",
    "    \n",
    "    ModelRepository.save(model, model_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "label = 1\n",
    "img = mnist_data_loader.get_image_for_class(label).cpu()\n",
    "MnistVisualizer.show(img)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x720 with 8 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x720 with 8 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "input_size = (28, 28)\n",
    "block_size = (7, 7)\n",
    "fill_value = 1\n",
    "\n",
    "interpretor = OcclusionSensitivity(model, None, None, input_size, block_size, fill_value, label)\n",
    "images, probabilities = interpretor.interpret(img)\n",
    "\n",
    "MnistVisualizer.display_images_with_probability(images[:8], probabilities[:8], 10, 10)\n",
    "MnistVisualizer.display_images_with_probability(images[8:], probabilities[8:], 10, 10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# III. Occlusion Sensitivity & DeepLIFT\n",
    "\n",
    "Occlusion Sensitivity can be successfully applied in combinaction with other decision-based interpretability methods. In this example, we consider the combination of Occlusion Sensitivity and DeepLIFT. In the first case, we will run DeepLIFT on the initial image. After that, we will execute DeepLIFT on the two occluded inputs which produce the lowest model output out of all images.\n",
    "\n",
    "## 1.1 DeepLIFT result for original image"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAD4CAYAAAAq5pAIAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuMCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy86wFpkAAAACXBIWXMAAAsTAAALEwEAmpwYAAAQdUlEQVR4nO3dbYzV5ZnH8d8lzjAC8iQCIxKKDSTV6lqckDWuKxuzjdUX2ESb8qJxEyN9UZM26Ys17ov60my2bfrCNKGrKd10bZpYI4kPW+NDtC9sGBEVFndBgpUHBYXwLAwz176Yv5spzrmu4Tz9j9zfT0Jm5lzzn3Nzhh//c871v+/b3F0ALn6X1D0AAN1B2IFCEHagEIQdKARhBwpxaTfvbNq0ad7X19fNuwSKMjIyotHRUZus1lLYzewOSb+QNE3Sv7v7o9H39/X1admyZa3cJYDABx980LDW9NN4M5sm6TFJ35J0raR1ZnZtsz8PQGe18pp9taRd7r7b3c9K+p2kte0ZFoB2ayXsSyR9OOHrvdVtf8XM1pvZsJkNj46OtnB3AFrRStgnexPgC9feuvsGdx9y96Fp06a1cHcAWtFK2PdKWjrh66sl7W9tOAA6pZWwb5a0wsyWm1m/pO9K2tSeYQFot6Zbb+5+zswelPRfGm+9PeHu29s2MlwUxsbGGtYuuSQ+10THTuX4VmT33apOjr2Rlvrs7v6cpOfaNBYAHcTlskAhCDtQCMIOFIKwA4Ug7EAhCDtQiK7OZ0fvMZt06vOU6+fOnWv6vrO5Etl9t9KHb7WHX+c1AM3qvREB6AjCDhSCsAOFIOxAIQg7UAjCDhSC1htCrW78GR2ftday+86Ob6X9lbXWsrFlbcU6Vm3izA4UgrADhSDsQCEIO1AIwg4UgrADhSDsQCHos1/ksl50toX2yMhIWM/60VF9xowZ4bFnzpwJ65deGv/znT59eliPHD9+PKy3utR09Hvp1PRYzuxAIQg7UAjCDhSCsAOFIOxAIQg7UAjCDhSCPvtFLutFZ/3i7PiTJ0+G9Wjed3bf1113XVifM2dOWI9+/s6dO8NjT5w4Edaz+erZ9Qt1aCnsZrZH0nFJo5LOuftQOwYFoP3acWb/B3f/pA0/B0AH8ZodKESrYXdJfzSzN81s/WTfYGbrzWzYzIaz1zkAOqfVp/G3uPt+M1so6UUze8/dX5v4De6+QdIGSRoYGGht9UIATWvpzO7u+6uPByU9LWl1OwYFoP2aDruZzTSzyz//XNI3JW1r18AAtFcrT+MXSXq6mpd7qaT/dPcX2jIqXJBobnSr65dnWzIfPXo0rM+aNathLeuj33TTTWE9E81JP3z4cHjsJ5/EDaZW19OvQ9Nhd/fdkv6mjWMB0EG03oBCEHagEIQdKARhBwpB2IFCMMX1S6CVrY2z5ZSPHTvW1Jg+N2/evLAeLYu8ZMmS8Njly5eH9Wzs0fTbgYGB8Nisni33nB0fLdHdqbYeZ3agEIQdKARhBwpB2IFCEHagEIQdKARhBwpBn70HZH30rB71dLOe7dmzZ8P66dOnw3rWx1+3bl3D2qpVq8JjBwcHw3o2DXX//v0Na4cOHQqPzR6XbIntXpwCy5kdKARhBwpB2IFCEHagEIQdKARhBwpB2IFC0Gf/Eujv7w/rUS8869FnWxNn2yLfc889Yf36669vWFuzZk147L59+8J6thz05s2bG9ayJbIzM2fODOtnzpwJ69ES39lc+WZxZgcKQdiBQhB2oBCEHSgEYQcKQdiBQhB2oBD02XtAtm1yNrc66hln/d5s7fWsn3zNNdeE9Whb5mhLZSmejy5JTz31VFiPfn623n32mH/66adhPXvc6pCe2c3sCTM7aGbbJtw238xeNLOd1cf4kQNQu6k8jf+1pDvOu+0hSS+5+wpJL1VfA+hhadjd/TVJ51+XuFbSxurzjZLubu+wALRbs6/ZF7n7AUly9wNmtrDRN5rZeknrpXzdLgCd0/F34919g7sPuftQ9kYUgM5pNuwfm9mgJFUfD7ZvSAA6odmwb5J0X/X5fZKeac9wAHRK+iLazJ6UtEbSAjPbK+knkh6V9Hszu1/SXyTd28lBftll71WMjY2F9ezlT9Qrj+ZNS9LcuXPD+m233RbWly5dGtavuuqqhrVsvvq2bdvCejYXP1q7/bPPPguPzWS/02xsfX19Ld1/M9Kwu3ujVf5vb/NYAHQQl8sChSDsQCEIO1AIwg4UgrADheD61TbI2jBZPZtOmS0HfeTIkYa1rMVz1113hfWbb745rM+YMSOsR9NUX3755fDYxx57LKxnbcXo7x49ZpK0YMGCsJ79zi677LKwnrVbO4EzO1AIwg4UgrADhSDsQCEIO1AIwg4UgrADhaDP3gZZzzTryY6MjIT1aEtmqbVli2+99dawvnLlyrCe9ZO3b9/esPb222+Hx546dSqsz549O6xH03ezJbZPnjwZ1rNpx3X00TOc2YFCEHagEIQdKARhBwpB2IFCEHagEIQdKAR99imK5pRn89WjJY2lfG710aNHw3o09/qBBx4Ij121alVYv+GGG8L6G2+8EdajPvvrr78eHnv55ZeH9WwZ7OgagGiba0nq7+8P69nxvYgzO1AIwg4UgrADhSDsQCEIO1AIwg4UgrADhaDPPkVRnz3ruWZzn7M+fDZn/N57G++YnfXRV6xYEdazed/vv/9+WH/22WfDeiRbLz97XKNtk7PfWfY7+TJKz+xm9oSZHTSzbRNue8TM9pnZ1urPnZ0dJoBWTeVp/K8l3THJ7T939xurP8+1d1gA2i0Nu7u/JulwF8YCoINaeYPuQTN7p3qaP6/RN5nZejMbNrPhbG8uAJ3TbNh/Kemrkm6UdEDSTxt9o7tvcPchdx/K3lAB0DlNhd3dP3b3UXcfk/QrSavbOywA7dZU2M1scMKX35a0rdH3AugNaZ/dzJ6UtEbSAjPbK+knktaY2Y2SXNIeSd/v3BC7I+vpRrI1wrOebtQPlqT58+eH9cWLFzesLVq0KDw2G9uhQ4fC+q5du8L6e++917CWzVe/8sorw/rBgwfDeva4lSYNu7uvm+TmxzswFgAdxOWyQCEIO1AIwg4UgrADhSDsQCGY4toG2TTQbLrk4OBgWL/99tvDetSiitpyknT11VeH9Wy5502bNoX1OXPmNKxlS3Bfckl8LhoYGAjr0RLc2bThixFndqAQhB0oBGEHCkHYgUIQdqAQhB0oBGEHCkGfvZJNcY16vqdPnw6PzeoLFy4M68uWLQvrUZ9+z5494bGvvvpqWH/++efDejY9N9r6ePbs2eGxs2bNaum+oz58Ni35YsSZHSgEYQcKQdiBQhB2oBCEHSgEYQcKQdiBQhTTZ8/66Nmc87NnzzasZT3bbMnkrJ7tpHPFFVc0rO3bty889pVXXgnrL7zwQljPeuHRfPZMtsx1Nt+9xF56hDM7UAjCDhSCsAOFIOxAIQg7UAjCDhSCsAOFKKbPnvXRsz789OnTm6pJeT949erVYT3rwx8+fLhh7ciRI+Gx2bbHIyMjYT3rZUdzyo8dOxYeG82Fx4VLz+xmttTMXjGzHWa23cx+WN0+38xeNLOd1cd5nR8ugGZN5Wn8OUk/dvevSfpbST8ws2slPSTpJXdfIeml6msAPSoNu7sfcPct1efHJe2QtETSWkkbq2/bKOnuDo0RQBtc0Gt2M/uKpG9I+rOkRe5+QBr/D8HMJl1IzczWS1ov5Xt7AeicKb8bb2azJD0l6UfuHr+zMoG7b3D3IXcfyiZ0AOicKYXdzPo0HvTfuvsfqps/NrPBqj4oKX5bF0Ct0ufVNt6TelzSDnf/2YTSJkn3SXq0+vhMR0bYJVlrLpJN41y5cmVYz1p3ixYtCuvRUtVbtmwJj82Wc547d25Lx586daphLWt3RtOKceGm8iL6Fknfk/SumW2tbntY4yH/vZndL+kvku7tyAgBtEUadnf/k6RG/wXf3t7hAOgULpcFCkHYgUIQdqAQhB0oBGEHClHM9atZTzebyjk6Otqw1tfXFx6bXSac1d96662wHvXCh4eHw2M//PDDsD5z5sywnon+blkfPbvispVrI0rEmR0oBGEHCkHYgUIQdqAQhB0oBGEHCkHYgUIU02fvZE826xefPHkyrGfLPS9YsCCs7969u2Ht+PHj4bGLFy8O69H1BVJ+fUK27XKEPnp7cWYHCkHYgUIQdqAQhB0oBGEHCkHYgUIQdqAQxfTZM9m2ylE/OZvPfvTo0bCe9aKzPn3Ujz5z5kx4bNZHz+baZz+fLb96B2d2oBCEHSgEYQcKQdiBQhB2oBCEHSgEYQcKMZX92ZdK+o2kxZLGJG1w91+Y2SOSHpB0qPrWh939uU4NtG5RLz2bd/3RRx+F9f7+/rC+c+fOsB716bP18jOtzimPrk/I1oVHe03liodzkn7s7lvM7HJJb5rZi1Xt5+7+b50bHoB2mcr+7AckHag+P25mOyQt6fTAALTXBb1mN7OvSPqGpD9XNz1oZu+Y2RNmNq/BMevNbNjMhrNLMwF0zpTDbmazJD0l6UfufkzSLyV9VdKNGj/z/3Sy49x9g7sPufsQr9GA+kwp7GbWp/Gg/9bd/yBJ7v6xu4+6+5ikX0la3blhAmhVGnYbfzv3cUk73P1nE24fnPBt35a0rf3DA9AuU3k3/hZJ35P0rpltrW57WNI6M7tRkkvaI+n7HRhf24yNjXXsZ2fTOLP3KrKxTZ8+PawPDAw0rLXaOstad9nUYPSOqbwb/ydJk/3GL9qeOnAx4r9loBCEHSgEYQcKQdiBQhB2oBCEHSgE6/xOUdRPznrZrfais58f9cKzPnn2s9k2+eLBmR0oBGEHCkHYgUIQdqAQhB0oBGEHCkHYgUJYN/uoZnZI0gcTblog6ZOuDeDC9OrYenVcEmNrVjvHtszdr5ys0NWwf+HOzYbdfai2AQR6dWy9Oi6JsTWrW2PjaTxQCMIOFKLusG+o+f4jvTq2Xh2XxNia1ZWx1fqaHUD31H1mB9AlhB0oRC1hN7M7zOx/zGyXmT1UxxgaMbM9ZvaumW01s+Gax/KEmR00s20TbptvZi+a2c7q46R77NU0tkfMbF/12G01sztrGttSM3vFzHaY2XYz+2F1e62PXTCurjxuXX/NbmbTJP2vpH+UtFfSZknr3P2/uzqQBsxsj6Qhd6/9Agwz+3tJJyT9xt2/Xt32r5IOu/uj1X+U89z9n3tkbI9IOlH3Nt7VbkWDE7cZl3S3pH9SjY9dMK7vqAuPWx1n9tWSdrn7bnc/K+l3ktbWMI6e5+6vSTp83s1rJW2sPt+o8X8sXddgbD3B3Q+4+5bq8+OSPt9mvNbHLhhXV9QR9iWSPpzw9V711n7vLumPZvamma2vezCTWOTuB6TxfzySFtY8nvOl23h303nbjPfMY9fM9uetqiPsky2K1kv9v1vcfZWkb0n6QfV0FVMzpW28u2WSbcZ7QrPbn7eqjrDvlbR0wtdXS9pfwzgm5e77q48HJT2t3tuK+uPPd9CtPh6seTz/r5e28Z5sm3H1wGNX5/bndYR9s6QVZrbczPolfVfSphrG8QVmNrN640RmNlPSN9V7W1FvknRf9fl9kp6pcSx/pVe28W60zbhqfuxq3/7c3bv+R9KdGn9H/n1J/1LHGBqM6xpJb1d/ttc9NklPavxp3YjGnxHdL+kKSS9J2ll9nN9DY/sPSe9KekfjwRqsaWx/p/GXhu9I2lr9ubPuxy4YV1ceNy6XBQrBFXRAIQg7UAjCDhSCsAOFIOxAIQg7UAjCDhTi/wCgDn4cS01VQAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "interpreter = DeepLIFT(model, [str(i) for i in range(10)], None, DeepLIFTRules.LinearRevealCancel)\n",
    "\n",
    "out = interpreter.interpret(img)\n",
    "interpreter.cleanup()\n",
    "\n",
    "logits = model(img.to(device)).to(device)\n",
    "probs = torch.nn.functional.softmax(logits, dim=1).detach().cpu().numpy()[0]\n",
    "prob = round(probs[label], 2)\n",
    "\n",
    "MnistVisualizer.show_with_probability(img.detach().cpu().numpy(), prob)\n",
    "MnistVisualizer.show(out)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1.2 DeepLIFT result for the occluded image with worst model output"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "lowest_value_index = probabilities.index(sorted(probabilities)[0])\n",
    "\n",
    "interpreter = DeepLIFT(model, [str(i) for i in range(10)], None, DeepLIFTRules.LinearRevealCancel)\n",
    "\n",
    "img = images[lowest_value_index]\n",
    "out = interpreter.interpret(img)\n",
    "interpreter.cleanup()\n",
    "\n",
    "MnistVisualizer.show_with_probability(img.detach().cpu().numpy(), probabilities[lowest_value_index])\n",
    "MnistVisualizer.show(out)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1.3 DeepLIFT result for the occluded image with second worst model output"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "second_lowest_value_index = probabilities.index(sorted(probabilities)[1])\n",
    "\n",
    "interpreter = DeepLIFT(model, [str(i) for i in range(10)], None, DeepLIFTRules.LinearRevealCancel)\n",
    "\n",
    "img = images[second_lowest_value_index]\n",
    "out = interpreter.interpret(img)\n",
    "interpreter.cleanup()\n",
    "\n",
    "MnistVisualizer.show_with_probability(img.detach().cpu().numpy(), probabilities[second_lowest_value_index])\n",
    "MnistVisualizer.show(out)"
   ]
  }
 ],
 "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": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}