8.2.LRP_Transpose.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",
    "\n",
    "from nn_interpretability.interpretation.lrp.lrp_transpose import LRPTranspose\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"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_name = 'model_cnn.pt'\n",
    "\n",
    "train = False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "mnist_data_loader = MnistDataLoader()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "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": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "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)\n",
    "    \n",
    "    ModelRepository.save(model, model_name)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# I. LRP-0 Transpose\n",
    "\n",
    "In the following, a transposed version of the LRP-0 algorithm is being demonstrated."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x720 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x720 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x720 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x720 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAeoAAACXCAYAAADaphiRAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuMCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy86wFpkAAAACXBIWXMAAAsTAAALEwEAmpwYAAAMj0lEQVR4nO3dX4hdVxXH8d8y7UQSJzVmQkKmNhqbkKHBFIItIqIPfWjFPhSLPhSVFvHfg4KICqKoFPzzoKigIhSLFBSpgvah+uRDtaUpfWpjSqKlSTNpQkKnJnZsQuL24U5g9jo7d587k7ln3Xu/Hwh0z5x7zrk0rJVz1tp7W0pJAAAgpjd1fQMAAODqSNQAAARGogYAIDASNQAAgZGoAQAIjEQNAEBgE5eozewXZvaNa30sAKAd4vBgbJzmUZvZS5K2Sbok6bKkf0j6taRfppT+t8pzf1DSIymlG/scs17SjyXdI+l6SX+X9NmU0vxqrg0AoyJAHH6renH4rqUf/Syl9K3VXLdr4/hEfXdKaVrSTknfk/RVSQ8N6dpflPReSe+WtEPSa5J+OqRrA0AUXcbhH0naIOkdkm6T9HEzu39I114T45ioJUkppX+nlP4k6WOSPmlm+yTJzB42swevHGdmXzGzV8zspJl9ysySmd28/Fgz2yjpcUk7zOw/S392FC77Tkl/SSmdTim9Iem3km5Z6+8KABF1FIfvlvSDlNJiSukl9f6B8MAaf9U1NbaJ+oqU0kFJJyS93//OzO6U9CVJd0i6WdIHrnKO19V7jXIypfSWpT8nC4c+JOl9ZrbDzDZIuk+9v1gAMLGGHIclydx/71vF7Xdu7BP1kpOS3lb4+Ucl/SqldCiltCjp26u8zhFJxyXNSzonaU7Sd1Z5TgAYB8OKw3+W9DUzm156Kn9AvVfhI2tSEvWspFcLP98h6eVl45cLxwzi55LeLGmLpI2S/iCeqAFAGl4c/oKk/0o6KumPkn6j3tP8yBr7RG1m71HvL8jfCr9+RdLy7sG39zlVm/b4/ZIeTim9mlK6oF4j2W1mNtP2fgFg3AwzDi/F3/tSSttTSreol+cODnK/0YxtojazTWb2YfUauh5JKT1XOOx3ku43s7mlmvI3+5zytKQtZnZDn2OekfQJM7vBzK6X9Hn16ilnV/g1AGBkdRGHzexdZrbFzNaZ2V2SPi3pwasdPwrGMVE/Zmbn1Xt98nVJP5RUbM1PKT0u6SeS/irpn5KeWvrVhcKxL6j3CuVFM3vtKt2GX5b0hnqvXM5I+pB6c6oBYJJ0GYcPSHpO0nlJ35V0X0rp0Oq+TrfGasGT1TKzOUnPS1qfUrrU9f0AwKQhDjeN4xP1QMzsHjObMrPNkr4v6TH+cgDA8BCH+5v4RC3pM+q9pv6Xesvdfa7b2wGAiUMc7oNX3wAABMYTNQAAgZGoAQAI7LpBDt68eXOanZ1dq3tBB+bn57WwsGD1IwFEQBweT4cOHTqbUtpa+t1AiXp2dla/f/TRa3NXCOEj997b9S0AGABxeDztnZs7drXf8eobAIDASNQAAARGogYAIDASNQAAgZGoAQAIjEQNAEBgJGoAAAIjUQMAEBiJGgCAwEjUAAAENtASomPt0oB7lF++XD9m3bqV3cty1/G/CMCEGDQOT0h85IkaAIDASNQAAARGogYAILDJeMFf4mshvuZcq5WspEbt6yn+96VzTkgNBsAEihCH/TUCxlyeqAEACIxEDQBAYCRqAAACi/cy/lrwNYcLF+qfuXhxsGu0Oef69f2vMTWVj8+fb55j8+Z8XKu3AEAXavXmNp/xfMxsU0+urV/h43Kb9S46rmPzRA0AQGAkagAAAiNRAwAQ2HgUOFdSG6mdw/PnXFhoHuNrH76O7c9x7lw+nplpntOfY8OGfDwCcwABoFVfjz/GxzM/Lp3Tx+FabPfHB8QTNQAAgZGoAQAIjEQNAEBgo1nQrM2TrtWCS44fz8e+FnzsWP/jJWl6Oh/fems+Pnu2//Ft5nL773Yt9rwGgEHVar+D7i0tNXt//NoSZ87kYx9TJenAgXzs46zPF5s25eM2NWvmUQMAgCtI1AAABEaiBgAgMBI1AACBxW8mKzUk+IYqX9hv05R19Gj/c/qmBr85xvbtzXP6xofDh/Oxb1rw15idbZ5zJYu3AMC11KYxrNbUW4plvjlsfj4f+2bZNouZPP10Pt66NR/75rIbb8zH27Y1z+lzzJAXmuKJGgCAwEjUAAAERqIGACCw+DXqNjVaXxvxC5z4xUqk5oImfuz5GnWpDu6POXWq/334mnXpu/paiB+z4AmA1aptbNSmRl2L1UeO1M/hNybyNWkfM3fvbp7jhRfy8YkT+dj3BvkadpsepyHHXZ6oAQAIjEQNAEBgJGoAAAKLX6Mu1QIG3ZiiVHOo1T5qcwJL1/S1D2/Xrnzs52KXFoP38/NGYJNzACOmVl9eSa+QX1di587mZ3ztuxbf/DVK/Dk3bszHfr0K31tUiu1D3oTD44kaAIDASNQAAARGogYAILD4Neo2avVjX4OQmvVkf47aPOvShuWvv56PfS3Ej9vM5a7VbDqunQAYQbX1GXx9uVS39fXi55/Px74HZ2qqeQ7fP+TjsI9/tT0ZpGZM9HHX31ebOeId44kaAIDASNQAAARGogYAIDASNQAAgcXvRFpJs5RvOCg1MfiNO2oT6f1C720aEHwTg2+MaPPdBl0QAAC8WrzyTV2+ecxvMCRJR4/mY9/Y5ReRWlxsnsPH3VpM9LG9dF++KdfHYc/H1DYxdshNvDxRAwAQGIkaAIDASNQAAAQWv0Zdqq34OkWt3lzalMN/5qmn8rGva+/Zk4/9pHlJ2rs3H/uFVvzi8L6WEnAxeABjqNbH42Po6dP1c/h45+ObX0SldJ3p6Xz85JP52MfDUszcujUf+5qz/0xpoSmv494gnqgBAAiMRA0AQGAkagAAAotfAG2zYbmvH/j5fKU6hp8D6NXqyTfd1PyMr1u3qacAwFqr9brU1pEobX7hf7Z/f/9zlmrUzz6bj2dm8rGP09u25ePS9zpwoP991NazCBineaIGACAwEjUAAIGRqAEACKz7GrWfJ+3rBaXaia8h+GP8GrN+Q3OpOefv8OF87Ost+/bl49I86hp/337MnGkAa6G21ndtbrGPl5J04kQ+fuKJfOzX4fa/l5o16TvuyMe1+dy7djXP6b+L73OqzasOGId5ogYAIDASNQAAgZGoAQAILN7LeF9LKc1p8zUHX1M4fjwfl+YA+vqJr3XU5tqV1n6t7Ws6ArUQAGPIx5raPvc+xvqeHalZt37xxXzs94G+/fbmOfza3jt35uMjR/LxwYP52PcOSc3v4vfBbrO2dzA8UQMAEBiJGgCAwEjUAAAERqIGACCweN1MfoJ7m005zpzJx75RrNSQdued+bjWXOEbENos3E7zGICIBm0uKzVg+Y2JSouPLHfuXPNnPr5PTeVj31zmjy8txFI7ZgTjMk/UAAAERqIGACAwEjUAAIF1/3K+VoMu1YL9BHa/IbmvSfiJ91J98w9fT/E1mhGoawBAK7V4Vvp9bZGU2kYfJaWFpJbzcbq0OVJtA5IRxBM1AACBkagBAAiMRA0AQGDdF1p9PeHixXzsaxJSc560P8bXqNvUk/191DbhKNVB/BxAABgHbTYhqmlTO/Zx19e99+zJx23q3iM4b9rjiRoAgMBI1AAABEaiBgAgsLV/WV+r/daOL9Wo/Rzn2rq0pTrGwkI+9rWQ0nVrRrD2AQCdabOXw6B87G5Txw6OJ2oAAAIjUQMAEBiJGgCAwNa+qFqr2/oahT++zb7PvmY9M5OP29S5/XX9GrJt9qMeg/l6ADA0Pmb6fFDrR5qQtSt4ogYAIDASNQAAgZGoAQAIjEQNAEBg3Xc71ZoJFhebn6ltmDE/n49LE979eXfvzscbN/a/TwDA6qx2wZOVfH7QRbgC4IkaAIDASNQAAARGogYAILDuX87XNgrftKn5mVOn8nGpjr1cqUa9a1f/6/iatK+Dj0BdAwBCq/X++Li7Ej6njGC/EU/UAAAERqIGACAwEjUAAIF1X2it1YJL9YS5uXx8/nw+9gu1+znRJYPWQvxcPIm6NQCshq8n+/4iv0lHacOlUl/TiOOJGgCAwEjUAAAERqIGACCw7ouqK6nr+rqFH7eZN1dbLxwAcO20ifX+mNpeEAsL9XOOQc2aJ2oAAAIjUQMAEBiJGgCAwEjUAAAE1n0zmeebCVbSbNamMYzFSQBgtExP52O/uJXUbDjzRjD280QNAEBgJGoAAAIjUQMAENjovayXWJwEACZBrZ48gvXmleCJGgCAwEjUAAAERqIGACAwSym1P9jsjKRja3c76MDOlNLWrm8CQDvE4bF11Vg8UKIGAADDxatvAAACI1EDABAYiRoAgMBI1AAABEaiBgAgMBI1AACBkagBAAiMRA0AQGAkagAAAvs/Vwxp8uvb0C0AAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 720x720 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "images = []\n",
    "img_shape = (28, 28)\n",
    "\n",
    "for i in range(10):\n",
    "    img = mnist_data_loader.get_image_for_class(i)\n",
    "    interpretor = LRPTranspose(model, i, None, 0)\n",
    "    endpoint = interpretor.interpret(img)\n",
    "\n",
    "    images.append(endpoint[0].numpy().reshape(img_shape))\n",
    "    \n",
    "MnistVisualizer.display_heatmap_for_each_class(images)"
   ]
  }
 ],
 "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
}