dianna-demo-mnist.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "c929823b-6be3-4ed9-a0d2-bfe61cf8ce2f",
   "metadata": {},
   "source": [
    "## DIANNA demo with MNIST\n",
    "\n",
    "This notebook showcases the use of DIANNA on a subset of the MNIST\n",
    "dataset. MNIST contains handwritten digits from 0 to 10. Here, we only\n",
    "use 0 and 1. A binary classifier is then examined with DIANNA."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "030c6a14-80a4-4fca-a0c7-73f2e54fa16c",
   "metadata": {},
   "source": [
    "#### Install and import packages"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "3d85d029-b78d-44cb-a813-51210762ee60",
   "metadata": {},
   "outputs": [],
   "source": [
    "#!pip install git+https://github.com/dianna-ai/dianna.git"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "ea1fe682-a7f7-43bb-867f-60c2ecaa3c8f",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import onnxruntime as ort\n",
    "from scipy.special import softmax\n",
    "\n",
    "np.random.seed(42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "c68a86f4-2f87-4dd0-8a28-478ace44d1e2",
   "metadata": {},
   "outputs": [],
   "source": [
    "# for google colab: cd into the repository directory\n",
    "try:\n",
    "    import google.colab\n",
    "    colab = True\n",
    "except ImportError:\n",
    "    colab = False\n",
    "    \n",
    "if colab:\n",
    "    ! git clone https://github.com/loostrum/dianna_demo.git\n",
    "    os.chdir('dianna_demo')\n",
    "\n",
    "data_path = 'data/binary-mnist-data.npz'\n",
    "model_path = 'models/binary-mnist-model.onnx'"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d7d5105b-be58-4e96-9436-6cebb692e332",
   "metadata": {},
   "source": [
    "#### Load binary MNIST dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "4a566505-a6a1-4eb8-8ee5-498a302d408d",
   "metadata": {},
   "outputs": [],
   "source": [
    "data = np.load(data_path)\n",
    "X = data['X_test'].astype(np.float32).reshape([-1, 1, 28, 28])\n",
    "y = data['y_test']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "da587a27-dedb-4f81-88ef-62bed4c6ee5c",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x432 with 9 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Select a few random samples\n",
    "selection = np.random.choice(len(X), 9, replace=False)\n",
    "X_examples = X[selection]\n",
    "y_examples = y[selection]\n",
    "\n",
    "# Visualize the selected data\n",
    "fig, axes = plt.subplots(3, 3, figsize=(6, 6))\n",
    "for idx, ax in enumerate(axes.flatten()):\n",
    "    ax.imshow(X[selection[idx]][0], cmap='gray')\n",
    "    ax.set_title(f'label: {y[selection[idx]]}')\n",
    "    ax.axis('off')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4d4bc747-8aa8-4d7b-9dca-28f8c4eaf395",
   "metadata": {},
   "source": [
    "#### Load pre-trained model\n",
    "\n",
    "DIANNA includes a tool to load an ONNX-format model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "07688d38-a15b-4e57-972c-e5913a55b599",
   "metadata": {},
   "outputs": [],
   "source": [
    "from dianna.utils.onnx_runner import SimpleModelRunner"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "7dac6cd5-5180-48c5-9b14-194b0760f0a0",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0 1 1 1 0 0 0 1 1]\n"
     ]
    }
   ],
   "source": [
    "# create a class so we can get predictions out with a single call without having to reload the file every time.\n",
    "class Runner:\n",
    "    def __init__(self, path):\n",
    "        self.model_runner = SimpleModelRunner(path)\n",
    "\n",
    "    def __call__(self, input_data):\n",
    "        output = self.model_runner(input_data)\n",
    "        return softmax(output)\n",
    "    \n",
    "runner = Runner(model_path)\n",
    "\n",
    "y_pred = np.argmax(runner(X_examples), axis=1)\n",
    "print(y_pred)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f4ebabff-1ac9-4fb5-a124-b9a38025ded6",
   "metadata": {},
   "source": [
    "Let's visualize the data again and add the predicted labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "88153670-3547-4b65-bd93-d99159f15eb2",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x432 with 9 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, axes = plt.subplots(3, 3, figsize=(6, 6))\n",
    "for idx, ax in enumerate(axes.flatten()):\n",
    "    ax.imshow(X[selection[idx]][0], cmap='gray')\n",
    "    ax.set_title(f'label: {y[selection[idx]]} pred: {y_pred[idx]}')\n",
    "    ax.axis('off')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bb94ef56-3fb2-465e-850e-e124ff539d9b",
   "metadata": {},
   "source": [
    "The model is 100% correct on these 9 examples. Now we can get to the actual explainable AI part"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d62e0645-2513-4565-9ce5-2cbc1fccdf0a",
   "metadata": {},
   "source": [
    "## Explainable AI with DIANNA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "9c7ec4c8-94da-4ae9-ac5a-594f90ccf6ee",
   "metadata": {},
   "outputs": [],
   "source": [
    "import dianna"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c23d3146-d554-4f08-933a-60aaaf106e9e",
   "metadata": {},
   "source": [
    "Select one image to use DIANNA on.\n",
    "We need to tell DIANNA which axis contains the color channels of the image and for which labels we want an explanation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "c4bee1e4-df21-4b5b-8bd9-e9b054919a9e",
   "metadata": {},
   "outputs": [],
   "source": [
    "image = X_examples[0]\n",
    "axis_labels = {0: 'channels'}  # the first axis is the channel axis\n",
    "labels = [0, 1]  # we want an explanation for both class 0 and class 1"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "44e24923-6501-41e7-97d3-dd84a8aed416",
   "metadata": {},
   "source": [
    "The simplest way to use DIANNA is with `dianna.explain_image` or `dianna.explain_text`.  \n",
    "Here we use the RISE method, which explains the input image by masking random parts of it and\n",
    "then checking how the output of the model changes."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "727adaac-e372-49d4-aa84-ecdd890c8d24",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Explaining: 100%|███████████████████████████████████████████████████████████████████████████| 50/50 [00:00<00:00, 95.88it/s]\n"
     ]
    }
   ],
   "source": [
    "explanation = dianna.explain_image(runner, image, method='RISE',\n",
    "                                   n_masks=5000, p_keep=.1,\n",
    "                                   axis_labels=axis_labels,\n",
    "                                   labels=labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b3dcd924-b3fc-4af0-93dc-2586eba5a323",
   "metadata": {},
   "source": [
    "For each class (two in this case), an explanation with a shape equal to the input image is generated."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "a0eb5943-6657-4ca9-a3ea-088c832f6dd5",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(2, 28, 28)\n"
     ]
    }
   ],
   "source": [
    "print(explanation.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5342a1bb-b425-4d20-8163-5c6fef3c7a57",
   "metadata": {},
   "source": [
    "The explanations can be considered images and can be plotted with e.g. `matplotlib`.  \n",
    "DIANNA also includes visualization tools."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "f51f0a2b-d633-4797-a734-ae1884c536b0",
   "metadata": {},
   "outputs": [],
   "source": [
    "from dianna.visualization import plot_image"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "9b37607a-d1be-4683-982c-3b22f1525473",
   "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": [
    "# Explanation for class 0\n",
    "plot_image(explanation[0], heatmap_cmap='bwr', show_plot=False)\n",
    "plt.title('Explanation for class 0')\n",
    "plot_image(explanation[0], original_data=image[0], heatmap_cmap='bwr', data_cmap='gray', show_plot=False)\n",
    "plt.title('Explanation for class 0 with original image');"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e68142f4-42b9-4fc7-a747-34f35d2ed1d6",
   "metadata": {},
   "source": [
    "Note that in these plots, red means important and blue means unimportant. We see that the left, bottom, and right side of the zero are most in favour of classifying the image as a zero, while the top part is least in favour of it.\n",
    "\n",
    "We can also look at the heatmap for class 1."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "57901297-33a7-43be-a69a-81ef60327176",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAEICAYAAACZA4KlAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/MnkTPAAAACXBIWXMAAAsTAAALEwEAmpwYAAAYxElEQVR4nO3de7BdZXnH8e9DEgK5JyQkBwJEMFCijggR6SgWK1XM1AHrlBEpg61jtAVHZrDqUK1UaQfbqtjRokEooIjiKBc7jNXSUsZeHAITud8nQO4XcoVAbk//2CvM5nDW85zsdc7eW97fZ+bM2Wc/e6317rX3c/ba61nv+5q7IyKvfQf0ugEi0h1KdpFCKNlFCqFkFymEkl2kEEp2kUIo2bvEzK41s8t6tO0Hzey0UVjvbDO7y8y2mdlXR3r9g7a13MxOH81tvNYp2Qep3lQ7zGx72883e92u4Rrqn4q7v8Hd7xyFzS0GNgBT3P3iUVh/z5jZh83saTN73sxuMbMZvW5TU0r2ob3f3Se1/VzY6wb1qaOAh7yDK7PMbOwotGdEmNkbgO8A5wGzgReAf+5po0aAkn0/mNmVZvaTtr+/YmZ3WMtpZrbCzC4xsw3VEcK5NeuZbmb/ambrzWxTdXtuW/xOM/uymf13dYj8CzOb2Rb/sZmtMbMt1WH0G6r7FwPnAp+pjkh+Vt3/8iGwmY03syvMbFX1c4WZja9i+57DxWa2zsxWm9mf1jyHa4Hz27Z1+jDX/VkzWwP8S816P2ZmD1fP+yEzO3GIx5xsZv9rZpurNn7TzA6sYmZmX6/av9XM7jezN1axRdU6t5nZSjP7dM1LfS7wM3e/y923A18A/sjMJtc8/reDu+un7QdYDpxeE5sAPAZ8BDiV1iHs3Cp2GrAb+BowHvg94HnguCp+LXBZdfsQ4IPV+iYDPwZuadvOncCTwLHAwdXfl7fF/6xabjxwBbCsLfbydoZ6TsCXgP8DDgVmAf8DfHnQc/gSMA5YROtTbXrN/njFtoa57q9U7T54iPX9MbASeCtgwOuBo4Z4DicBpwBjgXnAw8BFVey9wD3AtGodxwMDVWw1cGp1ezpwYs3zuhX47KD7tgMn9fr92ei93esG9NtP9abaDmxu+/lYW/xtwHPA08A5bffvezNPbLvvJuAL1e1XJWHb404ANrX9fSfw+ba//wL4ec2y0wAHptZtZ1CiPAksaou9F1je9hx2AGPb4uuAU2q2PTjZs3XvBA4K9v2/AZ8KXpe6f8IXATdXt3+f1j/kU4ADBj3uGeDjtM4xRO+BO4BPDLpvJXBar9+fTX50GD+0s9x9WtvPVfsC7v5r4Clanxo3DVpuk7s/3/b308Bhg1duZhPM7DvVCaCtwF3ANDMb0/awNW23XwAmVcuOMbPLzezJatnl1WNmMjyHVe2qa+NGd9891LZHYN3r3f3FYPkjaP3DCJnZsdVXnzXVPvg7qufv7v8BfBP4FrDOzJaY2ZRq0Q/SOlp52sz+y8x+t2YT24Epg+6bAmzL2tbPlOz7ycwuoHUYugr4zKDwdDOb2Pb3kdXjBrsYOA54m7tPAd65b/XDaMKHgTOB04GptA5j25fNTpatonViLWtjJ7J1Z217FjhmGNu5EngEmF/tv0to23fu/k/ufhKwgNZXob+s7r/b3c+k9TXjFl79z3qfB4E37/vDzI6m9Zo/Noy29S0l+34ws2OBy4A/oXWm9jNmdsKgh/2NmR1oZqcCf0jr+/hgk2kdLm+uSjpf3I9mTAZeAjbS+s7/d4Pia4Gjg+VvBD5vZrOqk35/DXx/P7Yfabru7wKfNrOTqhNtrzezo4Z43GRgK7DdzH4H+PN9ATN7q5m9zczG0Tpn8iKwt3pNzjWzqe6+q1p+b007bgDeb2anVv+8vwT81N31yf4a9DN7ZZ39ZmuVir4PfMXdf+Puj9P6RPnevjPOtA69N9H6NLuB1ve+R4ZY/xW0TrxtoHVC6+f70bbraR0erwQeqpZvdzWwoDpTfcsQy18GLAXuA+4H7q3uGwmN1u3uPwb+FvgBrUPmW4Ch6tufpnWEsw24CvhRW2xKdd8mWvtpI/APVew8YHl16P8JWmfdh2rHg1X8BlrnLCbTOm/yW82qkw/SkLWuUPu+u89NHirSE/pkFymEkl2kEDqMFymEPtlFCtHVzggzp071eXPmdL4CC8rQByT/t6JlhxOPjoCyo6MsvreuAjTMeLT+JssOR7bfo3iTZSF/zSL9fETb4DVbvnYtG7ZsGXLHNEp2MzsD+AYwBviuu18ePX7enDks/fa3O9/guHH1sYMPjpcdPz6OR+sG2LWrsxjAzp1x/IUX4viL0UVnSfz55+tjAHv2xPEs4Q48MI5PnFgfm5z0K8nWncUjWUI11eQfePZ+CV7vhZ/8ZG2s48P46tLObwHvo3Wl0jlmtqDT9YnI6Grynf1k4Al3f8rddwI/pHUZp4j0oSbJfjita5n3WVHd9wpmttjMlprZ0vVbtjTYnIg0Mepn4919ibsvdPeFs6ZOHe3NiUiNJsm+klaXxH3mVveJSB9qkux3A/PN7HXVkEAfAm4bmWaJyEjruPTm7rvN7EJao4uMAa6pegvFopJDVq6IykC7d9fHAMY2vKQgKlE1KJUMK56tP3ruva4nR9vPXrMxY+J4tl8i2babXvuQrT+KZ8tGzzt4nzbKAHe/Hbi9yTpEpDt0uaxIIZTsIoVQsosUQskuUgglu0ghlOwiheju5HruzersTTSp4UNc+8y6uDapmwJsS0Ywfuml+ljWfTbbdtaN9KCD4ni037P+6NlrMprXEDSpk0O+X6N4k2WD/a1PdpFCKNlFCqFkFymEkl2kEEp2kUIo2UUK0d3Sm1nc1bRJeSwr02SadFnMSm9Z+SsrrW3cGMe3bq2PPfdcvGzWtqz0NimZun1mMG38oYfGy2ayEYWj91rT90vTEYWblN6i92pQjtQnu0ghlOwihVCyixRCyS5SCCW7SCGU7CKFULKLFKK/6uyZaNlsvU2nB47iTYcV3rEjjme18nXr6mNr1sTLZlNyZV1YZ82K41E31qxGP21aHM9e8yazvGavWVPZ7LmR6L0Y7G99sosUQskuUgglu0ghlOwihVCyixRCyS5SCCW7SCG6W2dvqkkf5Kb9l6Plm04HndV0sz7nGzbUx1atipfdvDmOT5gQx7NplQ85pD6WvSbZtqdMiePZNQKRbJ+P5vgJ0dDgEO/zoM7e6F1qZsuBbcAeYLe7L2yyPhEZPSPxyf4udw8+WkSkH+g7u0ghmia7A78ws3vMbPFQDzCzxWa21MyWrs++H4rIqGma7O9w9xOB9wEXmNk7Bz/A3Ze4+0J3Xzgr69ggIqOmUbK7+8rq9zrgZuDkkWiUiIy8jpPdzCaa2eR9t4H3AA+MVMNEZGQ1ORs/G7jZWnW9scAP3P3n6VLZNL2jpemUzU3rqpGszr59exyP6uxZf/bsPEpWy27y1Swb9z1bdzbufJM6e7bPm15bEcneax2Oh99xi939KeDNnS4vIt2l0ptIIZTsIoVQsosUQskuUgglu0ghuj+UdNQ9LxteNyqfNS2tZSXBqNyRDVncyy6w2XTQWbzpc5s6tT42MBAvO29eGH5pUtB9FnjxxfpY1jN3UlZybDJMNXQ87TIQ73MNJS0iSnaRQijZRQqhZBcphJJdpBBKdpFCKNlFCvHbNZR0E1mdfdy4zpfPavzZkMhZV8ys7bt21ceyaxeymm4me27RUNJz54aLbhkb19GfeSzedFRnz8rkhxwSvx8OPzy+RsCyayOixkUx0JTNIhJTsosUQskuUgglu0ghlOwihVCyixRCyS5SiO73Z49qyjt3xss3mTY5i2f14qgw27ROng1bnA2ZHNWys6Gkoxo9wKRJcTwb7nn27PpYUmd/KqmjP/FEHG9SZ585M45nsyofc+Sc+AHRGATZdNFZDb+GPtlFCqFkFymEkl2kEEp2kUIo2UUKoWQXKYSSXaQQ3a2zjxkT122z4mUkm/43q4Vn9eKsDh+ZMSOOZ9cAZGO7R9MuZ9curFsXx6dPj+NzknpyUEtf/8LEcNGsjv7II3E8Klc3rbNnu3XChLg//EC037I6+9at9bEm/dnN7BozW2dmD7TdN8PMfmlmj1e/k3eEiPTacA7jrwXOGHTf54A73H0+cEf1t4j0sTTZ3f0u4LlBd58JXFfdvg44a2SbJSIjrdMTdLPdfXV1ew1QewG0mS02s6VmtnT9c4P/Z4hItzQ+G+/uDtSOWujuS9x9obsvnJWdqBKRUdNpsq81swGA6ndySldEeq3TZL8NOL+6fT5w68g0R0RGS1pnN7MbgdOAmWa2AvgicDlwk5l9FHgaOHtYWzvgAIjmvc7Gy45kdfRkvu3te+I6/cZn6mN5s+N68qFzjg/j09+VzBUeXWMweXK87MqVcXxi3HYWLIjjxx5bG3r04XjRRx+N448l/d2jYQLycePjeCbb7dNOqp+3/uDsmo+oyB+MnZAmu7ufUxN6d7asiPQPXS4rUgglu0ghlOwihVCyixRCyS5SiO53cY1KYFlXz0jSBXUHcWlt9eowzIoV9bHsKuBspOikKsj8+YeH8TcuWlQfzIahXr48jmevyZveFIZX76yvYT3+eLzqp56K41nTo/2ePa2oFynA+PFxPOsiOxDM+HzMocnC0RMLSm/6ZBcphJJdpBBKdpFCKNlFCqFkFymEkl2kEEp2kUJ0t85+wAH4QfX1btu7t+NV7xqbdFFdGy+fzWwc1XSzXqJZHX5cPOpwuv6NG+uf+4knnhouOznropq8JtsOmhXGl/2qPpZ1Uc3q6NG1DxCPyJzV2bNZkbNeqNn7acOG+ti8eXH/2DHRNSWqs4uIkl2kEEp2kUIo2UUKoWQXKYSSXaQQSnaRQnS3zt5UUBzNhnPOZj3OZi6Oat1Zv+us5rpnTxx/9tk4vnFjfSx7XvPmxWMmZ0MuZ8/tvvvqY9l+W5tcG5FdvxC9J7LnlY1Mno1R0CSezdg8OWp8kymbReS1QckuUgglu0ghlOwihVCyixRCyS5SCCW7SCG6W2d3x3bvqo9HU9Gm4oG8s1Xv2BHHozr9pk3xslm9OKurbt4cx6N6clbvzfqEJ8Pxp2174on62KpVzdad7beoT3rQ7bv/RZ3xm9TZzewaM1tnZg+03Xepma00s2XVTzBLgYj0g+H8f7sWOGOI+7/u7idUP7ePbLNEZKSlye7udwHJhYki0u+afHO50Mzuqw7zp9c9yMwWm9lSM1u6Php4S0RGVafJfiVwDHACsBr4at0D3X2Juy9094WzstnuRGTUdJTs7r7W3fe4+17gKuDkkW2WiIy0jpLdzNonnP0A8EDdY0WkP6R1djO7ETgNmGlmK4AvAqeZ2QmAA8uBjw9ra3v3xoXfbFLsoOg7YcbEcNGsf3I2dnskG+4+q/Fn9eImlx+4x/EGuxxoVsfP+qM3qaNDXEvP3g/Z8540KY5PmdL5+rO+9ukbrkaa7O5+zhB3X93R1kSkZ36bryMSkf2gZBcphJJdpBBKdpFCKNlFCtHdLq579sS1nmy856DWMiaphUycGE/pnJVampRKmk4PnJXeshJWJBuCOytRZctHV0hnZbsmpTWIX5esdDZjRhzPLgbNlp8czMo8fmwytni0Y4Jaqz7ZRQqhZBcphJJdpBBKdpFCKNlFCqFkFymEkl2kEN2vs0e19Oefz5evk4w7PG1aXGc/9NB403Pm1Mey0bayIZGzOnpWy47q+FlvyGzdTbvvNhnOuen1C1Et/bDD4mWPPDKOz5sXx+fOjeMDA0GwyRjawQumT3aRQijZRQqhZBcphJJdpBBKdpFCKNlFCqFkFylEd+vse/fGdfYmYyqPGRMuOinpwHzkkUEHY+K+141mmqb5tMiRrFadDaGdxbNaedZvPJL1pc/q7FGf8qyOftxxzeJHHx3HD5kSTF3+zOZ44egNp/7sIqJkFymEkl2kEEp2kUIo2UUKoWQXKYSSXaQQw5my+QjgemA2rSmal7j7N8xsBvAjYB6taZvPdvdN4cr27o1r6VnBOupcnfWFX7cuDA/MTQrSC8bXhrJadjZ9b9aXPusvH/VJz/qjZ3XyTJO++Flf+kxWh4/2a1ZnP/bYOH788XE8Wz+rgvdjNqB+hy/4cF7q3cDF7r4AOAW4wMwWAJ8D7nD3+cAd1d8i0qfSZHf31e5+b3V7G/AwcDhwJnBd9bDrgLNGqY0iMgL26yDOzOYBbwF+Dcx299VVaA2tw3wR6VPDTnYzmwT8BLjI3V8xYZu7O63v80Mtt9jMlprZ0vVNLvIWkUaGlexmNo5Wot/g7j+t7l5rZgNVfAAY8oyDuy9x94XuvnDWtGkj0GQR6USa7GZmwNXAw+7+tbbQbcD51e3zgVtHvnkiMlKG08X17cB5wP1mtqy67xLgcuAmM/so8DRwdrom97gWlNWBoviuoMsg5OWMVavC8EAwR+/kt8TdY6NhqCEv06xZE8ejb0dZeatJtRPi0b2zeLbuJkNFA8yaVR876qh42Wyo6CMOS574quRFi+bZzrp6Rzs16OKaJru7/wqwmvC7s+VFpD/oCjqRQijZRQqhZBcphJJdpBBKdpFCKNlFCtHdoaQzTfpbZgXfaAhryAvSW7fWhiYlVwbOnxOMaQwMDMR1+tWrw3BYsm0y+y/EUy43lb3c4+t7FQMwdWocj7q4Ztc+TD1wR/yAVUm/4y1b4nj0fsveyx3SJ7tIIZTsIoVQsosUQskuUgglu0ghlOwihVCyixSiu3V2s7i42qTOnnWOzjpuZ3X4SFTohnQs6EnJWNLz58VF4U0z6+dVzursWTf/rM6evWTRMNvZENxN6+yTJwT16vQ1q7+uAhjdCxSS6cc7rcPrk12kEEp2kUIo2UUKoWQXKYSSXaQQSnaRQijZRQrR/f7s0WDgTersWR09q3tm/dmj5ZtuO5Psl+lBnf6gg+pr8JCXi7PLF7JaeTSt8vgD68c4B/LGZfENwUUE2RTfL70Ux5sOet8kD6L3k9WN+q5PdpFiKNlFCqFkFymEkl2kEEp2kUIo2UUKoWQXKURaZzezI4DrgdmAA0vc/RtmdinwMWB99dBL3P32ZGWj12c9q2U3rYVH8SZ1UcjbltWTgzHtD04mMT94RlIobzpOwAvB9QsbkufVdHL5aL9lr8muXXF8XHz9QqP3ebZs9prUGM5FNbuBi939XjObDNxjZr+sYl9393/saMsi0lVpsrv7amB1dXubmT0MHD7aDRORkbVfxxpmNg94C/Dr6q4Lzew+M7vGzKbXLLPYzJaa2dL12RhJIjJqhp3sZjYJ+AlwkbtvBa4EjgFOoPXJ/9WhlnP3Je6+0N0XzkrmRBOR0TOsZDezcbQS/QZ3/ymAu6919z3uvhe4Cjh59JopIk2lyW5mBlwNPOzuX2u7f6DtYR8AHhj55onISBnO2fi3A+cB95vZsuq+S4BzzOwEWuW45cDH0zW5d1w2AOJSS9MyTVaKGaVpdIe17qxtWWku0rRsmO33qG2j/ZplyzfRtPQWdEVNResO1jucs/G/AoZaQ1xTF5G+oivoRAqhZBcphJJdpBBKdpFCKNlFCqFkFylE94eSbjKscpPhnLOhgT0Z1jiqi2Y11SyeTdHbpPttkxo85Pt1x444Hu33bNmm1x9E13Q0mWt6OMuPpg63rU92kUIo2UUKoWQXKYSSXaQQSnaRQijZRQqhZBcphHlWXx7JjZmtB55uu2smsKFrDdg//dq2fm0XqG2dGsm2HeXus4YKdDXZX7Vxs6XuvrBnDQj0a9v6tV2gtnWqW23TYbxIIZTsIoXodbIv6fH2I/3atn5tF6htnepK23r6nV1EuqfXn+wi0iVKdpFC9CTZzewMM3vUzJ4ws8/1og11zGy5md1vZsvMbGmP23KNma0zswfa7pthZr80s8er30POsdejtl1qZiurfbfMzBb1qG1HmNl/mtlDZvagmX2qur+n+y5oV1f2W9e/s5vZGOAx4A+AFcDdwDnu/lBXG1LDzJYDC9295xdgmNk7ge3A9e7+xuq+vweec/fLq3+U0939s33StkuB7b2exruarWigfZpx4CzgI/Rw3wXtOpsu7LdefLKfDDzh7k+5+07gh8CZPWhH33P3u4DnBt19JnBddfs6Wm+WrqtpW19w99Xufm91exuwb5rxnu67oF1d0YtkPxx4tu3vFfTXfO8O/MLM7jGzxb1uzBBmu/vq6vYaYHYvGzOEdBrvbho0zXjf7LtOpj9vSifoXu0d7n4i8D7ggupwtS956ztYP9VOhzWNd7cMMc34y3q57zqd/rypXiT7SuCItr/nVvf1BXdfWf1eB9xM/01FvXbfDLrV73U9bs/L+mka76GmGacP9l0vpz/vRbLfDcw3s9eZ2YHAh4DbetCOVzGzidWJE8xsIvAe+m8q6tuA86vb5wO39rAtr9Av03jXTTNOj/ddz6c/d/eu/wCLaJ2RfxL4q160oaZdRwO/qX4e7HXbgBtpHdbtonVu46PAIcAdwOPAvwMz+qht3wPuB+6jlVgDPWrbO2gdot8HLKt+FvV63wXt6sp+0+WyIoXQCTqRQijZRQqhZBcphJJdpBBKdpFCKNlFCqFkFynE/wPgfh/q3qHshAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAQ8AAAEICAYAAABBKnGGAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/MnkTPAAAACXBIWXMAAAsTAAALEwEAmpwYAAAdRklEQVR4nO3de5RdZZnn8e+TSyWpVG6QC0m4KRcdZGiQiNiAphtaEacb7HYYaWTAsQnMyBpZC6dlsXpG7VYXdjeKDg5tEAZoEcXmKm3TsrCFpi9IgZFbuBskITcggSSVkNszf+xd04ei9vOeenedS8Hvs1ZWqs5z3r3f/Z59ntrnvM/e29wdEZGRGtfpDojI2KTkISJZlDxEJIuSh4hkUfIQkSxKHiKSpWPJw8yuNrMvdWjdj5rZ4hYsd56Z3WNmm8zsktFe/pB1rTCzE1q5jlYys33NbLOZjQ+e42Z2YIvWf5yZPTHaz21iWT8zsz+qiF1kZt8ZjfW0QzJ5lDvp1vKFHvx3WTs6NxqGS1Lu/i53/1kLVrcEeBGY7u4XtGD5HWNmf2hmz5nZFjO7xcz2qLM8d/+1u/e5+65y+ZVvqlZw939093eM9nNr9ukr7t62Mair2SOP3y1f6MF/57W0V2PXfsBjnlF5Z2YTWtCfUWFm7wK+DZwBzAMGgP/T0U7V0M1jPaa4e/gPWAGcUBG7HLix4fevAncBBiwGVgIXUfw1XgGc3vDcq4EvlT/PAm4H1gMbyp/3bnjuz4A/A/4J2AT8BJjdEP8hsAZ4BbgHeFf5+BJgB7Ad2Az8aOg2AZOAS4EXyn+XApPK2OA2XACsA1YDn6wYi6uHrOuEJpf9ubLvf12x3LOB5eV2Pwa8e5htOAr4F2Bj2cfLgJ4yZsDXy/6/CjwMHFrGTiqXuQlYBXy2og9fAb7X8PsB5XZOG+a5XwT+d/nzRGAL8Bfl71OAbcAewP6AAxOALwO7ythm4LLy+Q6cCzxVbtu3AKvo44jGevCxhvbvBn5RjsUPgR/wb/vn0OeuAD4LPESxz/0AmDyCffmPKrbhC8B3y58Hx+eTwPPlss4F3lOud+PgODW8Jj8FXqJ4v10HzGxm+8r4fwCWlcv9Z+CwZG6omTx6gSeBs4Djyk7v3TDgO4GvlS/sByh2pHcMkzz2BP6gXN60cuNuGTLgzwAHU+yAPwMuboj/l7Ld4A60bLgkNdw2AX8K/CswF5hTDtyfDdmGP6V4I5xE8Vd3VpBAGl+QZpb91bLfU4ZZ3n+keFO/hyIJHAjsN8w2HAkcTfFG3J8i2Zxfxj4EPADMLJfx74D5ZWw1cFzDTv/uiu26FfjckMc2A0cO89zfBh4uf/7N8nW7ryH2yyFvjglVb6oyfnvZ930p3pAnVvRxRGNNQ0IAeoDngM+Ur/PvUyTHKHn8HFhAkQiXA+eOYF8eSfL4K2Ay8EGK5HpLuY0LKf4gfKB8/oHA75TbN4fij+ilTW7fEeWy3guMB84st3HSaCSPzRQZafDf2Q3x9wIvl507reHxwRdsasNjNwD/s+pN3fC8w4ENQwb8Txp+/2/AHRVtZ5aDPqPJ5PEMcFJD7EPAioZt2Eq5g5ePrQOObjJ5pJa9nfIvVsXy/h74TPC6VCX184GbG96wT1Ikl3FDnvdr4ByK72iifeAuyjdHw2OrgMXDPHfw6GJP4EKKI8+VQB/FUck3R5g8jh2y/1xY0ccRjTWvTx7vL7fHGuL3EiePTzT8/ufAX41gXx5J8ljYEH8J+E8Nv99I+UdimGWdAvyiye27nDLRNsSfoExMVf+a/c7jFHef2fDvisGAu98HPEvxV+2GIe02uPuWht+fo8jWr2NmvWb27fILuVcpsubMId/Er2n4eYBiZ8TMxpvZxWb2TNl2Rfmc2U1u24KyX1V9fMnddw637lFY9np33xa034fiTREys4PN7HYzW1OOwVcot9/df0rxMeZbwDozW2pm08umf0BxNPWcmd1tZu+rWMVmYPqQx6ZTHAK/jrtvBfopjjTfD9xNcRRwTPnY3antGWLY130YdcZ6AbDKy3dN6fmcfjW5L4/E2oaftw7z++B655nZ981sVbne7/Jv74HU9u0HXGBmGwf/Uex7b3ivNqo9VWtmn6Y4VHoB+OMh4VlmNrXh933L5w11AfAO4L3uPp1ip4MiIaX8IXAyxXcMMygydmNbH6ZNoxcoBi/VxxypZaf69jzFZ9mUy4HHgYPK8buIhrFz92+6+5HAIRQf/f5H+fj97n4yxWHwLbwx+Q96FPiNwV/M7O0Ur/mTFc+/m+KI5wjg/vL3D1F8N3NPRZvUWKTUGevVwEIza9zf9snsR519uY6vUGzjvy/X+4mGdaa273ngy0MOEHrd/fpohbWSh5kdDHyp7OgZwB+b2eFDnvZFM+sxs+MovpT54TCLmkaRRTeWU4CfH0E3pgGvURzS9VIMYqO1wNuD9tcDf2Jmc8xsNvC/KLL2aKi77O8AnzWzI61woJntN8zzplF8GbrZzN4J/NfBgJm9x8zea2aDX15uA3aXr8npZjbD3XeU7XdX9OM64HfLeoepFN8v3OTubzjyKN0N/GeKmaftlIfqwK/cfX1Fm9TrlFJnrP+F4gvb88xsgpmdTJHoctTZl+uYRnGE+IqZLaT8A1FKbd8VwLnlfmJmNtXMPmJm06IVNps8fjSkzuPmcrrru8BX3f2X7v4UxV+8vzazSWW7NRTfEr9AsQOe6+6PD7P8Syk+K79I8aXXHU32C+BaikPUVRQzB/86JH4lcEh5OHbLMO2/RHGY/RDFTMSD5WOjoday3f2HFDMR36P4iHALxRd0Q32W4ghsE8WO8IOG2PTysQ0U4/QS8Bdl7AxgRXmYey5wekU/Hi3j11F85zON4nunKv9M8XoOHmU8RpG0qo46AL4BfMzMNpjZN4PnVcke6zLB/T7wKYrv9D5B8UXtaxn9uJT8fbmOL1LMqLwC/C1w02AgtX3u3k8xq3cZxX7yNMUkSMhe/zFo9FhRwfldd9+7JSsQaSEzu4/iS9D/2+m+tMJobJ/ObREBzOwDZrZXeVh/JnAY7TtqaLlWbJ8q7UQK76D4wngqxezhx9x9dWe7NKpGffta9rFFRN7c9LFFRLK09WNL7+TJPrOv2fqqEbLENHoqXkfdo7dU+zrxusuuO25R+3E1/3a18jVttTr7TNB245YtDGzb1paBqZU8zOxEiim28cB33P3i6Pkz+/o4+yMfyV9htLNNnBi3HZ8o8EvFd+2qju2uKo9ooi3Ajh1xfOfO/Hhq2am+p96gqXGLXpdJk6pjzSw7FY+0+uN6naSd2l+C1/uKv/u7uO0oyk79Zbntt4APU1QunmZmh4xWx0Sku9U5bjwKeNrdny2LUL5PUSYuIm8BdZLHQl5/cs3K8rHXMbMlZtZvZv0D26JzwERkLGn5bIu7L3X3Re6+qHfy5FavTkTapE7yWMXrz8zbu3xMRN4C6iSP+4GDzOxtZtYDfBy4bXS6JSLdLnuq1t13mtl5FFe7Gg9cVZ59GTWqV5MQTRumphxTNQWp6bEWTa01FU8tP9r2bq4grjtN3Mp1162PSS0/iteZ+m/j612rzsPdfwz8eJT6IiJjiMrTRSSLkoeIZFHyEJEsSh4ikkXJQ0SyKHmISJb2X4awTp1Hq9YL6ZqCaO69Th1GM+23b4/jdU7JT607ddr7hMQuVOc17eT1OurUaUB6XKN4nbapfo0iHXmISBYlDxHJouQhIlmUPEQki5KHiGRR8hCRLO2dqjWLT42vM51ad1qvzinWqemx1HRpaip2YCCOvxbcj3nr1rhtqm+pqdqenjje21sdmzo1bpuSumJ+tK/V3V/qTKem4nUuD9FGOvIQkSxKHiKSRclDRLIoeYhIFiUPEcmi5CEiWZQ8RCRL+0/JT90CIbdtarmpef068bqX4U/VWqRqNbZsqY5t3hy3jWpEgBl77hnGTzv99DC+10EHVQdnzgzbXrJ0aRjflKqPSdWoRFp9avtYvVRBAx15iEgWJQ8RyaLkISJZlDxEJIuSh4hkUfIQkSxKHiKSpf11HnXUmd+uOzceta9TuwL160Ci631s2hQ2nb/HHmH89I9/PIz3zZgRxj26LUTC+48/Poz/7U9/Gi8gdVuISGrMU/tTon6mZXUebawBqZU8zGwFsAnYBex090Wj0SkR6X6jceTxW+7+4igsR0TGEH3nISJZ6iYPB35iZg+Y2ZLhnmBmS8ys38z6B7Ztq7k6EekWdT+2HOvuq8xsLnCnmT3u7vc0PsHdlwJLARbMnt0dV24VkdpqHXm4+6ry/3XAzcBRo9EpEel+2cnDzKaa2bTBn4EPAo+MVsdEpLvV+dgyD7jZinnlCcD33P2OUelVK9S5J0wz8TpSdR417uuyx6RJYdNPfuxjYXxiqoYl1bdw4fF9V4485pgwvmP69DD+Tz//eWVsS3QNFEhvV93ankhqX2vl/WhGIDt5uPuzwG+MYl9EZAzRVK2IZFHyEJEsSh4ikkXJQ0SyKHmISJbuuvVCasoymm6tOxWbEvU7dYn/Fp+yv//ee1fGPnjkkWHbialxS51Sn9q2aKq4ry9sarNmhfGjTvhQGB8/a35l7NZbbw7b9kyLp7hr3dYB4v21zjRxG6dqdeQhIlmUPEQki5KHiGRR8hCRLEoeIpJFyUNEsih5iEiWsXXrhTpS89+pefuofapWInHqeeoWAb1Tp4bxExYvrozN7+2N1526RUBKatui9SdOqX9t3JQw/spL8aqnTduvMjZ+/Oyw7aub42t6T58e16gka5ai+plUbU0bazkiOvIQkSxKHiKSRclDRLIoeYhIFiUPEcmi5CEiWZQ8RCRLe+s8zOKahl270u2rpK4rkYqn6hWiOpBEnUZyXj5x/Ybf+/CHw/jC6LoY69bF606NeU9PGF69YUMYf/FXv6oOJsZtz732COMbNsTjunPnjMrYYYf9dtj29ttvCOOpYZs1I1EHsmNHXgzSNSRtoiMPEcmi5CEiWZQ8RCSLkoeIZFHyEJEsSh4ikkXJQ0SytL/OI6obSF3HIFLzmhlMnlxv+ZEp8XUpxiWuJTLtbW+Llx/N+6cKErZsCcP3P/xwGL/ziSfC+I4DDqiMDTywPGx73HHVdRoA06e/PV53UC4xY8aBYdv168NwclgnToz/LvdFtTmpOo/oGizddN8WM7vKzNaZ2SMNj+1hZnea2VPl//HdeUTkTaeZjy1XAycOeexC4C53Pwi4q/xdRN5CksnD3e8BXh7y8MnANeXP1wCnjG63RKTb5X5hOs/dV5c/rwHmVT3RzJaYWb+Z9Q9s3Zq5OhHpNrVnW9zdgcorALv7Undf5O6LehNfHIrI2JGbPNaa2XyA8v/EqZsi8maTmzxuA84sfz4TuHV0uiMiY0WyzsPMrgcWA7PNbCXweeBi4AYz+xTwHHBqU2sbNw4mTaqOp+6dEknVcUTrBbbvjttvfaU6lipPGTcuXvfvnXJaGF/wnkPjFaxcWR1LbPevfvnLMH7XI4+E8R1z5oRx9tyzMvRiopbi29++MYyffvp/D+PRuKfqNF6Mb9uSlLgMCpMXVPdtQqrmKOp8G+s8ksnD3av27ONHuS8iMoaoPF1Esih5iEgWJQ8RyaLkISJZlDxEJEv7T8mPpg5Tt0eIJE6Z35nY1M2b48W/+mp1LFV139PTG8YXLHhfGF+/LV7+nIMOqoxtT4zL3//N34Tx1/baK1753LlhePOu6qril4eeMTXECy8MhPH+/qfD+MEHv6syZhbvD7NnV48pwIYNT4Xx3vglZ9q06tisqYnG0a06uumUfBGR4Sh5iEgWJQ8RyaLkISJZlDxEJIuSh4hkUfIQkSztr/OITp33yguSJe0eF2/KQKKOI1XnsXFjdWzTprjtkUceEcajM+qbWf7AQPW2r98Y10qsnT49XnhUkABsnxDXJKz5dXXspZfiVUdjDvDkk2vC+Pz51XUe48bF9RBm8Xan9pdUfCB4WWbOjM/nt6h2R3UeItLtlDxEJIuSh4hkUfIQkSxKHiKSRclDRLIoeYhIlvbWedQVXO8jdfuD6BIIAFu2xPGo1mLq1L3Dtocd9lthfO3aeG4+upYIwLp11Z279tqbw7YzZ8Z38UvdDSNVz7B2bXVsw4Z6y54xI77WyLZt1eOa2q7t2+PXJLU/1Ynv2BG37alzi5JRpCMPEcmi5CEiWZQ8RCSLkoeIZFHyEJEsSh4ikkXJQ0SytL/OY/fu6tiuXTUWHM99pxadqhN57bXq2Pz588K2AwNxzcCOHfF1TLYl7tuybNkvKmOrVsVtUzUkidu+JPsW3ZsldZ2SHTvi13TSpKlhfNeu6nFNXfbCa1xbpuWi+xt10/U8zOwqM1tnZo80PPYFM1tlZsvKfye1tpsi0m2a+dhyNXDiMI9/3d0PL//9eHS7JSLdLpk83P0eIHFjQBF5q6nzhel5ZvZQ+bFmVtWTzGyJmfWbWf9A6gQSERkzcpPH5cABwOHAauCSqie6+1J3X+Tui3qnxl9wicjYkZU83H2tu+9y993AFcBRo9stEel2WcnDzOY3/PpR4JGq54rIm1OyzsPMrgcWA7PNbCXweWCxmR0OOLACOKepte3eHV/IICqmgLDoYOKUuCAhul0MxFPnKamSgFSNSer6Dan2zzyzvjIW1VlArSEH0tetiOpItm6N25511tlhfObMuL4mKnlwjwf96aeXh/HE7WyYNCmOR+OavFxHtMO1sT4lmTzc/bRhHr6yBX0RkTFE5ekikkXJQ0SyKHmISBYlDxHJouQhIlnae0q+ezw3mJo3DE7nt56esOnEifGmpqYk60ytpaaBo6sUQHqq9tFHn81edupSBKkp7lT7vr65lbETTjg+bDt3bjwVm5rijl+XRGPieeTe3rj1lPiOFkS76/hxienW1IvaJjryEJEsSh4ikkXJQ0SyKHmISBYlDxHJouQhIlmUPEQkS3fVeaQm7qPTjRP3AJg8uS+Mpy5y1hc0T7VN1QTUvS1EVEdSd9nz5u0Vxt/3vmMT7fetjE2cGJ/Xnir7SdXP7NxZXavxox9dH7adMSNe9syZcXz69DgentKfup9F9D5p4yn5OvIQkSxKHiKSRclDRLIoeYhIFiUPEcmi5CEiWZQ8RCRLe+s8UrdeqHMPgug6+0BP4nofM2bE8ajbe+4ZNk3GU9cSSU3777tv9QqCO4ECcNxxvxnG3/nOg8O4WbwLbd9eXXeQum1D6loi27bF19x48MG7K2Nbt64M286eHa879ZrOioedKZOCa3K8knjBo/eB6jxEpNspeYhIFiUPEcmi5CEiWZQ8RCSLkoeIZFHyEJEsyToPM9sHuBaYBziw1N2/YWZ7AD8A9gdWAKe6+4ZwYe5xLUfq4hPRHHaqRmTLljDcNz1x85U51fG5c+P7aCxYEC86dT2QgYE4fv75nwiicf1L6n42KalbiNR5ubdsiTf89tvja3Js2FBdy7FXfJmSZB3HnDlxPHU9EDYF+2OqACZ1EZY2aebIYydwgbsfAhwNfNrMDgEuBO5y94OAu8rfReQtIpk83H21uz9Y/rwJWA4sBE4Grimfdg1wSov6KCJdaETfeZjZ/sARwH3APHdfXYbWUHysEZG3iKaTh5n1ATcC57v7q40xd3eK70OGa7fEzPrNrH8gdVFKERkzmkoeZjaRInFc5+43lQ+vNbP5ZXw+sG64tu6+1N0Xufui3kmTRqPPItIFksnDzAy4Elju7l9rCN0GnFn+fCZw6+h3T0S6VTPzdMcAZwAPm9my8rGLgIuBG8zsU8BzwKlNrTGabk2cVh/GU/N+qemvTZvCcF9w/4Q1a34Rtp0xY3EiHl+nf/PmMMy2bdXn9Kdm9erMjhfi12zbtupxf/zxJ8K2t956cxifMCGeJ164sDpW+9YK0xIDsynxom0NLieQKjuI5sfbeEp+Mnm4+71U7yHHj253RGSsUIWpiGRR8hCRLEoeIpJFyUNEsih5iEgWJQ8RydLeWy+kpOo8Iqn57VRpfKogImjfM3ly2PTO268N42edc14YT5SghCUDqds21CkpAFi/fm0Yv/fef6iMPf748rDt3LnxuhPDHl7qoK8vbjtpfGJ/2JS4TkJq4KP9LTXoXUJHHiKSRclDRLIoeYhIFiUPEcmi5CEiWZQ8RCSLkoeIZGl/nUdUy9HKOo+61/uIRIUWwJpHHw3j11729TB+6NFHh/FDDl1UGRs3Lq5X6O9/KIw/+OCyML5x48YwvnNndb1D6vYH4xN3w0jVefRMDPaJxGvGQKIuqG6BTGRc4m96l9SB6MhDRLIoeYhIFiUPEcmi5CEiWZQ8RCSLkoeIZFHyEJEs7a3zMIvnsOvUeaTqOFJz46nreUTtE+venVj3yifi+5esXLUqjN9xxx2VsZ27478PqXKFVPnMlClxfEKwh6XqOJKdS8UHgtqdVF1P3RvapGo16rwPWlUrNUI68hCRLEoeIpJFyUNEsih5iEgWJQ8RyaLkISJZlDxEJEuyzsPM9gGuBeYBDix192+Y2ReAs4H15VMvcvcfJ9fYqmt2pOo46taBRPHUNtXtW6qeIbinzISenrDphCmJYou610nZEdTPDCS2K1V7U2fc6r4mqSKVOvt5qm3qNWmTZorEdgIXuPuDZjYNeMDM7ixjX3f3v2xd90SkWyWTh7uvBlaXP28ys+XAwlZ3TES624i+8zCz/YEjgPvKh84zs4fM7Cozm1XRZomZ9ZtZ/0DqFnwiMmY0nTzMrA+4ETjf3V8FLgcOAA6nODK5ZLh27r7U3Re5+6Le1EUnRWTMaCp5mNlEisRxnbvfBODua919l7vvBq4AjmpdN0Wk2ySTh5kZcCWw3N2/1vD4/IanfRR4ZPS7JyLdqpnZlmOAM4CHzWxZ+dhFwGlmdjjF9O0K4Jym1lhnmimaPqs7rZeaumvl9FidaWJIT+VG6k4zp8Y96lurX7NU+1Zq5anxbTztPtLMbMu9wHC9Tdd0iMiblipMRSSLkoeIZFHyEJEsSh4ikkXJQ0SyKHmISJb23nrBPT03H6lx+4Pal9Kvc7n7VDx1mf46dSB1akCg/uUCovZ17/tQpzYn9Zq08pT7urqkzkNHHiKSRclDRLIoeYhIFiUPEcmi5CEiWZQ8RCSLkoeIZDFv42XczWw98FzDQ7OBF9vWgZFR30auW/sFb52+7efuc0ZpWaG2Jo83rNys390XdawDAfVt5Lq1X6C+tYI+tohIFiUPEcnS6eSxtMPrj6hvI9et/QL1bdR19DsPERm7On3kISJjlJKHiGTpSPIwsxPN7Akze9rMLuxEH6qY2Qoze9jMlplZf4f7cpWZrTOzRxoe28PM7jSzp8r/h71HcIf69gUzW1WO3TIzO6lDfdvHzP7BzB4zs0fN7DPl4x0du6BfXTFuI9X27zzMbDzwJPA7wErgfuA0d3+srR2pYGYrgEXu3vGCIjN7P7AZuNbdDy0f+3PgZXe/uEy8s9z9c13Sty8Am939L9vdnyF9mw/Md/cHzWwa8ABwCnAWHRy7oF+n0gXjNlKdOPI4Cnja3Z919+3A94GTO9CPrufu9wAvD3n4ZOCa8udrKHa+tqvoW1dw99Xu/mD58yZgObCQDo9d0K8xqRPJYyHwfMPvK+muAXTgJ2b2gJkt6XRnhjHP3VeXP68B5nWyM8M4z8weKj/WdOQjVSMz2x84AriPLhq7If2CLhu3ZugL0zc61t3fDXwY+HR5eN6VvPjM2U1z7ZcDBwCHA6uBSzrZGTPrA24Eznf3VxtjnRy7YfrVVePWrE4kj1XAPg2/710+1hXcfVX5/zrgZoqPWd1kbfnZefAz9LoO9+f/c/e17r7L3XcDV9DBsTOziRRv0Ovc/aby4Y6P3XD96qZxG4lOJI/7gYPM7G1m1gN8HLitA/14AzObWn6RhZlNBT4IPBK3arvbgDPLn88Ebu1gX15n8I1Z+igdGjszM+BKYLm7f60h1NGxq+pXt4zbSHWkwrSciroUGA9c5e5fbnsnhmFmb6c42oDithTf62TfzOx6YDHFKdtrgc8DtwA3APtSXN7gVHdv+xeXFX1bTHHo7cAK4JyG7xja2bdjgX8EHgYG789wEcX3Cx0bu6Bfp9EF4zZSKk8XkSz6wlREsih5iEgWJQ8RyaLkISJZlDxEJIuSh4hkUfIQkSz/D6OiNvx+XPYvAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Explanation for class 1\n",
    "plot_image(explanation[1], heatmap_cmap='bwr', show_plot=False)\n",
    "plt.title('Explanation for class 0')\n",
    "plot_image(explanation[1], original_data=image[0], heatmap_cmap='bwr', data_cmap='gray', show_plot=False)\n",
    "plt.title('Explanation for class 0 with original image');"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "563c1043-d9b8-4baa-8454-a7afefd59d37",
   "metadata": {},
   "source": [
    "This looks very similar to the explanation for class 0, but with inverted colours as expected: any pixel in favour of zero is against one, and vice-versa."
   ]
  }
 ],
 "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": "3.9.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}