explain_mnist.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "390a7280",
   "metadata": {},
   "source": [
    "<a href=\"https://colab.research.google.com/github/sdsc-bw/MCXAI/blob/main/explain_mnist.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6d6745d7",
   "metadata": {},
   "source": [
    "# Explain a Model of Covertype Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "ce81b040",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from tensorflow import keras\n",
    "from tensorflow.keras import layers\n",
    "import matplotlib.pyplot as plt\n",
    "from matplotlib import cm\n",
    "from PIL import Image, ImageDraw, ImageFont"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "5508782e",
   "metadata": {},
   "outputs": [],
   "source": [
    "from explainer import Explainer\n",
    "import util"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b7b28aed",
   "metadata": {},
   "source": [
    "## Load and Prepare Dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1bd143d2",
   "metadata": {},
   "source": [
    "We use the [keras mnist dataset](https://keras.io/examples/vision/mnist_convnet/). The code for the dataset preparation and modelling are based on this website."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "8834e928",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "x_train shape: (60000, 28, 28, 1)\n",
      "60000 train samples\n",
      "10000 test samples\n"
     ]
    }
   ],
   "source": [
    "# Model / data parameters\n",
    "num_classes = 10\n",
    "input_shape = (28, 28, 1)\n",
    "\n",
    "# the data, split between train and test sets\n",
    "(x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data()\n",
    "\n",
    "# Scale images to the [0, 1] range\n",
    "x_train = x_train.astype(\"float32\") / 255\n",
    "x_test = x_test.astype(\"float32\") / 255\n",
    "# Make sure images have shape (28, 28, 1)\n",
    "x_train = np.expand_dims(x_train, -1)\n",
    "x_test = np.expand_dims(x_test, -1)\n",
    "print(\"x_train shape:\", x_train.shape)\n",
    "print(x_train.shape[0], \"train samples\")\n",
    "print(x_test.shape[0], \"test samples\")\n",
    "\n",
    "\n",
    "# convert class vectors to binary class matrices\n",
    "y_train = keras.utils.to_categorical(y_train, num_classes)\n",
    "y_test = keras.utils.to_categorical(y_test, num_classes)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2dc3919f",
   "metadata": {},
   "source": [
    "## Train Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "5edee0ab",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d (Conv2D)              (None, 26, 26, 32)        320       \n",
      "_________________________________________________________________\n",
      "max_pooling2d (MaxPooling2D) (None, 13, 13, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_1 (Conv2D)            (None, 11, 11, 64)        18496     \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 5, 5, 64)          0         \n",
      "_________________________________________________________________\n",
      "flatten (Flatten)            (None, 1600)              0         \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, 1600)              0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 10)                16010     \n",
      "=================================================================\n",
      "Total params: 34,826\n",
      "Trainable params: 34,826\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = keras.Sequential(\n",
    "    [\n",
    "        keras.Input(shape=input_shape),\n",
    "        layers.Conv2D(32, kernel_size=(3, 3), activation=\"relu\", padding='valid', data_format = 'channels_last'),\n",
    "        layers.MaxPooling2D(pool_size=(2, 2)),\n",
    "        layers.Conv2D(64, kernel_size=(3, 3), activation=\"relu\", padding='valid', data_format = 'channels_last'),\n",
    "        layers.MaxPooling2D(pool_size=(2, 2)),\n",
    "        layers.Flatten(),\n",
    "        layers.Dropout(0.5),\n",
    "        layers.Dense(num_classes, activation=\"softmax\"),\n",
    "    ]\n",
    ")\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "7c56cf73",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "422/422 [==============================] - 13s 31ms/step - loss: 0.3593 - accuracy: 0.8898 - val_loss: 0.0867 - val_accuracy: 0.9773\n",
      "Epoch 2/15\n",
      "422/422 [==============================] - 13s 30ms/step - loss: 0.1169 - accuracy: 0.9641 - val_loss: 0.0592 - val_accuracy: 0.9843\n",
      "Epoch 3/15\n",
      "422/422 [==============================] - 13s 30ms/step - loss: 0.0878 - accuracy: 0.9725 - val_loss: 0.0486 - val_accuracy: 0.9862\n",
      "Epoch 4/15\n",
      "422/422 [==============================] - 13s 30ms/step - loss: 0.0736 - accuracy: 0.9776 - val_loss: 0.0420 - val_accuracy: 0.9885\n",
      "Epoch 5/15\n",
      "422/422 [==============================] - 12s 29ms/step - loss: 0.0660 - accuracy: 0.9792 - val_loss: 0.0379 - val_accuracy: 0.9902\n",
      "Epoch 6/15\n",
      "422/422 [==============================] - 12s 29ms/step - loss: 0.0577 - accuracy: 0.9812 - val_loss: 0.0391 - val_accuracy: 0.9888\n",
      "Epoch 7/15\n",
      "422/422 [==============================] - 12s 29ms/step - loss: 0.0525 - accuracy: 0.9835 - val_loss: 0.0384 - val_accuracy: 0.9887\n",
      "Epoch 8/15\n",
      "422/422 [==============================] - 13s 30ms/step - loss: 0.0488 - accuracy: 0.9845 - val_loss: 0.0351 - val_accuracy: 0.9900\n",
      "Epoch 9/15\n",
      "422/422 [==============================] - 13s 30ms/step - loss: 0.0460 - accuracy: 0.9852 - val_loss: 0.0332 - val_accuracy: 0.9913\n",
      "Epoch 10/15\n",
      "422/422 [==============================] - 13s 30ms/step - loss: 0.0436 - accuracy: 0.9858 - val_loss: 0.0316 - val_accuracy: 0.9910\n",
      "Epoch 11/15\n",
      "422/422 [==============================] - 13s 30ms/step - loss: 0.0411 - accuracy: 0.9865 - val_loss: 0.0329 - val_accuracy: 0.9908\n",
      "Epoch 12/15\n",
      "422/422 [==============================] - 13s 30ms/step - loss: 0.0408 - accuracy: 0.9870 - val_loss: 0.0295 - val_accuracy: 0.9922\n",
      "Epoch 13/15\n",
      "422/422 [==============================] - 13s 30ms/step - loss: 0.0375 - accuracy: 0.9878 - val_loss: 0.0301 - val_accuracy: 0.9915\n",
      "Epoch 14/15\n",
      "422/422 [==============================] - 13s 30ms/step - loss: 0.0367 - accuracy: 0.9876 - val_loss: 0.0298 - val_accuracy: 0.9917\n",
      "Epoch 15/15\n",
      "422/422 [==============================] - 13s 30ms/step - loss: 0.0357 - accuracy: 0.9881 - val_loss: 0.0280 - val_accuracy: 0.9920\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x1e60ef9a1f0>"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "batch_size = 128\n",
    "epochs = 15\n",
    "\n",
    "model.compile(loss=\"categorical_crossentropy\", optimizer=\"adam\", metrics=[\"accuracy\"])\n",
    "\n",
    "model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, validation_split=0.1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "cfb98c6e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test loss: 0.027469519525766373\n",
      "Test accuracy: 0.9900000095367432\n"
     ]
    }
   ],
   "source": [
    "score = model.evaluate(x_test, y_test, verbose=0)\n",
    "print(\"Test loss:\", score[0])\n",
    "print(\"Test accuracy:\", score[1])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "321ce30b",
   "metadata": {},
   "source": [
    "## Pick a Sample for Explanation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "fa3579a2",
   "metadata": {},
   "outputs": [],
   "source": [
    "sample = x_test[0].squeeze()\n",
    "target_label = np.argmax(y_test[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "2f646018",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(sample, cmap='gray')\n",
    "plt.axis(\"off\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "7aff1ad7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "7"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "target_label"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "94099f0b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[3.9499382e-09, 6.5734279e-11, 8.3598006e-06, 2.8119864e-06,\n",
       "        1.0289834e-12, 2.9822167e-10, 3.4116981e-16, 9.9998820e-01,\n",
       "        2.0498392e-08, 6.3432844e-07]], dtype=float32)"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.predict(np.expand_dims(sample.reshape(28, 28, 1), axis=0))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d0b8d7fb",
   "metadata": {},
   "source": [
    "## Explain the Sample"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "23528283",
   "metadata": {},
   "outputs": [],
   "source": [
    "kernel_size = 3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "f553d2dc",
   "metadata": {},
   "outputs": [],
   "source": [
    "def predict(data):\n",
    "    if data.shape == (28, 28):\n",
    "        return model.predict(np.expand_dims(data.reshape(28, 28, 1), axis=0))\n",
    "    else:\n",
    "        return model.predict(data.reshape(-1, 28, 28, 1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "0a6ab18c",
   "metadata": {},
   "outputs": [],
   "source": [
    "explainer = Explainer(sample, predict, target_label, hide_value=0, kernel_size=3, max_depth=20, max_episodes=10000)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "b91a8361",
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2021-11-02 12:54:32,591 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 12:54:32,592 - agent - INFO - Episode:\t0\n",
      "2021-11-02 12:54:32,593 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:00:12,968 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:00:12,968 - agent - INFO - Episode:\t1000\n",
      "2021-11-02 13:00:12,969 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:05:26,785 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:05:26,786 - agent - INFO - Episode:\t2000\n",
      "2021-11-02 13:05:26,786 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:10:56,499 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:10:56,500 - agent - INFO - Episode:\t3000\n",
      "2021-11-02 13:10:56,500 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:15:37,861 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:15:37,862 - agent - INFO - Episode:\t4000\n",
      "2021-11-02 13:15:37,862 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:19:24,884 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:19:24,884 - agent - INFO - Episode:\t5000\n",
      "2021-11-02 13:19:24,885 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:22:23,725 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:22:23,725 - agent - INFO - Episode:\t6000\n",
      "2021-11-02 13:22:23,726 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:24:03,397 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:24:03,397 - agent - INFO - Episode:\t7000\n",
      "2021-11-02 13:24:03,398 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:25:15,626 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:25:15,627 - agent - INFO - Episode:\t8000\n",
      "2021-11-02 13:25:15,627 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:26:12,634 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:26:12,635 - agent - INFO - Episode:\t9000\n",
      "2021-11-02 13:26:12,636 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:26:56,960 - agent - INFO - Maximum number of episodes reached\n",
      "2021-11-02 13:26:57,102 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:26:57,102 - agent - INFO - Episode:\t0\n",
      "2021-11-02 13:26:57,103 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:27:34,378 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:27:34,378 - agent - INFO - Episode:\t1000\n",
      "2021-11-02 13:27:34,379 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:28:08,539 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:28:08,539 - agent - INFO - Episode:\t2000\n",
      "2021-11-02 13:28:08,540 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:28:37,121 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:28:37,121 - agent - INFO - Episode:\t3000\n",
      "2021-11-02 13:28:37,122 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:29:00,857 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:29:00,858 - agent - INFO - Episode:\t4000\n",
      "2021-11-02 13:29:00,858 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:29:13,985 - agent - INFO - Distribution stable after \t4500 episodes\n",
      "2021-11-02 13:29:13,986 - explainer - INFO - Classification game finished\n",
      "2021-11-02 13:29:14,255 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:29:14,256 - agent - INFO - Episode:\t0\n",
      "2021-11-02 13:29:14,257 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:29:57,979 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:29:57,979 - agent - INFO - Episode:\t1000\n",
      "2021-11-02 13:29:57,979 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:30:16,331 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:30:16,331 - agent - INFO - Episode:\t2000\n",
      "2021-11-02 13:30:16,332 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:31:04,218 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:31:04,218 - agent - INFO - Episode:\t3000\n",
      "2021-11-02 13:31:04,219 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:31:52,010 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:31:52,010 - agent - INFO - Episode:\t4000\n",
      "2021-11-02 13:31:52,011 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:32:38,460 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:32:38,461 - agent - INFO - Episode:\t5000\n",
      "2021-11-02 13:32:38,461 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:33:19,911 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:33:19,912 - agent - INFO - Episode:\t6000\n",
      "2021-11-02 13:33:19,913 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:33:56,839 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:33:56,839 - agent - INFO - Episode:\t7000\n",
      "2021-11-02 13:33:56,840 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:34:31,798 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:34:31,798 - agent - INFO - Episode:\t8000\n",
      "2021-11-02 13:34:31,799 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:35:03,723 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:35:03,724 - agent - INFO - Episode:\t9000\n",
      "2021-11-02 13:35:03,724 - agent - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX\n",
      "2021-11-02 13:35:35,626 - explainer - INFO - Misclassification game finished\n"
     ]
    }
   ],
   "source": [
    "explainer.explain()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "f96307d3",
   "metadata": {},
   "outputs": [],
   "source": [
    "ranks_1, ranks_2 = explainer.get_explanation()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "061b0b46",
   "metadata": {},
   "source": [
    "Now the features are ranked by their importance, if the explainer was able to change the prediction (change the class with the highest probability). The lower the rank of the feature, the more important is this feature."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "e4ed78e5",
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_masked(ranks, sample, scale, ksize, offset=0):\n",
    "    img = Image.fromarray(np.uint8(sample*255))\n",
    "    img = img.resize((28 * scale, 28 * scale), resample=Image.BOX)\n",
    "    img = img.convert('RGBA')\n",
    "    \n",
    "    t_img = Image.new('RGBA', (28 * scale, 28 * scale), (255, 0, 0, 0))\n",
    "    fnt = ImageFont.truetype('/Library/Fonts/arial.ttf', (scale - offset) * ksize)\n",
    "    d_img = ImageDraw.Draw(img)\n",
    "    fillcolor = \"white\"\n",
    "    shadowcolor = \"black\"\n",
    "    \n",
    "    for x in range(0, 28, ksize):\n",
    "        for y in range(0, 28, ksize):\n",
    "            number = ranks[x][y]\n",
    "            if number != 0:\n",
    "                text = str(int(number))\n",
    "                xt = x * scale\n",
    "                yt = y * scale\n",
    "                \n",
    "                # thin border\n",
    "                d_img.text((yt-3, xt), text, font=fnt, fill=shadowcolor)\n",
    "                d_img.text((yt+3, xt), text, font=fnt, fill=shadowcolor)\n",
    "                d_img.text((yt, xt-3), text, font=fnt, fill=shadowcolor)\n",
    "                d_img.text((yt, xt+3), text, font=fnt, fill=shadowcolor)\n",
    "\n",
    "                # thicker border\n",
    "                d_img.text((yt-3, xt-3), text, font=fnt, fill=shadowcolor)\n",
    "                d_img.text((yt+3, xt-3), text, font=fnt, fill=shadowcolor)\n",
    "                d_img.text((yt-3, xt+3), text, font=fnt, fill=shadowcolor)\n",
    "                d_img.text((yt+3, xt+3), text, font=fnt, fill=shadowcolor)\n",
    "                \n",
    "                d_img.text((yt, xt), text, font=fnt, fill=fillcolor)\n",
    "\n",
    "    for x in range(0, img.width, ksize * scale):\n",
    "        line = ((x, 0), (x, img.height))\n",
    "        d_img.line(line, fill=(255, 0, 0))\n",
    "        \n",
    "    for y in range(0, img.height, ksize * scale):\n",
    "        line = ((0, y), (img.width, y))\n",
    "        d_img.line(line, fill=(255, 0, 0))\n",
    "    \n",
    "    img.paste(t_img, (0, 0), t_img)\n",
    "    plt.axis(\"off\")\n",
    "    plt.imshow(img)\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "18ba1ddb",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_masked(ranks_1, sample, 15, kernel_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "551bdc4a",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_masked(ranks_2, sample, 15, kernel_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a6ae7007",
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "mcxai",
   "language": "python",
   "name": "mcxai"
  },
  "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
}