digits_mnist.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "digits_mnist.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true,
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/ymiftah/machine_learning/blob/master/digits_mnist.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "w_LVLABepUv0",
        "colab_type": "text"
      },
      "source": [
        "# Digits MNIST\n",
        "\n",
        "This notebook retrace various experiments with the MNIST dataset. A final estimator was built for a Kaggle submission"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "N5CBD2Ii6JTK",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from sklearn.datasets import fetch_openml\n",
        "import pandas as pd\n",
        "import matplotlib.pyplot as plt\n",
        "import matplotlib as mpl\n",
        "import numpy as np"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WhuRZiJqp2a4",
        "colab_type": "text"
      },
      "source": [
        "We fetch the data from openml database"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qZxuQued6lN2",
        "colab_type": "code",
        "outputId": "1502a554-942d-4e21-884d-c0842b041f9e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        }
      },
      "source": [
        "mnist = fetch_openml('mnist_784')\n",
        "X, y = mnist['data'], mnist['target']\n",
        "y = y.astype(np.uint8)\n",
        "print(y.shape)"
      ],
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(70000,)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LUhWS5jcqPMk",
        "colab_type": "text"
      },
      "source": [
        "Looking at a histogram for each class, we see that they are not evenly distributed, but the difference should be minor"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7nnuLswPfq1y",
        "colab_type": "code",
        "outputId": "8662856a-d01f-4096-ab94-b0a7e6b54c91",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 265
        }
      },
      "source": [
        "bins = np.arange(0, 9 + 1.5) - 0.5\n",
        "# then you plot away\n",
        "fig, ax = plt.subplots()\n",
        "_ = ax.hist(y, bins)\n",
        "ax.set_xticks(bins+0.5)\n",
        "plt.show()"
      ],
      "execution_count": 21,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYMAAAD4CAYAAAAO9oqkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nMWI3CQAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "E4DmZcQErRie",
        "colab_type": "text"
      },
      "source": [
        "Lets look at one of the digit and its label"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GrfZ4sXo68K6",
        "colab_type": "code",
        "outputId": "cf54ec4c-4cbb-46c4-c6e0-3d8e399f4185",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 281
        }
      },
      "source": [
        "digit = X[9]\n",
        "digit_im = digit.reshape(28, 28)\n",
        "plt.imshow(digit_im, interpolation='nearest', cmap=mpl.cm.binary)\n",
        "print('This is a %d' %int(y[9]))"
      ],
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "This is a 4\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAD4CAYAAAAq5pAIAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ntKgqRzBLlCUAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "z49xIlQY7TpW",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from sklearn.model_selection import train_test_split\n",
        "\n",
        "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3,\n",
        "                                                    random_state=42)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HyYFHYjJrcuy",
        "colab_type": "text"
      },
      "source": [
        "We start by scaling our data between 0 and 1"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BE5WuNdD7MCc",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "X_train /= 255\n",
        "X_test /= 255"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "F7-uZS1blfmX",
        "colab_type": "text"
      },
      "source": [
        "We also prepare the Image data generator for the Deep Neural Network trained later"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "J_PbXnBYu_Dm",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from keras.preprocessing.image import ImageDataGenerator\n",
        "# applying transformation to image\n",
        "train_gen = ImageDataGenerator(rotation_range=20, \n",
        "                               width_shift_range=0.1, \n",
        "                               shear_range=0.3, \n",
        "                               height_shift_range=0.08, \n",
        "                               zoom_range=0.08 )\n",
        "training_set = train_gen.flow(X_train.reshape((X_train.shape[0], 28, 28, 1)), y_train, batch_size=128)\n",
        "test_set= train_gen.flow(X_test.reshape((X_test.shape[0], 28, 28, 1)), y_test, batch_size=128)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MPn0h37qy24w",
        "colab_type": "code",
        "outputId": "04615ef7-ce26-4491-ae01-9f2b466c0756",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        }
      },
      "source": [
        "def forest_training(iterations, generator):\n",
        "    X = []\n",
        "    Y = []\n",
        "    for _ in range(iterations):\n",
        "        x, y = generator.next()\n",
        "        x = x.reshape((x.shape[0], 28*28*1))\n",
        "        X.append(x)\n",
        "        Y.append(y)\n",
        "    X = np.concatenate(X)\n",
        "    Y = np.concatenate(Y)\n",
        "    return X, Y\n",
        "\n",
        "Xf, Yf = forest_training(1500, training_set)\n",
        "print(Yf.shape)"
      ],
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(191928,)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lKvD-17-bYZz",
        "colab_type": "text"
      },
      "source": [
        "## Neural Network\n",
        "\n",
        "In this section we will fit a neural network."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-lBrtTeTba7J",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from __future__ import absolute_import, division, print_function, unicode_literals\n",
        "\n",
        "import tensorflow as tf\n",
        "\n",
        "from tensorflow.keras import datasets, layers, models\n",
        "\n",
        "def create_model():\n",
        "    # Define Model\n",
        "    model = models.Sequential()\n",
        "    model.add(layers.Conv2D(128, (3, 3), activation='relu', input_shape=(28, 28, 1)))\n",
        "    model.add(layers.MaxPooling2D((2, 2)))\n",
        "    model.add(layers.Conv2D(64, (3, 3), activation='relu'))\n",
        "    model.add(layers.MaxPooling2D((2, 2)))\n",
        "    model.add(layers.Conv2D(64, (3, 3), activation='relu'))\n",
        "    model.add(layers.Flatten())\n",
        "    model.add(layers.Dense(64, activation='relu'))\n",
        "    model.add(layers.Dense(10, activation='softmax'))\n",
        "\n",
        "    # Compile Model\n",
        "    model.compile(optimizer='adam',\n",
        "                  loss='sparse_categorical_crossentropy',\n",
        "                  metrics=['accuracy'])\n",
        "    return model"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AOO8HuYQb4QC",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "model = create_model()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dkpSmDUyc1QC",
        "colab_type": "code",
        "outputId": "8b5d67f3-5968-407d-cdd2-35df8aa33b52",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "model.fit_generator(training_set, \n",
        "                     steps_per_epoch=60000//128, \n",
        "                     validation_data= test_set, \n",
        "                     validation_steps=10000//128, \n",
        "                     epochs=20)"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.4118 - acc: 0.8693Epoch 1/20\n",
            "468/468 [==============================] - 33s 70ms/step - loss: 0.4114 - acc: 0.8694 - val_loss: 0.1558 - val_acc: 0.9512\n",
            "Epoch 2/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.1273 - acc: 0.9611Epoch 1/20\n",
            "468/468 [==============================] - 31s 67ms/step - loss: 0.1272 - acc: 0.9611 - val_loss: 0.0961 - val_acc: 0.9697\n",
            "Epoch 3/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0948 - acc: 0.9710Epoch 1/20\n",
            "468/468 [==============================] - 32s 68ms/step - loss: 0.0950 - acc: 0.9709 - val_loss: 0.0739 - val_acc: 0.9757\n",
            "Epoch 4/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0758 - acc: 0.9762Epoch 1/20\n",
            "468/468 [==============================] - 31s 67ms/step - loss: 0.0758 - acc: 0.9763 - val_loss: 0.0697 - val_acc: 0.9779\n",
            "Epoch 5/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0630 - acc: 0.9800Epoch 1/20\n",
            "468/468 [==============================] - 31s 67ms/step - loss: 0.0630 - acc: 0.9800 - val_loss: 0.0687 - val_acc: 0.9781\n",
            "Epoch 6/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0578 - acc: 0.9816Epoch 1/20\n",
            "468/468 [==============================] - 31s 67ms/step - loss: 0.0579 - acc: 0.9816 - val_loss: 0.0589 - val_acc: 0.9833\n",
            "Epoch 7/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0537 - acc: 0.9828Epoch 1/20\n",
            "468/468 [==============================] - 32s 68ms/step - loss: 0.0537 - acc: 0.9827 - val_loss: 0.0452 - val_acc: 0.9848\n",
            "Epoch 8/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0494 - acc: 0.9850Epoch 1/20\n",
            "468/468 [==============================] - 32s 68ms/step - loss: 0.0494 - acc: 0.9849 - val_loss: 0.0469 - val_acc: 0.9859\n",
            "Epoch 9/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0451 - acc: 0.9860Epoch 1/20\n",
            "468/468 [==============================] - 31s 66ms/step - loss: 0.0451 - acc: 0.9859 - val_loss: 0.0517 - val_acc: 0.9850\n",
            "Epoch 10/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0436 - acc: 0.9863Epoch 1/20\n",
            "468/468 [==============================] - 33s 70ms/step - loss: 0.0435 - acc: 0.9863 - val_loss: 0.0546 - val_acc: 0.9824\n",
            "Epoch 11/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0404 - acc: 0.9879Epoch 1/20\n",
            "468/468 [==============================] - 32s 68ms/step - loss: 0.0403 - acc: 0.9879 - val_loss: 0.0395 - val_acc: 0.9870\n",
            "Epoch 12/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0384 - acc: 0.9881Epoch 1/20\n",
            "468/468 [==============================] - 32s 68ms/step - loss: 0.0383 - acc: 0.9881 - val_loss: 0.0416 - val_acc: 0.9872\n",
            "Epoch 13/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0368 - acc: 0.9888Epoch 1/20\n",
            "468/468 [==============================] - 31s 67ms/step - loss: 0.0367 - acc: 0.9888 - val_loss: 0.0386 - val_acc: 0.9880\n",
            "Epoch 14/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0353 - acc: 0.9889Epoch 1/20\n",
            "468/468 [==============================] - 32s 68ms/step - loss: 0.0353 - acc: 0.9889 - val_loss: 0.0398 - val_acc: 0.9876\n",
            "Epoch 15/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0339 - acc: 0.9897Epoch 1/20\n",
            "468/468 [==============================] - 31s 67ms/step - loss: 0.0339 - acc: 0.9897 - val_loss: 0.0378 - val_acc: 0.9887\n",
            "Epoch 16/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0300 - acc: 0.9906Epoch 1/20\n",
            "468/468 [==============================] - 31s 66ms/step - loss: 0.0300 - acc: 0.9906 - val_loss: 0.0460 - val_acc: 0.9862\n",
            "Epoch 17/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0314 - acc: 0.9903Epoch 1/20\n",
            "468/468 [==============================] - 32s 68ms/step - loss: 0.0314 - acc: 0.9903 - val_loss: 0.0389 - val_acc: 0.9882\n",
            "Epoch 18/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0299 - acc: 0.9903Epoch 1/20\n",
            "468/468 [==============================] - 31s 66ms/step - loss: 0.0299 - acc: 0.9903 - val_loss: 0.0363 - val_acc: 0.9890\n",
            "Epoch 19/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0290 - acc: 0.9909Epoch 1/20\n",
            "468/468 [==============================] - 33s 70ms/step - loss: 0.0290 - acc: 0.9909 - val_loss: 0.0352 - val_acc: 0.9897\n",
            "Epoch 20/20\n",
            "467/468 [============================>.] - ETA: 0s - loss: 0.0276 - acc: 0.9910Epoch 1/20\n",
            "468/468 [==============================] - 32s 68ms/step - loss: 0.0276 - acc: 0.9910 - val_loss: 0.0376 - val_acc: 0.9891\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7f4bdaec5940>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 29
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ckKjE6nBdF-8",
        "colab_type": "code",
        "outputId": "f96bd493-7510-49ae-c7f1-8ace7d584b26",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        }
      },
      "source": [
        "test_loss, test_acc = model.evaluate(X_test.reshape((X_test.shape[0], 28, 28, 1)), y_test)"
      ],
      "execution_count": 30,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "21000/21000 [==============================] - 3s 135us/sample - loss: 0.0274 - acc: 0.9921\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "k4yrhbQ-d0as",
        "colab_type": "code",
        "outputId": "5dc8aada-7ee9-462d-8b3a-4c2847f8d124",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        }
      },
      "source": [
        "print(test_acc)"
      ],
      "execution_count": 31,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.99209523\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EyoGLIyIynPk",
        "colab_type": "text"
      },
      "source": [
        "## Random Forest classifier\n",
        "\n",
        "In this section we will implement a random forest classifier"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "W3atCU-1ytJ7",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from sklearn.ensemble import RandomForestClassifier\n",
        "\n",
        "forest_clf = RandomForestClassifier(n_estimators=256, max_depth=32,\n",
        "                             random_state=0)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ti6-4kcs-ZWz",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 134
        },
        "outputId": "f26e0dcb-ce40-43b8-df9e-f2aabee8b63e"
      },
      "source": [
        "forest_clf.fit(Xf, Yf)"
      ],
      "execution_count": 33,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "RandomForestClassifier(bootstrap=True, class_weight=None, criterion='gini',\n",
              "                       max_depth=32, max_features='auto', max_leaf_nodes=None,\n",
              "                       min_impurity_decrease=0.0, min_impurity_split=None,\n",
              "                       min_samples_leaf=1, min_samples_split=2,\n",
              "                       min_weight_fraction_leaf=0.0, n_estimators=256,\n",
              "                       n_jobs=None, oob_score=False, random_state=0, verbose=0,\n",
              "                       warm_start=False)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 33
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "F51IfVxg0EWs",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        },
        "outputId": "f8817dba-3700-40b2-f441-a916e44898e3"
      },
      "source": [
        "forest_clf.score(X_train, y_train) "
      ],
      "execution_count": 34,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.9646938775510204"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 34
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TembDSIq0hpZ",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        },
        "outputId": "c0590da4-20e9-45e6-c3a7-ca570703d8e8"
      },
      "source": [
        "forest_clf.score(X_test, y_test) "
      ],
      "execution_count": 35,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.9614285714285714"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 35
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rdmncdD61gGK",
        "colab_type": "text"
      },
      "source": [
        "## Voting Classifier"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qf2NR_aT1iK8",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 518
        },
        "outputId": "e53321c2-0b11-46f6-c266-a0b36bf57158"
      },
      "source": [
        "ann_pred = model.predict_proba(X_test.reshape((X_test.shape[0], 28, 28, 1)))\n",
        "rf_pred = forest_clf.predict_proba(X_test)\n",
        "joint_proba = np.sqrt(ann_pred**2 + rf_pred**2)\n",
        "\n",
        "joint_pred = np.argmax(joint_proba, axis=1)\n",
        "joint_poba = np.max(joint_proba, axis=1)\n",
        "\n",
        "from sklearn import metrics\n",
        "\n",
        "expected = y_test\n",
        "predicted = joint_pred\n",
        "print(\"Classification report for classifier %s:\\n%s\\n\"\n",
        "      % ('Joint', metrics.classification_report(expected, predicted, digits=4)))\n",
        "print(\"Confusion matrix:\\n%s\" % metrics.confusion_matrix(expected, predicted))"
      ],
      "execution_count": 37,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Classification report for classifier Joint:\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0     0.9980    0.9922    0.9951      2058\n",
            "           1     0.9966    0.9949    0.9958      2364\n",
            "           2     0.9967    0.9930    0.9948      2133\n",
            "           3     0.9927    0.9968    0.9947      2176\n",
            "           4     0.9943    0.9861    0.9901      1936\n",
            "           5     0.9896    0.9943    0.9919      1915\n",
            "           6     0.9986    0.9904    0.9945      2088\n",
            "           7     0.9911    0.9947    0.9929      2248\n",
            "           8     0.9836    0.9930    0.9883      1992\n",
            "           9     0.9848    0.9904    0.9876      2090\n",
            "\n",
            "    accuracy                         0.9927     21000\n",
            "   macro avg     0.9926    0.9926    0.9926     21000\n",
            "weighted avg     0.9927    0.9927    0.9927     21000\n",
            "\n",
            "\n",
            "Confusion matrix:\n",
            "[[2042    0    0    1    1    2    3    0    9    0]\n",
            " [   0 2352    1    0    1    1    0    7    1    1]\n",
            " [   1    2 2118    5    0    0    0    5    2    0]\n",
            " [   0    0    0 2169    0    4    0    1    2    0]\n",
            " [   0    0    0    0 1909    1    0    3    1   22]\n",
            " [   0    1    0    6    0 1904    0    0    4    0]\n",
            " [   1    2    1    0    2    4 2068    0   10    0]\n",
            " [   0    1    2    1    2    0    0 2236    1    5]\n",
            " [   0    2    2    0    0    4    0    2 1978    4]\n",
            " [   2    0    1    3    5    4    0    2    3 2070]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NIPHv_dj5dT1",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        },
        "outputId": "274f47ae-1547-4184-eb47-cc366d2d7758"
      },
      "source": [
        "fail = expected != predicted\n",
        "fail_exp = expected[fail]\n",
        "fail_pred = predicted[fail]\n",
        "X_fail = X_test[fail,:]\n",
        "proba_fail = joint_poba[fail]\n",
        "X_fail.shape"
      ],
      "execution_count": 38,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(154, 784)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 38
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IJfe8m2AHfTn",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        },
        "outputId": "1774f281-e168-4850-befc-0bc033f9deef"
      },
      "source": [
        "len(proba_fail[proba_fail > 0.95])"
      ],
      "execution_count": 39,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "54"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 39
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rrlJJZ9z5tDi",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 315
        },
        "outputId": "025899f4-816f-4094-e812-9dcc6caf52b5"
      },
      "source": [
        "num = 17\n",
        "\n",
        "digit = X_fail[num]\n",
        "digit_im = digit.reshape(28, 28)\n",
        "plt.imshow(digit_im, interpolation='nearest', cmap=mpl.cm.binary)\n",
        "print('This is a %d' %int(fail_exp[num]))\n",
        "print('The model predicted %d' %(int(fail_pred[num])))\n",
        "print('with certainty %f' %(proba_fail[num]))"
      ],
      "execution_count": 40,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "This is a 9\n",
            "The model predicted 8\n",
            "with certainty 0.960344\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAD4CAYAAAAq5pAIAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nA0EQdiAIwg4EQdiBIAg7EMQ/AE8qUgFWmon9AAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "w1GEOVD8MIUA",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 201
        },
        "outputId": "b3bfb466-8415-4f7b-ff9e-a1274db7cefa"
      },
      "source": [
        "fig, axs = plt.subplots(1, 20, figsize=(50, 5))\n",
        "\n",
        "for c, ax in enumerate(axs):\n",
        "    digit = X_fail[c]\n",
        "    digit = digit.reshape(28, 28)\n",
        "    ax.imshow(digit, interpolation='nearest', cmap=mpl.cm.binary)\n",
        "    ax.set_title('Is %d, predicted %d' %(int(fail_exp[c]), int(fail_pred[c])))\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": 52,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAACwgAAACkCAYAAACJ8OhaAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<Figure size 3600x360 with 20 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "V2pW8fPQOMun",
        "colab_type": "text"
      },
      "source": [
        "As we can see above some of the mismatches are actually difficult to ascertain even for a human.\n",
        "The model is achieving pretty good !"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "y0bD_Z28OWOt",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}