Copy_of_Seminar6.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Copy of Seminar6.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "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/wallykop/iad-deep-learning/blob/master/Copy_of_Seminar6.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hgGelUsnoxlW",
        "colab_type": "text"
      },
      "source": [
        "# Трюки для обучения нейронных сетей\n",
        "\n",
        "### **1. Аугментации**\n",
        "\n",
        "На практике часто оказывается, что нет достаточно большого количества размеченных данных, чтобы адекватно обучить модель, а общедоступные датасеты плохо приближают генеральную совокупность данных, возникающую в реальной задаче.\n",
        "\n",
        "Однако в таких ситуациях может оказаться полезным изменить объекты из доступной нам выборки. Например, если речь идёт о работе с изображениями, картинки — в зависимости от конкретной задачи — можно поворачивать, искривлять, накладывать дополнительные объекты, добавлять шум, менять яркость, расфокусировать и т. д.\n",
        "\n",
        "Многие из таких функций реализованы в [preprocessing.image](https://www.tensorflow.org/api_docs/python/tf/keras/preprocessing/image) из TensorFlow.\n",
        "\n",
        "Возьмём произвольный объект из датасета CIFAR-10:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LC7LYkcSokgJ",
        "colab_type": "code",
        "outputId": "58f6d21c-a429-42ae-b139-be1feb30f1be",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 118
        }
      },
      "source": [
        "from keras.datasets import cifar10\n",
        "(x_train, y_train), (x_test, y_test) = cifar10.load_data()\n",
        "\n",
        "NUM_CLASSES = 10\n",
        "cifar10_classes = [\"airplane\", \"automobile\", \"bird\", \"cat\", \"deer\", \n",
        "                   \"dog\", \"frog\", \"horse\", \"ship\", \"truck\"]"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Using TensorFlow backend.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<p style=\"color: red;\">\n",
              "The default version of TensorFlow in Colab will soon switch to TensorFlow 2.x.<br>\n",
              "We recommend you <a href=\"https://www.tensorflow.org/guide/migrate\" target=\"_blank\">upgrade</a> now \n",
              "or ensure your notebook will continue to use TensorFlow 1.x via the <code>%tensorflow_version 1.x</code> magic:\n",
              "<a href=\"https://colab.research.google.com/notebooks/tensorflow_version.ipynb\" target=\"_blank\">more info</a>.</p>\n"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Downloading data from https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz\n",
            "170500096/170498071 [==============================] - 2s 0us/step\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bDRZWvHkSF8o",
        "colab_type": "code",
        "outputId": "0ef3848c-5deb-415c-ecd7-0ed10121cb9c",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 281
        }
      },
      "source": [
        "import tensorflow as tf\n",
        "%matplotlib inline\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "\n",
        "rand_index = np.random.randint(0, len(y_train))\n",
        "\n",
        "x_cur = x_train[rand_index,:]\n",
        "y_cur = y_train[rand_index, 0]\n",
        "\n",
        "def draw_image(x, y):\n",
        "  plt.imshow(x)\n",
        "  plt.title(cifar10_classes[y])\n",
        "  plt.show()\n",
        "  \n",
        "draw_image(x_cur, y_cur)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAEICAYAAACZA4KlAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjEsIGh0\ndHA6Ly9tYXRwbG90bGliLm9yZy8QZhcZAAAgAElEQVR4nO2de5DcV3Xnv6df0z0PzUuyNHpYL9vY\nMsay0ToGDEXCQgF52CZgsLOOKyEYsnh3qWV3y4HaQFLZXdhdoEglRUpgB4ewtjHYawIG/FiIYwzG\nspFlPWy9sKzHaCRr3q+efpz9o1ubsfZ+74xGmh7B7/upmpqee/revnP7d/p232+fc8zdIYT41Se1\n0BMQQjQGObsQCUHOLkRCkLMLkRDk7EIkBDm7EAlBzi7OKmb2IzP7I2I738xGzSw9033F2UfO/ivI\nuepE7v6yu7e6e2Wh55JE5OxCJAQ5+zmMmd1uZvvMbMTMdprZ9fX2T5vZ30+73xozczPLmNl/AfBm\nAH9Vf8v8V/X7vNHMnjazofrvN07r/yMz+wsze7Le5x/MrNvMvm5mw/X7r5l2fzpWnfVm9rN63wfN\nrOvUeZL/9w/NbJeZDZjZD8xs9VlaSgE5+7nOPtQctx3AnwH4ezPriXVw908C+CcAt9XfMt9Wd7bv\nAvhLAN0APg/gu2bWPa3rBwDcDGAFgPUAfgLgbwF0AdgF4FMAMMuxfh/AHwLoAVCu3zeKmV0L4BMA\n3gNgSf1/uHumfmL2yNnPYdz9Pnc/4u5Vd78XwB4AV81hqN8EsMfdv+buZXe/G8ALAH572n3+1t33\nufsQgO8B2Ofuj7p7GcB9AK44jbG+5u7b3X0MwH8GcMPJQ7kIHwHw39x9V/0x/yuAjdrdzx5y9nMY\nM/t9M9tqZoNmNgjgtQAWz2Go5QAOnNJ2ALVd/CR9025PBP5uPY2xDp5iy2Lmea8G8MVp/2s/ADtl\nXHEGyNnPUeo72pcB3Aag2907AGxHzQHGADRPu/uyU7qfGsp4BDVnms75AA7PYWqzGWvVKbYSgFdm\nGPcggA+7e8e0n4K7PzmHOYoAcvZzlxbUnPY4AJjZH6C2swPAVgBvqevW7QD+5JS+fQDWTfv7IQAX\nmdlN9UO89wPYAOA7c5jXbMb6V2a2wcyaAfw5gG/OQm77GwB/YmaXAoCZtZvZ++YwP0GQs5+juPtO\nAJ9D7aCsD8BlAH5ctz0C4F4A2wA8g//fab8I4L31U+2/dPcTAH4LwMcBnADwnwD8lrvPtNuG5jWb\nsb4G4KsAjgLIA/i3sxj3AQCfBXCPmQ2j9i7mXac7P8ExJa8QIhloZxciIcjZhUgIcnYhEoKcXYiE\nEPyO8nyRzaS8KUteX4z3M/KaVK1U5zSPSvXsHkpGR5uX88/TH9QstsDcFusWe9LYwW90HrHD4rmZ\n+BTnOF46zb8ImE5zdyqVytTW2dURbB8fG6N9pqamwo9TLqNcqQb/6zNydjN7J2oyTxrAV9z9M7H7\nN2VTuGxte9CWzvCLIJvKBdtHh8P/MABUI68DY+OT1OYp/maHDVmp8Msjev1WY2+sImOidNrdsrnw\nGgJAJhO5gHN8jmbcViqFZfVcls/DyQUMAJUyl+nLkefaMuFLvOq8UzVy8bS2hR0TANo7+JcEjx7j\nKuf73v+eYPuzT/HvEx08eDDYvv/Icdpnzm/j6991/mvUtNANAG40sw1zHU8IMb+cyWf2qwDsdff9\n7j4F4B4A156daQkhzjZn4uwr8OqAh0MIBC2Y2a1mtsXMtpTK+gKPEAvFvJ/Gu/tmd9/k7puykc/l\nQoj55Uyc/TBeHd20EnOLohJCNIAzOY1/GsCFZrYWNSf/AICbZuxFpJdsNiZbhE+fU5FT5LHhCWor\nR6S3mDQ0PMKlEEYqlaW20hSXY2B8jlXjJ9MpMv1cNdInzV/zC6kmasvm+Ppn8+GJlErjtA8ic6xG\nZA1Lxa6d8Al/NbLPxa6B0ZERaktH8nMs7mijtv7j4ZP6oYEB2qdMpLxYrMucnd3dy2Z2G4AfoCa9\n3enuO+Y6nhBifjkjnd3dH0ItvlkIcY6jr8sKkRDk7EIkBDm7EAlBzi5EQmhoWqpCLuXrloalqNY2\nLvEUSVAFyrzP2GgkSCbyGrd67QXUduWVrw+2ty1aRPuMjnCpaWqKB7SMT/B+o2Nc/imRMYeGB2mf\n4YicNDbObYNDJ6iNzb+1lT9ni9t5kMlYZI7uXCqrVMJyWDpXoH3KkUCYcrFIbXkSdAMA2Rz/vy0b\n9gnz2PURnsehE0MolsrBBdHOLkRCkLMLkRDk7EIkBDm7EAlBzi5EQmhoDrpUylDIh08eUzTpE5DO\nhPuMjvFAkliwS0cXTx/00X/z76ltcXdXsD0dCchpbeEn9c2FFmqbmORqQqkUS1lF2iOqS6nMT30n\nprgqMDTYT23btm0Ntj/62Pdpn3zkahwd4/PIsOgfAFVyHUxO8tRkFklNlokEDVXL/DlL53hAVDe5\nro4cOrV+5j9TZrm4IuKadnYhEoKcXYiEIGcXIiHI2YVICHJ2IRKCnF2IhNBQ6S2TTmNJZ1hmGBvl\ngRoDQ2GZpOp52qcS0SBuvuUPqC1f4LnCzl8dDpJ5nshMANDS0klte/e9RG2bN99JbVOliDREgjEK\nBR74EStb1NbOpcP2iG3psuXB9muu+U3a5zsP3kNt7ly6cnAJNkVkOY/k/yuzwCvEKwblSUALAExM\ncKnv6OFwntZqJZKTj5g8ct1rZxciIcjZhUgIcnYhEoKcXYiEIGcXIiHI2YVICA2V3nLZLFb19ARt\nJ47z152jx/uC7WXjkXJXv/HN1HbxhtdS22oirwFAJhuW+grNXK7r6lpCbcVJLq2wSD8AaG7ikiMr\nXdTeyfO7HT9+nNoqw/x5GYqUwxocCts6O7kUefHF4Rx/APDTnz5MbSnjslahKZz7LRe58qem+PNi\nkf1xikWiAUAkCrNSCUf0pdM8ms+rxBaJejsjZzezlwCMAKgAKLv7pjMZTwgxf5yNnf3X3T1cmU4I\ncc6gz+xCJIQzdXYH8LCZPWNmt4buYGa3mtkWM9syGStRLISYV870bfw17n7YzM4D8IiZveDuj0+/\ng7tvBrAZAJZ0tDSuIoUQ4lWc0c7u7ofrv48BeADAVWdjUkKIs8+cd3YzawGQcveR+u13APjzWJ90\nKoVFzUQ2ipT+aWkaDbbnOnjiyJt+7yZqy2Rz1NbVzaWy8dHwPFrb2mmfWGmipnwztSGSRLFY5qWQ\nsiSxYRmR8lqV8P8FAOUil5NyOS4Bjhcngu35Yivts279pdR24MCL1FYu8/Ph4YGBYPva1efTPr0k\nCg0AJiPrMVniH1MzkcjClIelPq/yvdiM2Phlc0Zv45cCeKCu62YA/C9356lDhRALypyd3d33A7j8\nLM5FCDGPSHoTIiHI2YVICHJ2IRKCnF2IhNDQqLdyuYITJ4bDE8lyzaBC9IRrr7ue9mki0U4AsHQZ\nl+yyTVxaqQwXg+2trVxOyuW4zNccSQKZz/H5Dw3w5Jyd3eGowmKJ1yErRqK8Chn+PSgSYAcAmJgM\nR3KNjXPZEJH1uOxyHhH3Su8L1HbeJeuD7V1tXPY8tphLqc8+xx+rJc9r/o1NhKVIAICF1ziWPJLV\nsFOtNyGEnF2IpCBnFyIhyNmFSAhydiESQmNP46sVDAyHT+PLxoMI3n39dcH2D/3RH9M+j//Tj6it\nvY2f+jY18XlUq+ET5u7O7sh4kQCIdn6Kv6iN2/bs30NtaXJEnsrwoJUiT+EGr3JjNs2Vhmo1rGpk\ns3x/GSvyk3qmMgBA30t7qa1Agky6m7mUUFjOn8/Dh3nJq3wkMOj4MZ7nr5QJn+L39/P1yJAgmYhA\nop1diKQgZxciIcjZhUgIcnYhEoKcXYiEIGcXIiE0VHpryjVh7fmrg7YDR1+m/W668X3B9tgr1drz\n11JbpcR7Zq2F2orFcDBJf5nLKl/5H1+hto0beXBHUxOfY6XMg1pO9B8LtmdyXG4045JRtcr7xcoT\nsbWamOAlo2I5+aameIDSspXhYBcAyKTDQUMjE1xSHBjmkle+wEt9rVm1itpa83wdCyQo51hErtu7\n52Cw3SKRMNrZhUgIcnYhEoKcXYiEIGcXIiHI2YVICHJ2IRJCQ6U3rzqmiqWgbdkSHmm0lNjyOT79\nnqXnUdv4RDiXHAAcOdhLbZVSOIJq7z4ehfbkT56gtt17eL/VEemwOMnnn86FX79LZS55ZbNc1rJI\nHFWxyOfB8tNVKjyqcHEXvwbSkQi7lgLP1/fC1rBEVUjzklddnTyybfF5PPqu4hFJNxIRVyYy4JoV\nS2mfrpZwVOQrW3fTPjPu7GZ2p5kdM7Pt09q6zOwRM9tT/9050zhCiIVlNm/jvwrgnae03Q7gMXe/\nEMBj9b+FEOcwMzp7vd56/ynN1wK4q377LgDh7BJCiHOGuR7QLXX3kx9uj6JW0TWImd1qZlvMbMvE\nVPjzuhBi/jnj03h3d0RS07v7Znff5O6bCqR2uBBi/pmrs/eZWQ8A1H+Hoy+EEOcMc5Xevg3gFgCf\nqf9+cDadzAzpNJFJeOUc7HlhZ7B95fn8Y0Euy6Waxd28vM/wSDipJAC0kySQK5atoH3ec93vUtvO\nnS9Sm5OEggCw4TWvo7bRcSKxZfh4E5EIsEqknNDgIC9DNT4eLnd06NAh2ufgyweobcWKcLQkACw+\nj5fzyi0KC0UtOR691rGYi0stkQUpF/m105Tl72onyfpXnEc3dnSFpbx0OpLglFrqmNndAH4C4DVm\ndsjMPoiak7/dzPYA+Jf1v4UQ5zAz7uzufiMxve0sz0UIMY/o67JCJAQ5uxAJQc4uREKQswuREKz2\nnZjGsKy70295V/hcr1QZov08G5YZNl5xBe1z5eu57YKLL6W2tPEzyx079gXbuyJ1yLzKo8aeeOIn\n1Pbkk09T2/oLufS2atWaYHvnkiW0TzkSiRarvzY2xiPpMpnwOrIacABQmgzXAQSATJ5Lqesveg21\ndbSFE4ge3s+jwx79/veobarM1yoLLgVnI4kgS0R3TkXGGzweTkZ5/z/+GMcHh4IXnXZ2IRKCnF2I\nhCBnFyIhyNmFSAhydiESgpxdiITQ2FpvTQWsvWBD0DZZ5HWttu7YEWw/eCAshQHAP/4fLp9c8trL\nqO2yyy+ntmolLAFefNEltM/wIJcUL1i9nNpSzuvAtbSvpLbR0bBcMzoUjkIDgHwzr0O2qJVHgC3v\nOZ/amprCUlk6zcMb89lwQk8AyOR5P09xW8rCtsuvuJr2aYrUc3vouw9RW2n81IRO/0ysPt8UkXtT\nEVXccuEITMTWgg8nhPhVQs4uREKQswuREOTsQiQEObsQCaGhp/GZTBZdS8InyZbiJXd27vtFsH10\nmJcfOnqY58Dcs/t/U9ujDz9MbZe+NhyAUp7kpYTe8KZrqO3KK9ZT29o1y6jt5WP8ZP2lX7wSbP/F\nfl5qqr2ri9osw093S+WXqa2tLfx8NkdO/psiV2O6KXLKnOUd85mwgtJM1AIAWLueB0r97g0d1PbI\nd75JbePDJ6iNJfrzMlcnCk3h/yvF6m5BO7sQiUHOLkRCkLMLkRDk7EIkBDm7EAlBzi5EQmio9Faq\nVHH8RDjPmGOA9jt0+Giw/af7DtI+a1fwnGte5WV6Jsd5wMKzTz8ZbN+17We0zyPfv4jafvt3eKXr\nq97wFmrr6OJPW/tguLxWbx/P/bZ1K893130eL22VSvOSRnv3hOXSlpZwTjgAaGtupjbL8n0pGykY\nWiDSW0sLlwAPHu2jtuU9/Lp64zX8OXvyRz+gtmw1HLzkkaLH6Wz4ec6cSSCMmd1pZsfMbPu0tk+b\n2WEz21r/efdM4wghFpbZvI3/KoB3Btq/4O4b6z887k8IcU4wo7O7++MAeKCuEOKXgjM5oLvNzLbV\n3+bTDAdmdquZbTGzLaNj/GulQoj5Za7O/iUA6wFsBNAL4HPsju6+2d03ufum1haSXUMIMe/Mydnd\nvc/dK+5eBfBlAFed3WkJIc42c5LezKzH3Xvrf14PYHvs/v+vHxyZTDjC59AhnoPuxLFwJNevX3Ml\n7dPSEpZcACDXwssn7X6RR4cd6+sNtheLPDpp+4791LZnF31DhA99hH/kecf7bqa2oaFwJGDPsm7a\nZ//+sEwGAE8+9UNqKxS4jJbysAQ0MT5J++Qi0WvpNI/mykSkt6ZM2JbNcZmvuY3noFuxfCm1dS7i\n69HTcyG1vdL7YrB9kuTPAwDLkH06EvU2o7Ob2d0A3gpgsZkdAvApAG81s40AHMBLAD480zhCiIVl\nRmd39xsDzXfMw1yEEPOIvi4rREKQswuREOTsQiQEObsQCaGhUW/uVUxOjgVtkxM82ixLInnykQyF\nu3bvpLbuZbzs0sWX8tJQGy/fFGw/cJBH3x04eJjaxgd4Usz77r2f2h76MVc63/62UBgDsGwpL9W0\nank7tW3bHo44BIBfvDxIbZlUWPqcmuLRd8USvwYszfcli8hNTkoopdM84WQmy2Xbtki0XCGSxLKn\ni8t5G9YvDranM/z6LpFklA5eM0o7uxAJQc4uREKQswuREOTsQiQEObsQCUHOLkRCaKz0BsCJFlLI\nc7mjtTUcTdS1pIf2GXqeS28Hn99NbS+8yOuXrVm1Kti+bj2v2bZi9Tpq6+87Qm0///lWavuHJx6g\ntoFXhoLtn7z947RPz3lcuvrT//ivqe3AAS4rHj9OEosal6eefo4/Zy8f4o9VLvHMjIVCOLqtUi3T\nPsUir6W3bClPOPkvXs+jMEf7eVRnLh+OzJuqchktmwvv0zEZUju7EAlBzi5EQpCzC5EQ5OxCJAQ5\nuxAJoaGn8fBaMEyIzq4u2m3DhkuD7a//tWton4cff4ra+gd5AEdTjueTGx0Jn+Lv3sdzuJ23lOcs\nW7okHAABAIMjPFfbpRfxklITo+H/7cH776Z99r7IA2s+cutHqO31GzdQ29PPPB9snyzx/WX9Oh6s\nc/RoOP8fAKQtdhmHT93LJV4CbGyEl0lobuJz9DIfc/26sJIDAMWJcL7BycGwsgIAxI2iaGcXIiHI\n2YVICHJ2IRKCnF2IhCBnFyIhyNmFSAizqQizCsDfAViKWizLZnf/opl1AbgXwBrUqsLc4O4DsbHc\nHSUStOAVHpiwZElYvuolZaEAwCLlfa6+5q3UtmMbD0AZHTwRbM+VePmhiYkD1HbgJW4bHYtIb+ev\npLbOtnDQUEuelyZ6zQWXUNu999xHbYu6l1HbvgN9wfZj/eEAGQAYm+ABLS/uDpdIAuLBH6Uyua4i\n2lUuzcsuvbCTB2yNDvDLf/kyHkCzbv3qYHu5wmXglJN9msfOzGpnLwP4uLtvAHA1gI+a2QYAtwN4\nzN0vBPBY/W8hxDnKjM7u7r3u/mz99giAXQBWALgWwF31u90F4Lr5mqQQ4sw5rc/sZrYGwBUAngKw\ndFol16Oovc0XQpyjzNrZzawVwLcAfMzdX/XBy2sZKYKfFszsVjPbYmZbxib41wmFEPPLrJzdzLKo\nOfrX3f1k9YI+M+up23sABCseuPtmd9/k7ptaSNYQIcT8M6OzW+2o8w4Au9z989NM3wZwS/32LQAe\nPPvTE0KcLWYT9fYmADcDeN7MTupSnwDwGQDfMLMPAjgA4IaZBnKvolgMS0qpiHzSlGsKtltEIhmf\nLFJbPs9L+Nz4gfdT287tzwXbn/t5uB0Ahod5hF2uKfx/AYBl+OtwoYlLMs3NYWno+ee5dPWWN7+V\n2sarPGfcE09tobbRYljaqhqXKStTvPzTyuU832AUcl1ZZJ/LZ7i8Vog8ZyMjPHddn3GZOJMPX8cr\nVvIIuxEiYVZZvSvMwtnd/QkAzBPfNlN/IcS5gb5BJ0RCkLMLkRDk7EIkBDm7EAlBzi5EQmhowslq\ntYrx8bGgLZvhEo8hLE3s3MXLBR3p5eWCeo+GI7IA4OB+Hsl16cUXBNtv/j2uOr5ynEsujz3+Y2o7\nfoKXC7owUm5qYCD8LcVnfh5OAAkAzz7Hy2EtWcklr9aO86itIxeWr05Ekn1WipG9JyIBlllkG2rX\nXIjSFI96myDyMABMTnJ5zSs8au/wIT7m/kN7g+03vJ8nqWxrawu2p9N8DbWzC5EQ5OxCJAQ5uxAJ\nQc4uREKQswuREOTsQiSEhktvE5NEgsjxqLdyKdxnxfrltM/EBJdI3vhrb6C2Sy7kY/YeDCeIfHlf\nWDoBgBU9fLyb3v9eantmK6+/duXr3kRtng4/pY/8kNe+K1e4LPTibv6/xSIVc7lwdFslkkQxlgSy\nVOayVizhJKMSqZXmsT3QI0kgI9keLdKvuTN8jQwO81pvi7Lh3BAeiXrTzi5EQpCzC5EQ5OxCJAQ5\nuxAJQc4uREJo6Gk84KhWw0ELUyWeM644GQ7u+Nn9T9M+U5EcdH3HeqltfIgH0Bg5Lc5GlnHHDn6q\nvmP3Pmprbu2ktiNH+qmtRE59i2We320yEkji/BAclUi/ibHwPNKpSKkm57ZIVSPwrGk0BR0ih/Hw\nyKOlY7bI/8YCcgAgS/LajQzxUlmVTPj6jqkd2tmFSAhydiESgpxdiIQgZxciIcjZhUgIcnYhEsKM\n0puZrQLwd6iVZHYAm939i2b2aQAfAnAyWdon3P2hGQZDJkPkCeOSQToXfk1KgeclW7aE50c73nuE\n2o5WuETl5fAcUxHpJ52KvJ5GbO0dPJCnq4fLcr19wfqaWNS2iPZZdx6vtr13/35qGx+PBLWQS8sj\n8pRVeTmvqamIBhghmw0H5JSJBAwA1WrkWiTjAUBLazgvHACMjo1Q2/LlK4Lt2ch11fdK+HmOBQzN\nRmcvA/i4uz9rZm0AnjGzR+q2L7j7/5zFGEKIBWY2td56AfTWb4+Y2S4A4ZciIcQ5y2l9ZjezNQCu\nAHAyOPo2M9tmZneaGX9vKYRYcGbt7GbWCuBbAD7m7sMAvgRgPYCNqO38nyP9bjWzLWa2hSauEELM\nO7NydjPLouboX3f3+wHA3fvcveLuVQBfBnBVqK+7b3b3Te6+qZDnda+FEPPLjM5utZw/dwDY5e6f\nn9Y+vVTI9QB4xIcQYsGZzWn8mwDcDOB5M9tab/sEgBvNbCNqctxLAD4800AGIJMOyyuxqKC0hV+T\nrrj8ctrnogvCpZoAoK+PR7319/NyTf0D4Wiz4dFR2mdyikt5lYjNRrlUc+IVXr5qkMyxfyBSdikS\nUhaTDpty4WgtAEiRfrHnOZfmslYsz9xcbOmIrOXO3SIdeayYPGjGZUWWrw9V/sQ05cNrz9YdmN1p\n/BMIxxDGNXUhxDmFvkEnREKQswuREOTsQiQEObsQCUHOLkRCaHDCScAqTArh0kSeJOTLF1ppn1yK\nyzgZIuUBQHsbj1zq6OwItvcPc1lrYGCA2kYjkl2lwqOyJiMRVEcOhxNmjo/zKLrJEpeMPFInKRVZ\nRzpepDyRZfj/zCRbgF8fAJDJhC/xsch6FIs8WWkpMv+Obh5ZuH79Omor5MJfNiuVePRdvqkQbLfI\nc6KdXYiEIGcXIiHI2YVICHJ2IRKCnF2IhCBnFyIhNFZ6c6AyFZYTaOQPgKZMWJrIE8kCACoR2aKQ\nC8sWtY6Rmlxkjm1tXALsiCR6HB7mtbwG+09QW+wVulgkkXSRaKhKpMZa2nm/amStWLRZLCrLYnXU\njNtyWS7Ltbe3B9tj9dxGIhGHseSiq5avpLblS3uobaoYlgGzJNEqAFozMRZVqJ1diIQgZxciIcjZ\nhUgIcnYhEoKcXYiEIGcXIiE0VHpzAFUiyViG122rkkgeUnoNAJDLN1NbqRKp5RWJNit4WOLJRSKy\nClkuD3a2hmUhAOhs4XLe97/3A2rbdyhcA6wckWSyOb722XRkP4hEgJXJGlcjfSolvvZTkRpmsQi2\nEyfCEmYqw6XedDqScDJSqy7bxNdxMhJJVyVrNWk8IWmVTCMWVaidXYiEIGcXIiHI2YVICHJ2IRKC\nnF2IhDDjabyZ5QE8DqCpfv9vuvunzGwtgHsAdAN4BsDN7s6PD2uDASRfmOX5SWaqKXyi3dTK88WV\nyvxkNxcJqpggAQYAkEuF55gCP3HPRkoC5XJ8ufKFFmqrRvLrFYbClXJzkZPuYkSdqERssfJElWqk\nHxsOXDGIVKiKhKYAJTJHL/PT8Vg5qVRk7ffu30dt4xNcMejuDlc7Z3MHACeqQOUMT+OLAH7D3S9H\nrTzzO83sagCfBfAFd78AwACAD85iLCHEAjGjs3uNk2lQs/UfB/AbAL5Zb78LwHXzMkMhxFlhtvXZ\n0/UKrscAPAJgH4BBdz/53vAQgBXzM0UhxNlgVs7u7hV33whgJYCrAFw82wcws1vNbIuZbZmYDH+e\nFELMP6d1Gu/ugwB+COANADrM7OQB30oAweoE7r7Z3Te5+6ZCnh9kCSHmlxmd3cyWmFlH/XYBwNsB\n7ELN6d9bv9stAB6cr0kKIc6c2QTC9AC4y8zSqL04fMPdv2NmOwHcY2Z/AeDnAO6YaaAqgDEibRUn\nuTSRIkEE/RMxpY/LJ1WPyFARyShFpBCLyB2IlOOpRIIxkOa2pedfQG2dy1cH24fGeampyRKXoSrF\niBwWCa5hARmx3G9ztc2FqJQXk94itmzkuc4T+RgACs3hnIgeeawSuU5jOf5mdHZ33wbgikD7ftQ+\nvwshfgnQN+iESAhydiESgpxdiIQgZxciIcjZhUgIFstZddYfzOw4gAP1PxcDeKVhD87RPF6N5vFq\nftnmsdrdl4QMDXX2Vz2w2RZ337QgD655aB4JnIfexguREOTsQiSEhXT2zQv42NPRPF6N5vFqfmXm\nsWCf2YUQjUVv44VICHJ2IRLCgji7mb3TzF40s71mdvtCzKE+j5fM7Hkz22pmWxr4uHea2TEz2z6t\nrcvMHjGzPfXf4ZSj8z+PT5vZ4fqabDWzdzdgHqvM7IdmttPMdpjZv6u3N3RNIvNo6JqYWd7MfmZm\nz9Xn8Wf19rVm9lTdb+41M56SOYS7N/QHQBq1HHbrAOQAPAdgQ6PnUZ/LSwAWL8DjvgXAlQC2T2v7\n7wBur9++HcBnF2genwbwHxq8Hj0ArqzfbgOwG8CGRq9JZB4NXRPUkjG01m9nATwF4GoA3wDwgXr7\n3wD449MZdyF29qsA7HX3/aV1qWwAAAHxSURBVF7LM38PgGsXYB4Lhrs/DqD/lOZrUcvSCzQoWy+Z\nR8Nx9153f7Z+ewS1TEgr0OA1icyjoXiNs57ReSGcfQWAg9P+XsjMtA7gYTN7xsxuXaA5nGSpu/fW\nbx8FsHQB53KbmW2rv82f948T0zGzNaglS3kKC7gmp8wDaPCazEdG56Qf0F3j7lcCeBeAj5rZWxZ6\nQkDtlR3xzEnzyZcArEetIEgvgM816oHNrBXAtwB8zN2Hp9sauSaBeTR8TfwMMjozFsLZDwNYNe1v\nmpl2vnH3w/XfxwA8gIVNs9VnZj0AUP99bCEm4e599QutCuDLaNCamFkWNQf7urvfX29u+JqE5rFQ\na1J/7NPO6MxYCGd/GsCF9ZPFHIAPAPh2oydhZi1m1nbyNoB3ANge7zWvfBu1LL3AAmbrPelcda5H\nA9bEahke7wCwy90/P83U0DVh82j0msxbRudGnTCectr4btROOvcB+OQCzWEdakrAcwB2NHIeAO5G\n7e1gCbXPXh9ErUDmYwD2AHgUQNcCzeNrAJ4HsA01Z+tpwDyuQe0t+jYAW+s/7270mkTm0dA1AfA6\n1DI2b0PtheVPp12zPwOwF8B9AJpOZ1x9XVaIhJD0AzohEoOcXYiEIGcXIiHI2YVICHJ2IRKCnF2I\nhCBnFyIh/F/jrD1OTcFaSAAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pYJPXmIPVksI",
        "colab_type": "code",
        "outputId": "f2cc3541-3221-41fd-e7a8-d285392abe4d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 281
        }
      },
      "source": [
        "x = tf.keras.preprocessing.image.apply_affine_transform(x_cur, theta=30)\n",
        "draw_image(x, y_cur)\n",
        "#Повернём картинку на 30 градусов"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAEICAYAAACZA4KlAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nECR2IRLC/wOgqpHpriPCsQAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3quPypzyZqPH",
        "colab_type": "code",
        "outputId": "97328ded-6629-472b-9c29-0672c947d1fd",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 281
        }
      },
      "source": [
        "x = tf.keras.preprocessing.image.apply_affine_transform(x_cur, shear=35)\n",
        "draw_image(x, y_cur)\n",
        "#Наклоним картинку на 35 градусов"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAEICAYAAACZA4KlAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nILELEQkSuxCRILELEQn/H9DRCB/QLP89AAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pNDvGZock8Qg",
        "colab_type": "code",
        "outputId": "4684fdfd-a597-431a-8471-a002794da3b9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 281
        }
      },
      "source": [
        "x = tf.keras.preprocessing.image.apply_brightness_shift(x_cur, 0.5)\n",
        "draw_image(x.astype(int), y_cur)\n",
        "#уменьшим яркость"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAEICAYAAACZA4KlAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lIS5Lcv1iIdX",
        "colab_type": "text"
      },
      "source": [
        "Попробуйте реализовать зеркальное отображение картинки относительно вертикальной оси (подсказка: в numpy есть функция для изменения порядка элементов в массиве на обратный по заданному измерению):"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8UpRrflymazB",
        "colab_type": "code",
        "outputId": "35548f6b-a1bf-435f-ba6a-fb7dd99cd769",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 281
        }
      },
      "source": [
        "def horizontal_reflect(image):\n",
        "  return np.flip(image, axis=1)\n",
        "\n",
        "draw_image(horizontal_reflect(x_cur), y_cur)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAEICAYAAACZA4KlAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nsAuREBTsQiQEBbsQCUHBLkRC+P+YySdaqbLZGwAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "G2QjMl7MZmz5",
        "colab_type": "text"
      },
      "source": [
        "Обучим на CIFAR-10 свёрточную нейронную сеть без аугментации (см. семинар 4)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "00XDnF0NoAR0",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "x_train2 = x_train / 255 - 0.5\n",
        "x_test2 = x_test / 255 - 0.5\n",
        "\n",
        "y_train2 = tf.keras.utils.to_categorical(y_train, 10)\n",
        "y_test2 = tf.keras.utils.to_categorical(y_test, 10)\n",
        "\n",
        "from tensorflow.keras.models import Sequential\n",
        "from tensorflow.keras.layers import Conv2D, MaxPooling2D, Flatten, Dense, Activation, Dropout, LeakyReLU\n",
        "\n",
        "def make_model():\n",
        "    model = Sequential()\n",
        "    model.add(Conv2D(filters=16, padding='same', kernel_size=(3,3), input_shape=(32,32,3)))\n",
        "    model.add(LeakyReLU(0.1))\n",
        "    model.add(Conv2D(filters=32, padding='same', kernel_size=(3,3)))  \n",
        "    model.add(LeakyReLU(0.1))\n",
        "    model.add(MaxPooling2D(pool_size=(2,2), padding='same'))\n",
        "    model.add(Conv2D(filters=64, padding='same', kernel_size=(3,3)))  \n",
        "    model.add(LeakyReLU(0.1))\n",
        "    model.add(MaxPooling2D(pool_size=(2,2), padding='same')) \n",
        "    model.add(Flatten())\n",
        "    model.add(Dense(256))                \n",
        "    model.add(LeakyReLU(0.1))\n",
        "    model.add(Dense(10))             \n",
        "    model.add(Activation(\"softmax\"))\n",
        "    \n",
        "    return model"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_kiItBmfrJ-U",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def reset_tf_session():\n",
        "    curr_session = tf.get_default_session()\n",
        "    if curr_session is not None:\n",
        "        curr_session.close()\n",
        "    tf.keras.backend.clear_session()\n",
        "    config = tf.ConfigProto()\n",
        "    config.gpu_options.allow_growth = True\n",
        "    s = tf.InteractiveSession(config=config)\n",
        "    tf.keras.backend.set_session(s)\n",
        "    return s"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xkqaychOpe1R",
        "colab_type": "code",
        "outputId": "605105f0-c55a-478a-f296-edb54e81515e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 469
        }
      },
      "source": [
        "def train_model(x_train, y_train, x_test, y_test, make_model):\n",
        "  s = reset_tf_session()\n",
        "  model = make_model()\n",
        "  INIT_LR = 5e-3\n",
        "  BATCH_SIZE = 32\n",
        "  EPOCHS = 10\n",
        "  model.compile(\n",
        "      loss='categorical_crossentropy',\n",
        "      optimizer=tf.keras.optimizers.Adam(lr=INIT_LR),\n",
        "      metrics=['accuracy']\n",
        "  )\n",
        "\n",
        "  def lr_scheduler(epoch):\n",
        "      return INIT_LR * 0.9 ** epoch\n",
        "  \n",
        "  model.fit(\n",
        "      x_train, y_train,\n",
        "      batch_size=BATCH_SIZE,\n",
        "      epochs=EPOCHS,\n",
        "      callbacks=[tf.keras.callbacks.LearningRateScheduler(lr_scheduler)],\n",
        "      validation_data=(x_test, y_test),\n",
        "      shuffle=True,\n",
        "      verbose=1,\n",
        "      initial_epoch=0\n",
        "  )\n",
        "  \n",
        "train_model(x_train2, y_train2, x_test2, y_test2, make_model)\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow_core/python/ops/resource_variable_ops.py:1630: calling BaseResourceVariable.__init__ (from tensorflow.python.ops.resource_variable_ops) with constraint is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "If using Keras pass *_constraint arguments to layers.\n",
            "Train on 50000 samples, validate on 10000 samples\n",
            "Epoch 1/10\n",
            "50000/50000 [==============================] - 22s 439us/sample - loss: 1.2798 - acc: 0.5471 - val_loss: 1.0573 - val_acc: 0.6328\n",
            "Epoch 2/10\n",
            "50000/50000 [==============================] - 18s 361us/sample - loss: 0.9298 - acc: 0.6783 - val_loss: 0.9271 - val_acc: 0.6879\n",
            "Epoch 3/10\n",
            "50000/50000 [==============================] - 18s 359us/sample - loss: 0.7474 - acc: 0.7440 - val_loss: 1.0189 - val_acc: 0.6768\n",
            "Epoch 4/10\n",
            "50000/50000 [==============================] - 17s 331us/sample - loss: 0.5885 - acc: 0.7986 - val_loss: 1.0113 - val_acc: 0.7135\n",
            "Epoch 5/10\n",
            "50000/50000 [==============================] - 18s 353us/sample - loss: 0.4390 - acc: 0.8496 - val_loss: 1.0840 - val_acc: 0.7116\n",
            "Epoch 6/10\n",
            "50000/50000 [==============================] - 18s 365us/sample - loss: 0.3193 - acc: 0.8907 - val_loss: 1.1684 - val_acc: 0.7214\n",
            "Epoch 7/10\n",
            "50000/50000 [==============================] - 18s 363us/sample - loss: 0.2195 - acc: 0.9251 - val_loss: 1.4454 - val_acc: 0.7199\n",
            "Epoch 8/10\n",
            "50000/50000 [==============================] - 17s 348us/sample - loss: 0.1748 - acc: 0.9420 - val_loss: 1.7519 - val_acc: 0.7181\n",
            "Epoch 9/10\n",
            "50000/50000 [==============================] - 18s 362us/sample - loss: 0.1191 - acc: 0.9607 - val_loss: 2.0877 - val_acc: 0.7036\n",
            "Epoch 10/10\n",
            "50000/50000 [==============================] - 18s 360us/sample - loss: 0.0881 - acc: 0.9718 - val_loss: 2.3502 - val_acc: 0.7120\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dMp-gWprw63D",
        "colab_type": "text"
      },
      "source": [
        "Добавим аугментацию, используя ImageDataGenerator из preprocessing.image."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ollyZq5vxDcg",
        "colab_type": "code",
        "outputId": "f8c56e56-71b7-4804-c510-ba4ddb4b6916",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 757
        }
      },
      "source": [
        "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
        "\n",
        "#datagen = ImageDataGenerator(\n",
        "#    horizontal_flip=True,\n",
        "#    vertical_flip = True,\n",
        "#    rotation_range=20,\n",
        "#\t\tzoom_range=0.15,\n",
        "#\t\twidth_shift_range=0.2,\n",
        "#\t\theight_shift_range=0.2,\n",
        "#\t\tshear_range=0.15\n",
        "#)\n",
        "\n",
        "datagen = ImageDataGenerator(\n",
        "    horizontal_flip = True,\n",
        "    width_shift_range = 0.1,\n",
        "    vertical_flip = True,\n",
        "    height_shift_range = 0.1 \n",
        ")\n",
        "\n",
        "datagen.fit(x_train2)\n",
        "\n",
        "BATCH_SIZE = 32\n",
        "\n",
        "def train_gen_model(train_gen, x_test, y_test, make_model):\n",
        "  s = reset_tf_session()\n",
        "  model = make_model()\n",
        "  INIT_LR = 5e-3\n",
        "  EPOCHS = 10\n",
        "  model.compile(\n",
        "      loss='categorical_crossentropy',\n",
        "      optimizer=tf.keras.optimizers.Adam(lr=INIT_LR),\n",
        "      metrics=['accuracy']\n",
        "  )\n",
        "\n",
        "  def lr_scheduler(epoch):\n",
        "      return INIT_LR * 0.9 ** epoch\n",
        "  \n",
        "  model.fit_generator(\n",
        "      train_gen,\n",
        "      steps_per_epoch = len(x_train) / BATCH_SIZE,\n",
        "      epochs=EPOCHS,\n",
        "      callbacks=[tf.keras.callbacks.LearningRateScheduler(lr_scheduler)],\n",
        "      validation_data=(x_test, y_test),\n",
        "      shuffle=True,\n",
        "      verbose=1,\n",
        "      initial_epoch=0\n",
        "  )\n",
        "  \n",
        "train_gen_model(datagen.flow(x_train2, y_train2, batch_size=BATCH_SIZE), x_test2, y_test2, make_model)\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/10\n",
            "1562/1562 [============================>.] - ETA: 0s - loss: 1.6029 - acc: 0.4154Epoch 1/10\n",
            "10000/1562 [================================================================================================================================================================================================] - 3s 256us/sample - loss: 1.5284 - acc: 0.5223\n",
            "1563/1562 [==============================] - 44s 28ms/step - loss: 1.6028 - acc: 0.4154 - val_loss: 1.3811 - val_acc: 0.5223\n",
            "Epoch 2/10\n",
            "1562/1562 [============================>.] - ETA: 0s - loss: 1.3214 - acc: 0.5285Epoch 1/10\n",
            "10000/1562 [================================================================================================================================================================================================] - 2s 211us/sample - loss: 1.1709 - acc: 0.5754\n",
            "1563/1562 [==============================] - 43s 27ms/step - loss: 1.3216 - acc: 0.5284 - val_loss: 1.2374 - val_acc: 0.5754\n",
            "Epoch 3/10\n",
            "1561/1562 [============================>.] - ETA: 0s - loss: 1.2083 - acc: 0.5720Epoch 1/10\n",
            "10000/1562 [================================================================================================================================================================================================] - 2s 225us/sample - loss: 1.1637 - acc: 0.6067\n",
            "1563/1562 [==============================] - 43s 27ms/step - loss: 1.2082 - acc: 0.5719 - val_loss: 1.1214 - val_acc: 0.6067\n",
            "Epoch 4/10\n",
            "1562/1562 [============================>.] - ETA: 0s - loss: 1.1250 - acc: 0.6023Epoch 1/10\n",
            "10000/1562 [================================================================================================================================================================================================] - 2s 218us/sample - loss: 1.3420 - acc: 0.6311\n",
            "1563/1562 [==============================] - 42s 27ms/step - loss: 1.1251 - acc: 0.6023 - val_loss: 1.0988 - val_acc: 0.6311\n",
            "Epoch 5/10\n",
            "1560/1562 [============================>.] - ETA: 0s - loss: 1.0755 - acc: 0.6216Epoch 1/10\n",
            "10000/1562 [================================================================================================================================================================================================] - 2s 223us/sample - loss: 1.0934 - acc: 0.6392\n",
            "1563/1562 [==============================] - 43s 27ms/step - loss: 1.0757 - acc: 0.6214 - val_loss: 1.0668 - val_acc: 0.6392\n",
            "Epoch 6/10\n",
            "1561/1562 [============================>.] - ETA: 0s - loss: 1.0169 - acc: 0.6443Epoch 1/10\n",
            "10000/1562 [================================================================================================================================================================================================] - 2s 219us/sample - loss: 1.1275 - acc: 0.6389\n",
            "1563/1562 [==============================] - 42s 27ms/step - loss: 1.0168 - acc: 0.6444 - val_loss: 1.0912 - val_acc: 0.6389\n",
            "Epoch 7/10\n",
            "1562/1562 [============================>.] - ETA: 0s - loss: 0.9734 - acc: 0.6575Epoch 1/10\n",
            "10000/1562 [================================================================================================================================================================================================] - 2s 212us/sample - loss: 1.0131 - acc: 0.6820\n",
            "1563/1562 [==============================] - 43s 27ms/step - loss: 0.9733 - acc: 0.6575 - val_loss: 0.9480 - val_acc: 0.6820\n",
            "Epoch 8/10\n",
            "1561/1562 [============================>.] - ETA: 0s - loss: 0.9295 - acc: 0.6732Epoch 1/10\n",
            "10000/1562 [================================================================================================================================================================================================] - 2s 217us/sample - loss: 1.0187 - acc: 0.6852\n",
            "1563/1562 [==============================] - 43s 27ms/step - loss: 0.9296 - acc: 0.6732 - val_loss: 0.9273 - val_acc: 0.6852\n",
            "Epoch 9/10\n",
            "1561/1562 [============================>.] - ETA: 0s - loss: 0.8919 - acc: 0.6861Epoch 1/10\n",
            "10000/1562 [================================================================================================================================================================================================] - 2s 220us/sample - loss: 1.1677 - acc: 0.6887\n",
            "1563/1562 [==============================] - 43s 27ms/step - loss: 0.8918 - acc: 0.6861 - val_loss: 0.9064 - val_acc: 0.6887\n",
            "Epoch 10/10\n",
            "1561/1562 [============================>.] - ETA: 0s - loss: 0.8597 - acc: 0.6987Epoch 1/10\n",
            "10000/1562 [================================================================================================================================================================================================] - 2s 247us/sample - loss: 1.0945 - acc: 0.6956\n",
            "1563/1562 [==============================] - 43s 27ms/step - loss: 0.8596 - acc: 0.6988 - val_loss: 0.9165 - val_acc: 0.6956\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tJc859FrBFA6",
        "colab_type": "text"
      },
      "source": [
        "Обратите внимание, что происходит с loss и accuracy на валидации. Сравните с предыдущим результатом."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Z8VM2orWou45",
        "colab_type": "text"
      },
      "source": [
        "### **2. Dropout**\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kJJd7ME8BM9a",
        "colab_type": "text"
      },
      "source": [
        "При обучении моделей из предыдущего пункта можно заметить, что после нескольких эпох функция потерь на тестовой выборке начинает расти (при использовании аугментации это происходит позже; как вы думаете, почему?), то есть происходит переобучение. Чтобы избежать этой проблемы, существуют различные методы, например, [Dropout](https://http://jmlr.org/papers/v15/srivastava14a.html).\n",
        "\n",
        "Идея метода заключается в том, чтобы при обучении случайно исключать из некоторых слоёв часть нейронов. Предполагается, что это уменьшит \"совместную адаптацию\" (когда одни нейроны обучаются исправлять ошибки других) и позволит избежать переобучения.\n",
        "\n",
        "Добавим в нашу модель слои Dropout."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OyMyWgFdp64R",
        "colab_type": "code",
        "outputId": "70abdab5-4006-49dc-ff6a-19c631447650",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 395
        }
      },
      "source": [
        "def make_dropout_model():\n",
        "    model = Sequential()\n",
        "    model.add(Conv2D(filters=16, padding='same', kernel_size=(3,3), input_shape=(32,32,3)))\n",
        "    model.add(LeakyReLU(0.1))\n",
        "    model.add(Dropout(0.2))\n",
        "    model.add(Conv2D(filters=32, padding='same', kernel_size=(3,3)))  \n",
        "    model.add(LeakyReLU(0.1))\n",
        "    model.add(MaxPooling2D(pool_size=(2,2), padding='same'))\n",
        "    model.add(Dropout(0.2))\n",
        "    model.add(Conv2D(filters=64, padding='same', kernel_size=(3,3)))  \n",
        "    model.add(LeakyReLU(0.1))\n",
        "    model.add(MaxPooling2D(pool_size=(2,2), padding='same')) \n",
        "    model.add(Flatten())\n",
        "    model.add(Dropout(0.2))\n",
        "    model.add(Dense(256))                \n",
        "    model.add(LeakyReLU(0.1))\n",
        "    model.add(Dropout(0.2))\n",
        "    model.add(Dense(10))             \n",
        "    model.add(Activation(\"softmax\"))\n",
        "    #YOUR CODE HERE\n",
        "    #используйте модель из make_model и добавьте между её слоями слои Dropout(p), где p - вероятность исключения одного нейрона\n",
        "   \n",
        "    return model\n",
        "  \n",
        "train_model(x_train2, y_train2, x_test2, y_test2, make_dropout_model)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train on 50000 samples, validate on 10000 samples\n",
            "Epoch 1/10\n",
            "50000/50000 [==============================] - 19s 381us/sample - loss: 1.4188 - acc: 0.4991 - val_loss: 1.1267 - val_acc: 0.6214\n",
            "Epoch 2/10\n",
            "50000/50000 [==============================] - 19s 376us/sample - loss: 1.1623 - acc: 0.5978 - val_loss: 1.0300 - val_acc: 0.6381\n",
            "Epoch 3/10\n",
            "50000/50000 [==============================] - 19s 372us/sample - loss: 1.0527 - acc: 0.6366 - val_loss: 0.9514 - val_acc: 0.6757\n",
            "Epoch 4/10\n",
            "50000/50000 [==============================] - 19s 383us/sample - loss: 0.9805 - acc: 0.6620 - val_loss: 0.9827 - val_acc: 0.6648\n",
            "Epoch 5/10\n",
            "50000/50000 [==============================] - 19s 383us/sample - loss: 0.9039 - acc: 0.6891 - val_loss: 0.8894 - val_acc: 0.6915\n",
            "Epoch 6/10\n",
            "50000/50000 [==============================] - 19s 387us/sample - loss: 0.8344 - acc: 0.7105 - val_loss: 0.8573 - val_acc: 0.7185\n",
            "Epoch 7/10\n",
            "50000/50000 [==============================] - 18s 368us/sample - loss: 0.7731 - acc: 0.7323 - val_loss: 0.8277 - val_acc: 0.7169\n",
            "Epoch 8/10\n",
            "50000/50000 [==============================] - 19s 380us/sample - loss: 0.7129 - acc: 0.7521 - val_loss: 0.7607 - val_acc: 0.7419\n",
            "Epoch 9/10\n",
            "50000/50000 [==============================] - 18s 365us/sample - loss: 0.6612 - acc: 0.7720 - val_loss: 0.7769 - val_acc: 0.7394\n",
            "Epoch 10/10\n",
            "50000/50000 [==============================] - 19s 383us/sample - loss: 0.6101 - acc: 0.7876 - val_loss: 0.7399 - val_acc: 0.7508\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ElUftz9jrF7j",
        "colab_type": "text"
      },
      "source": [
        "Попробуйте поменять параметр Dropout. Как это влияет на результат?"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fCJeSkblrM8R",
        "colab_type": "text"
      },
      "source": [
        "### **3. Batch-нормализация**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MuiH_dHjsJsD",
        "colab_type": "text"
      },
      "source": [
        "Как вы наверняка помните, нейронным сетям легче обучаться на нормализованных данных, однако после прохождения через несколько слоёв сети матожидание и дисперсия сигнала, попадающего на вход слежующим слоям, могут значительно искажаться и таким образом вносить несоответствие в градиенты разных слоёв.\n",
        "\n",
        "Метод [Batch normalization](https://arxiv.org/abs/1502.03167) предлагает нормализовать данные не только на входе, но и внутри сети.\n",
        "\n",
        "Добавим в модель слои BatchNormalization."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CfEBgqh_sIvc",
        "colab_type": "code",
        "outputId": "e8d87876-bd38-40fd-90b1-a043be8a902d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 590
        }
      },
      "source": [
        "from tensorflow.keras.layers import BatchNormalization\n",
        "\n",
        "def make_batchnorm_model():\n",
        "    model = Sequential()\n",
        "    model.add(Conv2D(filters=16, padding='same', kernel_size=(3,3), input_shape=(32,32,3)))\n",
        "    model.add(LeakyReLU(0.1))\n",
        "    model.add(BatchNormalization())\n",
        "    model.add(Conv2D(filters=32, padding='same', kernel_size=(3,3)))  \n",
        "    model.add(LeakyReLU(0.1))\n",
        "    model.add(BatchNormalization()) \n",
        "    model.add(MaxPooling2D(pool_size=(2,2), padding='same'))\n",
        "    model.add(Conv2D(filters=64, padding='same', kernel_size=(3,3)))  \n",
        "    model.add(LeakyReLU(0.1))\n",
        "    model.add(BatchNormalization())\n",
        "    model.add(MaxPooling2D(pool_size=(2,2), padding='same')) \n",
        "    model.add(Flatten())\n",
        "    model.add(Dense(256))                \n",
        "    model.add(LeakyReLU(0.1))\n",
        "    model.add(BatchNormalization())\n",
        "    model.add(Dense(10))             \n",
        "    model.add(Activation(\"softmax\"))\n",
        "    #YOUR CODE HERE\n",
        "    #используйте модель из make_model и добавьте между её слоями слои BatchNormalization()\n",
        "    \n",
        "    return model\n",
        "  \n",
        "train_model(x_train2, y_train2, x_test2, y_test2, make_batchnorm_model)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train on 50000 samples, validate on 10000 samples\n",
            "Epoch 1/10\n",
            "50000/50000 [==============================] - 26s 510us/sample - loss: 1.1869 - acc: 0.5870 - val_loss: 0.9440 - val_acc: 0.6773\n",
            "Epoch 2/10\n",
            "50000/50000 [==============================] - 25s 491us/sample - loss: 0.7824 - acc: 0.7289 - val_loss: 0.7855 - val_acc: 0.7334\n",
            "Epoch 3/10\n",
            "50000/50000 [==============================] - 26s 518us/sample - loss: 0.6287 - acc: 0.7821 - val_loss: 0.7378 - val_acc: 0.7521\n",
            "Epoch 4/10\n",
            "50000/50000 [==============================] - 25s 493us/sample - loss: 0.4864 - acc: 0.8293 - val_loss: 0.7984 - val_acc: 0.7363\n",
            "Epoch 5/10\n",
            "50000/50000 [==============================] - 25s 508us/sample - loss: 0.3602 - acc: 0.8744 - val_loss: 0.8402 - val_acc: 0.7541\n",
            "Epoch 6/10\n",
            "  288/50000 [..............................] - ETA: 22s - loss: 0.1987 - acc: 0.9167"
          ],
          "name": "stdout"
        },
        {
          "output_type": "error",
          "ename": "KeyboardInterrupt",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-26-59b7ac94a0b8>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     25\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     26\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 27\u001b[0;31m \u001b[0mtrain_model\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx_train2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_train2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mx_test2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_test2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmake_batchnorm_model\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
            "\u001b[0;32m<ipython-input-12-ad316a630349>\u001b[0m in \u001b[0;36mtrain_model\u001b[0;34m(x_train, y_train, x_test, y_test, make_model)\u001b[0m\n\u001b[1;32m     22\u001b[0m       \u001b[0mshuffle\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mTrue\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     23\u001b[0m       \u001b[0mverbose\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 24\u001b[0;31m       \u001b[0minitial_epoch\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     25\u001b[0m   )\n\u001b[1;32m     26\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/engine/training.py\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_freq, max_queue_size, workers, use_multiprocessing, **kwargs)\u001b[0m\n\u001b[1;32m    725\u001b[0m         \u001b[0mmax_queue_size\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mmax_queue_size\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    726\u001b[0m         \u001b[0mworkers\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mworkers\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 727\u001b[0;31m         use_multiprocessing=use_multiprocessing)\n\u001b[0m\u001b[1;32m    728\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    729\u001b[0m   def evaluate(self,\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/engine/training_arrays.py\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, model, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_freq, **kwargs)\u001b[0m\n\u001b[1;32m    673\u001b[0m         \u001b[0mvalidation_steps\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mvalidation_steps\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    674\u001b[0m         \u001b[0mvalidation_freq\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mvalidation_freq\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 675\u001b[0;31m         steps_name='steps_per_epoch')\n\u001b[0m\u001b[1;32m    676\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    677\u001b[0m   def evaluate(self,\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/engine/training_arrays.py\u001b[0m in \u001b[0;36mmodel_iteration\u001b[0;34m(model, inputs, targets, sample_weights, batch_size, epochs, verbose, callbacks, val_inputs, val_targets, val_sample_weights, shuffle, initial_epoch, steps_per_epoch, validation_steps, validation_freq, mode, validation_in_fit, prepared_feed_values_from_dataset, steps_name, **kwargs)\u001b[0m\n\u001b[1;32m    392\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    393\u001b[0m         \u001b[0;31m# Get outputs.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 394\u001b[0;31m         \u001b[0mbatch_outs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mf\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mins_batch\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    395\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbatch_outs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlist\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    396\u001b[0m           \u001b[0mbatch_outs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mbatch_outs\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/backend.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, inputs)\u001b[0m\n\u001b[1;32m   3474\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3475\u001b[0m     fetched = self._callable_fn(*array_vals,\n\u001b[0;32m-> 3476\u001b[0;31m                                 run_metadata=self.run_metadata)\n\u001b[0m\u001b[1;32m   3477\u001b[0m     \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_call_fetch_callbacks\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfetched\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_fetches\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3478\u001b[0m     output_structure = nest.pack_sequence_as(\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/client/session.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1470\u001b[0m         ret = tf_session.TF_SessionRunCallable(self._session._session,\n\u001b[1;32m   1471\u001b[0m                                                \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_handle\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1472\u001b[0;31m                                                run_metadata_ptr)\n\u001b[0m\u001b[1;32m   1473\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mrun_metadata\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1474\u001b[0m           \u001b[0mproto_data\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtf_session\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTF_GetBuffer\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mrun_metadata_ptr\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_AA--BiNvNuq",
        "colab_type": "text"
      },
      "source": [
        "### **4. L2 - регуляризация**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gklK3QjDvZQf",
        "colab_type": "text"
      },
      "source": [
        "Так же, как в случае других моделей машинного обучения, нейронные сети можно штрафовать за слишком большие веса, добавляя к функции потерь норму весов с некоторым коэффициентом. Добавить регуляризацию по весам слоя можно установкой параметра kernel_regularizer."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KUgEvvcV_teK",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "regularizers."
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AB7wwQ6FrLyi",
        "colab_type": "code",
        "outputId": "d10c3aab-3627-4956-873d-755c08a412ea",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 410
        }
      },
      "source": [
        "from keras import regularizers\n",
        "\n",
        "def make_model_with_l2():\n",
        "    model = Sequential()\n",
        "    model.add(Conv2D(filters=16, padding='same', kernel_size=(3,3), input_shape=(32,32,3)))\n",
        "    model.add(LeakyReLU(0.1))\n",
        "    model.add(BatchNormalization())\n",
        "    model.add(Conv2D(filters=32, padding='same', kernel_size=(3,3)))  \n",
        "    model.add(LeakyReLU(0.1))\n",
        "    model.add(BatchNormalization()) \n",
        "    model.add(MaxPooling2D(pool_size=(2,2), padding='same'))\n",
        "    model.add(Conv2D(filters=64, padding='same', kernel_size=(3,3)))  \n",
        "    model.add(LeakyReLU(0.1))\n",
        "    model.add(BatchNormalization())\n",
        "    model.add(MaxPooling2D(pool_size=(2,2), padding='same')) \n",
        "    model.add(Flatten())\n",
        "    model.add(Dense(256, kernel_regularizer= regularizers.l1(2.0)))                \n",
        "    model.add(LeakyReLU(0.1))\n",
        "\n",
        "    model.add(Dense(10))             \n",
        "    model.add(Activation(\"softmax\"))\n",
        "    #YOUR CODE HERE\n",
        "    #используйте модель из make_model и добавьте в параметры слоёв l2-регуляризацию с помощью параметра kernel_regularizer\n",
        "    #подсказка: поищите подходящий регуляризатор в tf.keras.regularizers\n",
        "    \n",
        "    return model\n",
        "\n",
        "train_model(x_train2, y_train2, x_test2, y_test2, make_model_with_l2)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train on 50000 samples, validate on 10000 samples\n",
            "Epoch 1/10\n",
            "25120/50000 [==============>...............] - ETA: 10s - loss: 1465.0470 - acc: 0.1612"
          ],
          "name": "stdout"
        },
        {
          "output_type": "error",
          "ename": "KeyboardInterrupt",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-30-0ff4cfe8ceb9>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     26\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     27\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 28\u001b[0;31m \u001b[0mtrain_model\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx_train2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_train2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mx_test2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_test2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmake_model_with_l2\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
            "\u001b[0;32m<ipython-input-12-ad316a630349>\u001b[0m in \u001b[0;36mtrain_model\u001b[0;34m(x_train, y_train, x_test, y_test, make_model)\u001b[0m\n\u001b[1;32m     22\u001b[0m       \u001b[0mshuffle\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mTrue\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     23\u001b[0m       \u001b[0mverbose\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 24\u001b[0;31m       \u001b[0minitial_epoch\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     25\u001b[0m   )\n\u001b[1;32m     26\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/engine/training.py\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_freq, max_queue_size, workers, use_multiprocessing, **kwargs)\u001b[0m\n\u001b[1;32m    725\u001b[0m         \u001b[0mmax_queue_size\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mmax_queue_size\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    726\u001b[0m         \u001b[0mworkers\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mworkers\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 727\u001b[0;31m         use_multiprocessing=use_multiprocessing)\n\u001b[0m\u001b[1;32m    728\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    729\u001b[0m   def evaluate(self,\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/engine/training_arrays.py\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, model, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_freq, **kwargs)\u001b[0m\n\u001b[1;32m    673\u001b[0m         \u001b[0mvalidation_steps\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mvalidation_steps\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    674\u001b[0m         \u001b[0mvalidation_freq\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mvalidation_freq\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 675\u001b[0;31m         steps_name='steps_per_epoch')\n\u001b[0m\u001b[1;32m    676\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    677\u001b[0m   def evaluate(self,\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/engine/training_arrays.py\u001b[0m in \u001b[0;36mmodel_iteration\u001b[0;34m(model, inputs, targets, sample_weights, batch_size, epochs, verbose, callbacks, val_inputs, val_targets, val_sample_weights, shuffle, initial_epoch, steps_per_epoch, validation_steps, validation_freq, mode, validation_in_fit, prepared_feed_values_from_dataset, steps_name, **kwargs)\u001b[0m\n\u001b[1;32m    392\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    393\u001b[0m         \u001b[0;31m# Get outputs.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 394\u001b[0;31m         \u001b[0mbatch_outs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mf\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mins_batch\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    395\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbatch_outs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlist\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    396\u001b[0m           \u001b[0mbatch_outs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mbatch_outs\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/backend.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, inputs)\u001b[0m\n\u001b[1;32m   3474\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3475\u001b[0m     fetched = self._callable_fn(*array_vals,\n\u001b[0;32m-> 3476\u001b[0;31m                                 run_metadata=self.run_metadata)\n\u001b[0m\u001b[1;32m   3477\u001b[0m     \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_call_fetch_callbacks\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfetched\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_fetches\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3478\u001b[0m     output_structure = nest.pack_sequence_as(\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/client/session.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1470\u001b[0m         ret = tf_session.TF_SessionRunCallable(self._session._session,\n\u001b[1;32m   1471\u001b[0m                                                \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_handle\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1472\u001b[0;31m                                                run_metadata_ptr)\n\u001b[0m\u001b[1;32m   1473\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mrun_metadata\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1474\u001b[0m           \u001b[0mproto_data\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtf_session\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTF_GetBuffer\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mrun_metadata_ptr\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qRzlh5K5s2k0",
        "colab_type": "text"
      },
      "source": [
        "Что будет, если увеличить коэффициент перед весами? А если уменьшить?\n",
        "\n",
        "P. S. Обратите внимание, что регуляризация замедляет скорость обучения. Улучшится ли результат, если увеличить количество эпох?"
      ]
    }
  ]
}