Knowledge_Distillation_Notebook.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Knowledge-Distillation-Notebook",
      "version": "0.3.2",
      "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/patrickphatnguyen/Knowledge-Distillation-Keras/blob/master/Knowledge_Distillation_Notebook.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "AXtR3INVPcHJ",
        "colab_type": "text"
      },
      "source": [
        "# Distilling Knowledge in Neural Network\n",
        "\n",
        "The term \"Knowledge Distillation\" (a.k.a Teacher-Student Model) was first introduced by (Bu-cilu et al., 2006; Ba & Caruana,2014) and has been popularized by (Hinton et al., 2015), as a way to let smaller deep learning models learn how bigger ones generalize to large datasets, hence increase the performance of the smaller one. In this notebook, I'll try to explain the idea of knowledge distillation alongside with hands-on implementation of it.\n",
        "\n",
        "# The main idea\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ce65yuvootlq",
        "colab_type": "text"
      },
      "source": [
        "# Install and import requirements\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xWnbC2ooAfNf",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import tensorflow as tf\n",
        "import matplotlib.pyplot as plt \n",
        "import numpy as np\n",
        "from tensorflow.keras.models import Sequential, load_model, Model\n",
        "from tensorflow.keras.layers import Conv2D,GlobalAveragePooling2D,Dense,Softmax,Flatten,MaxPooling2D,Dropout,Activation, Lambda, concatenate\n",
        "from tensorflow.keras.callbacks import ModelCheckpoint\n",
        "from tensorflow.keras.losses import kullback_leibler_divergence as KLD_Loss, categorical_crossentropy as logloss\n",
        "from tensorflow.keras.utils import to_categorical\n",
        "from tensorflow.keras.metrics import categorical_accuracy\n",
        "import seaborn as sns"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2wQvkPoB7Cnr",
        "colab_type": "text"
      },
      "source": [
        "#  Load and preprocess the data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "N4nKTacmAjXh",
        "colab_type": "code",
        "outputId": "9212669a-2d36-4b8d-8d04-18529cff753f",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "NUM_CLASSES = 10\n",
        "(x_train, y_train), (x_test, y_test) = tf.keras.datasets.cifar10.load_data()\n",
        "print(\"x_train shape:\", x_train.shape, \"y_train shape:\", y_train.shape)\n",
        "\n",
        "# Normalize the dataset\n",
        "x_train = x_train.astype('float32') / 255\n",
        "x_test = x_test.astype('float32') / 255\n",
        "\n",
        "# Reshape each example to WIDTH*HEIGHT*CHANNELS for Convolution operation\n",
        "# x_test = x_test.reshape(-1,,28,1)\n",
        "# x_train = x_train.reshape(-1,28,28,1)\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "x_train shape: (50000, 32, 32, 3) y_train shape: (50000, 1)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rYDhVMdU7dWN",
        "colab_type": "text"
      },
      "source": [
        "# Create teacher model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "B_EpiY-nA8u6",
        "colab_type": "code",
        "outputId": "f544b054-5860-467b-f271-01b4a5c1f9c8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 558
        }
      },
      "source": [
        "Teacher = Sequential() # Must define the input shape in the first layer of the neural network\n",
        "Teacher.add(Conv2D(filters=32, kernel_size=2, padding='same', activation='relu', input_shape=(32,32,3)))\n",
        "Teacher.add(MaxPooling2D(pool_size=2))\n",
        "Teacher.add(Conv2D(filters=64, kernel_size=2, padding='same', activation='relu'))\n",
        "Teacher.add(MaxPooling2D(pool_size=2))\n",
        "Teacher.add(Flatten())\n",
        "Teacher.add(Dense(256, activation='relu'))\n",
        "Teacher.add(Dropout(0.5))\n",
        "Teacher.add(Dense(10))\n",
        "Teacher.add(Activation('softmax'))\n",
        "\n",
        "Teacher.compile(loss='sparse_categorical_crossentropy',\n",
        "             optimizer='adam',\n",
        "             metrics=['accuracy'])\n",
        "\n",
        "# Take a look at the model summary\n",
        "\n",
        "Teacher.summary()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING: Logging before flag parsing goes to stderr.\n",
            "W0809 14:32:38.185163 139750472083328 deprecation.py:506] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/init_ops.py:1251: calling VarianceScaling.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Call initializer instance with the dtype argument instead of passing it to the constructor\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "conv2d (Conv2D)              (None, 32, 32, 32)        416       \n",
            "_________________________________________________________________\n",
            "max_pooling2d (MaxPooling2D) (None, 16, 16, 32)        0         \n",
            "_________________________________________________________________\n",
            "conv2d_1 (Conv2D)            (None, 16, 16, 64)        8256      \n",
            "_________________________________________________________________\n",
            "max_pooling2d_1 (MaxPooling2 (None, 8, 8, 64)          0         \n",
            "_________________________________________________________________\n",
            "flatten (Flatten)            (None, 4096)              0         \n",
            "_________________________________________________________________\n",
            "dense (Dense)                (None, 256)               1048832   \n",
            "_________________________________________________________________\n",
            "dropout (Dropout)            (None, 256)               0         \n",
            "_________________________________________________________________\n",
            "dense_1 (Dense)              (None, 10)                2570      \n",
            "_________________________________________________________________\n",
            "activation (Activation)      (None, 10)                0         \n",
            "=================================================================\n",
            "Total params: 1,060,074\n",
            "Trainable params: 1,060,074\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WEcS4uUpBD6Q",
        "colab_type": "code",
        "outputId": "5fdb3c87-789e-40e7-ac94-9608612ea211",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 746
        }
      },
      "source": [
        "myCP = ModelCheckpoint(save_best_only=True,filepath='teacher.h5',monitor = 'val_acc')\n",
        "Teacher.fit(x_train,\n",
        "         y_train,\n",
        "         batch_size=128,\n",
        "         epochs=20,\n",
        "         validation_split = 0.2,\n",
        "         callbacks=[myCP])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train on 40000 samples, validate on 10000 samples\n",
            "Epoch 1/20\n",
            "40000/40000 [==============================] - 4s 112us/sample - loss: 1.6893 - acc: 0.3891 - val_loss: 1.3523 - val_acc: 0.5280\n",
            "Epoch 2/20\n",
            "40000/40000 [==============================] - 2s 60us/sample - loss: 1.3560 - acc: 0.5121 - val_loss: 1.2365 - val_acc: 0.5690\n",
            "Epoch 3/20\n",
            "40000/40000 [==============================] - 2s 60us/sample - loss: 1.2333 - acc: 0.5610 - val_loss: 1.0995 - val_acc: 0.6171\n",
            "Epoch 4/20\n",
            "40000/40000 [==============================] - 2s 59us/sample - loss: 1.1362 - acc: 0.5949 - val_loss: 1.1003 - val_acc: 0.6065\n",
            "Epoch 5/20\n",
            "40000/40000 [==============================] - 2s 59us/sample - loss: 1.0761 - acc: 0.6201 - val_loss: 1.0189 - val_acc: 0.6416\n",
            "Epoch 6/20\n",
            "40000/40000 [==============================] - 2s 59us/sample - loss: 1.0135 - acc: 0.6423 - val_loss: 0.9939 - val_acc: 0.6524\n",
            "Epoch 7/20\n",
            "40000/40000 [==============================] - 2s 59us/sample - loss: 0.9617 - acc: 0.6590 - val_loss: 0.9710 - val_acc: 0.6588\n",
            "Epoch 8/20\n",
            "40000/40000 [==============================] - 2s 59us/sample - loss: 0.9243 - acc: 0.6726 - val_loss: 0.9274 - val_acc: 0.6762\n",
            "Epoch 9/20\n",
            "40000/40000 [==============================] - 2s 58us/sample - loss: 0.8883 - acc: 0.6877 - val_loss: 0.9090 - val_acc: 0.6753\n",
            "Epoch 10/20\n",
            "40000/40000 [==============================] - 2s 59us/sample - loss: 0.8513 - acc: 0.6981 - val_loss: 0.9046 - val_acc: 0.6791\n",
            "Epoch 11/20\n",
            "40000/40000 [==============================] - 2s 58us/sample - loss: 0.8138 - acc: 0.7106 - val_loss: 0.9123 - val_acc: 0.6791\n",
            "Epoch 12/20\n",
            "40000/40000 [==============================] - 2s 59us/sample - loss: 0.7861 - acc: 0.7225 - val_loss: 0.8634 - val_acc: 0.6963\n",
            "Epoch 13/20\n",
            "40000/40000 [==============================] - 2s 59us/sample - loss: 0.7540 - acc: 0.7309 - val_loss: 0.8670 - val_acc: 0.7008\n",
            "Epoch 14/20\n",
            "40000/40000 [==============================] - 2s 59us/sample - loss: 0.7320 - acc: 0.7389 - val_loss: 0.8548 - val_acc: 0.7036\n",
            "Epoch 15/20\n",
            "40000/40000 [==============================] - 2s 59us/sample - loss: 0.6956 - acc: 0.7528 - val_loss: 0.8413 - val_acc: 0.7092\n",
            "Epoch 16/20\n",
            "40000/40000 [==============================] - 2s 59us/sample - loss: 0.6750 - acc: 0.7581 - val_loss: 0.8652 - val_acc: 0.7038\n",
            "Epoch 17/20\n",
            "40000/40000 [==============================] - 2s 58us/sample - loss: 0.6490 - acc: 0.7708 - val_loss: 0.8683 - val_acc: 0.7040\n",
            "Epoch 18/20\n",
            "40000/40000 [==============================] - 2s 58us/sample - loss: 0.6259 - acc: 0.7766 - val_loss: 0.8640 - val_acc: 0.7045\n",
            "Epoch 19/20\n",
            "40000/40000 [==============================] - 2s 59us/sample - loss: 0.5948 - acc: 0.7863 - val_loss: 0.8789 - val_acc: 0.7102\n",
            "Epoch 20/20\n",
            "40000/40000 [==============================] - 2s 59us/sample - loss: 0.5751 - acc: 0.7937 - val_loss: 0.8910 - val_acc: 0.6996\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7f1a0da32ac8>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 4
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ivHJ880q7xhy",
        "colab_type": "code",
        "outputId": "f3c17df4-7e3b-44e7-c19e-dad60bc255bb",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 176
        }
      },
      "source": [
        "# Retrieve best model from saved\n",
        "Teacher = load_model('teacher.h5')\n",
        "\n",
        "# Evaluation with test set\n",
        "Teacher.evaluate(x_test,y_test)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "W0809 14:33:28.917092 139750472083328 deprecation.py:506] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/init_ops.py:97: calling GlorotUniform.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Call initializer instance with the dtype argument instead of passing it to the constructor\n",
            "W0809 14:33:28.918682 139750472083328 deprecation.py:506] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/init_ops.py:97: calling Zeros.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Call initializer instance with the dtype argument instead of passing it to the constructor\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "10000/10000 [==============================] - 1s 66us/sample - loss: 0.8755 - acc: 0.7075\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[0.8754792424201965, 0.7075]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 5
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ppnopnPM6DE1",
        "colab_type": "text"
      },
      "source": [
        "# Understand temperature"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IgGxH0YOBGLm",
        "colab_type": "code",
        "outputId": "58104de8-fe67-49e9-ef12-c2a5f76ad0a0",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 729
        }
      },
      "source": [
        "Teacher_logits = Model(Teacher.input,Teacher.layers[-2].output)\n",
        "\n",
        "logits_plot = []\n",
        "\n",
        "class_names = [\"airplane\",\"automobile\",\"bird\",\"cat\",\"deer\",\"dog\",\"frog\",\"horse\",\"ship\",\"truck\"]\t\t\t\t\t\t\t\t\t\t\n",
        "\n",
        "\n",
        "# Choose the item to visualize temperature\n",
        "item_idx = 7\n",
        "\n",
        "item_image = x_train[item_idx]\n",
        "\n",
        "plt.imshow(item_image)\n",
        "\n",
        "Temperatures = [1,5,10,20,35,50]\n",
        "\n",
        "\n",
        "for Temperature in Temperatures:\n",
        "  # Create temperature layer that produces logits with temperature\n",
        "  T_layer = Lambda(lambda x:x/Temperature)(Teacher_logits.output)\n",
        "  \n",
        "  # Create a softmax layer\n",
        "  Softmax_layer = Softmax()(T_layer)\n",
        "  # Add the teacher T_layer to the whole model\n",
        "  Teacher_logits_soften = Model(Teacher.input,Softmax_layer)\n",
        "                             \n",
        "  # Append for plotting\n",
        "  logits_plot.append(Teacher_logits_soften.predict(np.array([item_image])))\n",
        "  \n",
        "  plt.figure(figsize=(14, 6))\n",
        "for i in range(len(Temperatures)):\n",
        "  sns.lineplot(class_names,logits_plot[i][0],legend=\"full\")\n",
        "  plt.title('This is a '+ class_names[y_train[item_idx][0]])\n",
        "  plt.legend(Temperatures,title=\"Temperatures\")"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nYK8MyQAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 1008x432 with 0 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 1008x432 with 0 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 1008x432 with 0 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 1008x432 with 0 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 1008x432 with 0 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAzUAAAF1CAYAAADRK8SpAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<Figure size 1008x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VC7j3YBUWe_2",
        "colab_type": "text"
      },
      "source": [
        "# Create a teacher model that create softened output\n",
        "As mentioned in **Hinton's paper**:  \"When the distilled net had 300 or more units in each of its two hidden layers, all temperatures above gave fairly similar results. But when this was radically reduced to 30 units per layer, temperatures in the range 2.5 to 4 worked significantly better than high or lower temperatures.\"  \n",
        "In this notebook, I'll use temperature **3.25**, feel free to change to the Temperature to any number of your interest."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RG0cEG8GRZm0",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "Temperature = 3.25\n",
        "T_layer = Lambda(lambda x:x/Temperature)(Teacher_logits.output)\n",
        "Softmax_layer = Activation('softmax')(T_layer)\n",
        "Teacher_soften = Model(Teacher.input,Softmax_layer)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5eS3Z8rBWY3a",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Predict and convert to sparse categorical matrix\n",
        "y_train_new = Teacher_soften.predict(x_train)\n",
        "y_test_new = Teacher_soften.predict(x_test)\n",
        "\n",
        "y_train_new = np.c_[to_categorical(y_train),y_train_new]\n",
        "y_test_new = np.c_[to_categorical(y_test),y_test_new]"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "C0wHj1lobBaa",
        "colab_type": "text"
      },
      "source": [
        "# Create a student model that produces with and without soften output"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sgDCsYaxC0Zv",
        "colab_type": "text"
      },
      "source": [
        "The student model we'll in this notebook is a really shallow neural network with only 1 hidden layers with 64 units, followed by a 10 softmax unit for the output"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cCOh7PFP6fPV",
        "colab_type": "code",
        "outputId": "d10b2c8e-13e9-4f0d-9815-37c08bff6429",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 260
        }
      },
      "source": [
        "Student = Sequential() #a Must define the input shape in the first layer of the neural network\n",
        "Student.add(Flatten(input_shape=(32,32,3)))\n",
        "Student.add(Dense(64, activation='relu'))\n",
        "Student.add(Dense(10))\n",
        "Student.summary()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential_5\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "flatten_5 (Flatten)          (None, 3072)              0         \n",
            "_________________________________________________________________\n",
            "dense_10 (Dense)             (None, 64)                196672    \n",
            "_________________________________________________________________\n",
            "dense_11 (Dense)             (None, 10)                650       \n",
            "=================================================================\n",
            "Total params: 197,322\n",
            "Trainable params: 197,322\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2ic-g-z1ODuY",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "student_logits = Student.layers[-1].output\n",
        "\n",
        "# Compute softmax\n",
        "probs = Activation(\"softmax\")(student_logits)\n",
        "\n",
        "# Compute softmax with softened logits\n",
        "logits_T = Lambda(lambda x:x/Temperature)(student_logits)\n",
        "probs_T = Activation(\"softmax\")(logits_T)\n",
        "\n",
        "CombinedLayers = concatenate([probs,probs_T])\n",
        "\n",
        "StudentModel = Model(Student.input,CombinedLayers)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Bs3G-pvGOOFE",
        "colab_type": "text"
      },
      "source": [
        "<center><img src=\"https://nervanasystems.github.io/distiller/imgs/knowledge_distillation.png\" width=500></center>\n",
        "<center> "
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kpXESRefDGtb",
        "colab_type": "text"
      },
      "source": [
        "$$ \\text{Let } a_{t}  \\text{ and } a_{s} \\text{ be the logits (the inputs to the final softmax) of the teacher and student network, respectively, with the ground-truth label } y_{r} .\\text{ We calculate the cross-entropy between the softmax} (a_{s},y_{r}) \\text{ and } y_{r} \\text{ as follow:}$$\n",
        "$$ \\mathcal{L}_{SL}=\\mathcal{H}(\\text{softmax}(a_{s},y_{r})) $$\n",
        "\n",
        "$$ \\text{In knowledge distillation (in all 3 papers), we tries to match the softened outputs of the student } y_{s} = \\text{softmax}(a_{s}/\\mathcal{T})   \\text{ and teacher's softened outputs }  y_{t}=\\text{softmax}(a_{t}/\\mathcal{T}) \\text{via a KL-divergence loss}$$\n",
        "$$\\mathcal{L}_{KD}=\\mathcal{T}^2\\text{KL}(y_{s},y_{t})$$ \n",
        "$$ \\text{The student model will then be trained on a \"combined\" loss between } \\mathcal{L}_{SL} \\text{ and } \\mathcal{L}_{KD} \\text{ with } \\lambda \\text{ representing the trade off of 2 losses }$$\n",
        "$$\\mathcal{L}_{\\text{student}} = \\lambda\\mathcal{L}_{SL} + (1-\\lambda)\\mathcal{L}_{KD}$$"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "n2-QsDZ5jtA8",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def KD_loss(y_true,y_pred,lambd=0.5,T=10.0):\n",
        "  y_true,y_true_KD = y_true[:,:NUM_CLASSES],y_true[:,NUM_CLASSES:]\n",
        "  y_pred,y_pred_KD = y_pred[:,:NUM_CLASSES],y_pred[:,NUM_CLASSES:]\n",
        "  # Classic cross-entropy (without temperature)\n",
        "  CE_loss = logloss(y_true,y_pred)\n",
        "  # KL-Divergence loss for softened output (with temperature)\n",
        "  KL_loss = T**2*KLD_Loss(y_true_KD,y_pred_KD)\n",
        "  \n",
        "  return lambd*CE_loss + (1-lambd)*KL_loss\n",
        "\n",
        "def accuracy(y_true,y_pred):\n",
        "  return categorical_accuracy(y_true,y_pred)\n",
        "  "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0pbswIyRhYUG",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "StudentModel.compile(optimizer='adam',loss=lambda y_true,y_pred: KD_loss(y_true, y_pred,lambd=0.5,T=Temperature),metrics=[accuracy])"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "am0Ufoz6jc3-",
        "colab_type": "code",
        "outputId": "11ed429d-11db-4e79-a12e-99a9510dbb91",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "myCP = ModelCheckpoint(save_best_only=True,filepath='student.h5',monitor = 'val_accuracy')\n",
        "\n",
        "StudentModel.fit(x_train,y_train_new,epochs=50,validation_split=0.15,batch_size=128,callbacks=[myCP])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train on 42500 samples, validate on 7500 samples\n",
            "Epoch 1/50\n",
            "42500/42500 [==============================] - 3s 60us/sample - loss: 3.7140 - accuracy: 0.3105 - val_loss: 3.2836 - val_accuracy: 0.3513\n",
            "Epoch 2/50\n",
            "42500/42500 [==============================] - 2s 44us/sample - loss: 3.2945 - accuracy: 0.3738 - val_loss: 3.1989 - val_accuracy: 0.3471\n",
            "Epoch 3/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 3.1572 - accuracy: 0.3867 - val_loss: 3.0909 - val_accuracy: 0.3820\n",
            "Epoch 4/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 3.0415 - accuracy: 0.4030 - val_loss: 3.3710 - val_accuracy: 0.3577\n",
            "Epoch 5/50\n",
            "42500/42500 [==============================] - 2s 44us/sample - loss: 2.9621 - accuracy: 0.4102 - val_loss: 3.0683 - val_accuracy: 0.3695\n",
            "Epoch 6/50\n",
            "42500/42500 [==============================] - 2s 44us/sample - loss: 2.8845 - accuracy: 0.4166 - val_loss: 2.8167 - val_accuracy: 0.4116\n",
            "Epoch 7/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.7905 - accuracy: 0.4280 - val_loss: 3.2658 - val_accuracy: 0.3464\n",
            "Epoch 8/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.8092 - accuracy: 0.4230 - val_loss: 2.8306 - val_accuracy: 0.4009\n",
            "Epoch 9/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.7244 - accuracy: 0.4365 - val_loss: 2.7771 - val_accuracy: 0.4063\n",
            "Epoch 10/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.6815 - accuracy: 0.4416 - val_loss: 2.6925 - val_accuracy: 0.4255\n",
            "Epoch 11/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.6444 - accuracy: 0.4460 - val_loss: 2.7171 - val_accuracy: 0.4116\n",
            "Epoch 12/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.6298 - accuracy: 0.4464 - val_loss: 2.7660 - val_accuracy: 0.4127\n",
            "Epoch 13/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.5935 - accuracy: 0.4519 - val_loss: 2.6507 - val_accuracy: 0.4235\n",
            "Epoch 14/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.5858 - accuracy: 0.4544 - val_loss: 2.6539 - val_accuracy: 0.4239\n",
            "Epoch 15/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.5647 - accuracy: 0.4545 - val_loss: 2.5736 - val_accuracy: 0.4307\n",
            "Epoch 16/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.5307 - accuracy: 0.4604 - val_loss: 2.9367 - val_accuracy: 0.3903\n",
            "Epoch 17/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.5486 - accuracy: 0.4581 - val_loss: 2.5169 - val_accuracy: 0.4421\n",
            "Epoch 18/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.5094 - accuracy: 0.4636 - val_loss: 2.7981 - val_accuracy: 0.4024\n",
            "Epoch 19/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.5171 - accuracy: 0.4628 - val_loss: 2.6668 - val_accuracy: 0.4217\n",
            "Epoch 20/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.5102 - accuracy: 0.4643 - val_loss: 2.6968 - val_accuracy: 0.4331\n",
            "Epoch 21/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.4748 - accuracy: 0.4695 - val_loss: 2.4857 - val_accuracy: 0.4523\n",
            "Epoch 22/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.4607 - accuracy: 0.4705 - val_loss: 2.6590 - val_accuracy: 0.4215\n",
            "Epoch 23/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.4755 - accuracy: 0.4678 - val_loss: 2.4940 - val_accuracy: 0.4481\n",
            "Epoch 24/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.4382 - accuracy: 0.4712 - val_loss: 2.6369 - val_accuracy: 0.4236\n",
            "Epoch 25/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.4490 - accuracy: 0.4707 - val_loss: 2.4935 - val_accuracy: 0.4505\n",
            "Epoch 26/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.4230 - accuracy: 0.4724 - val_loss: 2.6235 - val_accuracy: 0.4256\n",
            "Epoch 27/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.4142 - accuracy: 0.4762 - val_loss: 2.5793 - val_accuracy: 0.4359\n",
            "Epoch 28/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.4222 - accuracy: 0.4740 - val_loss: 2.5541 - val_accuracy: 0.4379\n",
            "Epoch 29/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.4126 - accuracy: 0.4740 - val_loss: 2.4779 - val_accuracy: 0.4580\n",
            "Epoch 30/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.3876 - accuracy: 0.4793 - val_loss: 2.5100 - val_accuracy: 0.4483\n",
            "Epoch 31/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.3927 - accuracy: 0.4787 - val_loss: 2.8003 - val_accuracy: 0.4123\n",
            "Epoch 32/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.3916 - accuracy: 0.4790 - val_loss: 2.5754 - val_accuracy: 0.4219\n",
            "Epoch 33/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.3742 - accuracy: 0.4811 - val_loss: 2.5109 - val_accuracy: 0.4399\n",
            "Epoch 34/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.3760 - accuracy: 0.4803 - val_loss: 2.7033 - val_accuracy: 0.4223\n",
            "Epoch 35/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.3707 - accuracy: 0.4814 - val_loss: 2.4971 - val_accuracy: 0.4500\n",
            "Epoch 36/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.3497 - accuracy: 0.4840 - val_loss: 2.5929 - val_accuracy: 0.4324\n",
            "Epoch 37/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.3456 - accuracy: 0.4816 - val_loss: 2.7595 - val_accuracy: 0.4384\n",
            "Epoch 38/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.3646 - accuracy: 0.4814 - val_loss: 2.5590 - val_accuracy: 0.4269\n",
            "Epoch 39/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.3490 - accuracy: 0.4832 - val_loss: 2.4646 - val_accuracy: 0.4485\n",
            "Epoch 40/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.3334 - accuracy: 0.4850 - val_loss: 2.4714 - val_accuracy: 0.4476\n",
            "Epoch 41/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.3326 - accuracy: 0.4846 - val_loss: 2.5795 - val_accuracy: 0.4444\n",
            "Epoch 42/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.3318 - accuracy: 0.4873 - val_loss: 2.6374 - val_accuracy: 0.4331\n",
            "Epoch 43/50\n",
            "42500/42500 [==============================] - 2s 44us/sample - loss: 2.3180 - accuracy: 0.4895 - val_loss: 2.4602 - val_accuracy: 0.4589\n",
            "Epoch 44/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.3154 - accuracy: 0.4884 - val_loss: 2.4701 - val_accuracy: 0.4433\n",
            "Epoch 45/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.3238 - accuracy: 0.4858 - val_loss: 2.4439 - val_accuracy: 0.4463\n",
            "Epoch 46/50\n",
            "42500/42500 [==============================] - 2s 43us/sample - loss: 2.3165 - accuracy: 0.4868 - val_loss: 2.4843 - val_accuracy: 0.4427\n",
            "Epoch 47/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.2992 - accuracy: 0.4876 - val_loss: 2.4383 - val_accuracy: 0.4544\n",
            "Epoch 48/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.3092 - accuracy: 0.4912 - val_loss: 2.4675 - val_accuracy: 0.4521\n",
            "Epoch 49/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.2938 - accuracy: 0.4907 - val_loss: 2.5032 - val_accuracy: 0.4436\n",
            "Epoch 50/50\n",
            "42500/42500 [==============================] - 2s 42us/sample - loss: 2.2932 - accuracy: 0.4902 - val_loss: 2.4657 - val_accuracy: 0.4508\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7f194f1e3c18>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 42
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-QujVzs7rRxN",
        "colab_type": "code",
        "outputId": "f6fd2c26-101c-4e18-9e16-73af89d0a024",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "source": [
        "StudentModel.load_weights('student.h5')\n",
        "StudentModel.evaluate(x_train,y_train_new)\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "50000/50000 [==============================] - 3s 64us/sample - loss: 2.3504 - accuracy: 0.4861\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[2.3504372370910644, 0.4861]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 43
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1k5LsqTWzh56",
        "colab_type": "text"
      },
      "source": [
        "# Create a standalone student"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vfInFTnqr6Yg",
        "colab_type": "code",
        "outputId": "0871bbcb-d3e7-4946-bee7-acfe901c7648",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 260
        }
      },
      "source": [
        "AloneModel = Sequential() #a Must define the input shape in the first layer of the neural netAloneStudent = Sequential() #a Must define the input shape in the first layer of the neural network\n",
        "AloneModel.add(Flatten(input_shape=(32,32,3)))\n",
        "AloneModel.add(Dense(64, activation='relu'))\n",
        "AloneModel.add(Dense(10,activation=\"softmax\"))\n",
        "AloneModel.summary()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential_6\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "flatten_6 (Flatten)          (None, 3072)              0         \n",
            "_________________________________________________________________\n",
            "dense_12 (Dense)             (None, 64)                196672    \n",
            "_________________________________________________________________\n",
            "dense_13 (Dense)             (None, 10)                650       \n",
            "=================================================================\n",
            "Total params: 197,322\n",
            "Trainable params: 197,322\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "N9rxNJ9AzkVj",
        "colab_type": "code",
        "outputId": "ade89b2c-cb7a-4a39-c9f4-715b4329cb87",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "AloneModel.compile(optimizer='adam',loss='sparse_categorical_crossentropy',metrics=['accuracy'])\n",
        "\n",
        "myCP = ModelCheckpoint(_best_only=True,filepath='alone.h5',monitor = 'val_acc')\n",
        "\n",
        "AloneModel.fit(x_train,y_train,epochs=50,validation_split=0.15,batch_size=128,callbacks=[myCP])\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train on 42500 samples, validate on 7500 samples\n",
            "Epoch 1/50\n",
            "42500/42500 [==============================] - 2s 52us/sample - loss: 2.0431 - acc: 0.2463 - val_loss: 1.9507 - val_acc: 0.2873\n",
            "Epoch 2/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.8805 - acc: 0.3196 - val_loss: 1.9021 - val_acc: 0.3144\n",
            "Epoch 3/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.8343 - acc: 0.3402 - val_loss: 1.8557 - val_acc: 0.3289\n",
            "Epoch 4/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.8064 - acc: 0.3496 - val_loss: 1.8348 - val_acc: 0.3376\n",
            "Epoch 5/50\n",
            "42500/42500 [==============================] - 1s 35us/sample - loss: 1.7856 - acc: 0.3571 - val_loss: 1.8426 - val_acc: 0.3353\n",
            "Epoch 6/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.7756 - acc: 0.3619 - val_loss: 1.8208 - val_acc: 0.3356\n",
            "Epoch 7/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.7637 - acc: 0.3663 - val_loss: 1.8080 - val_acc: 0.3509\n",
            "Epoch 8/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.7516 - acc: 0.3723 - val_loss: 1.7973 - val_acc: 0.3532\n",
            "Epoch 9/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.7472 - acc: 0.3756 - val_loss: 1.8044 - val_acc: 0.3535\n",
            "Epoch 10/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.7410 - acc: 0.3787 - val_loss: 1.7833 - val_acc: 0.3597\n",
            "Epoch 11/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.7312 - acc: 0.3825 - val_loss: 1.7926 - val_acc: 0.3563\n",
            "Epoch 12/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.7306 - acc: 0.3851 - val_loss: 1.7787 - val_acc: 0.3624\n",
            "Epoch 13/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.7238 - acc: 0.3859 - val_loss: 1.7630 - val_acc: 0.3735\n",
            "Epoch 14/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.7201 - acc: 0.3864 - val_loss: 1.7799 - val_acc: 0.3568\n",
            "Epoch 15/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.7190 - acc: 0.3861 - val_loss: 1.7817 - val_acc: 0.3676\n",
            "Epoch 16/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.7170 - acc: 0.3862 - val_loss: 1.7568 - val_acc: 0.3709\n",
            "Epoch 17/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.7112 - acc: 0.3870 - val_loss: 1.8002 - val_acc: 0.3616\n",
            "Epoch 18/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.7094 - acc: 0.3871 - val_loss: 1.7986 - val_acc: 0.3584\n",
            "Epoch 19/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6979 - acc: 0.3937 - val_loss: 1.7436 - val_acc: 0.3721\n",
            "Epoch 20/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.6918 - acc: 0.3941 - val_loss: 1.7915 - val_acc: 0.3605\n",
            "Epoch 21/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.6883 - acc: 0.3942 - val_loss: 1.7701 - val_acc: 0.3733\n",
            "Epoch 22/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.6829 - acc: 0.3968 - val_loss: 1.7264 - val_acc: 0.3803\n",
            "Epoch 23/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.6792 - acc: 0.3965 - val_loss: 1.7635 - val_acc: 0.3691\n",
            "Epoch 24/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.6798 - acc: 0.3956 - val_loss: 1.8555 - val_acc: 0.3349\n",
            "Epoch 25/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.6823 - acc: 0.3966 - val_loss: 1.8043 - val_acc: 0.3607\n",
            "Epoch 26/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6770 - acc: 0.3979 - val_loss: 1.7220 - val_acc: 0.3836\n",
            "Epoch 27/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.6743 - acc: 0.3976 - val_loss: 1.7976 - val_acc: 0.3645\n",
            "Epoch 28/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.6698 - acc: 0.3988 - val_loss: 1.8179 - val_acc: 0.3485\n",
            "Epoch 29/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6746 - acc: 0.3990 - val_loss: 1.7522 - val_acc: 0.3716\n",
            "Epoch 30/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6651 - acc: 0.4006 - val_loss: 1.7548 - val_acc: 0.3677\n",
            "Epoch 31/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6635 - acc: 0.4024 - val_loss: 1.7113 - val_acc: 0.3860\n",
            "Epoch 32/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.6613 - acc: 0.3998 - val_loss: 1.7345 - val_acc: 0.3788\n",
            "Epoch 33/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6597 - acc: 0.4031 - val_loss: 1.7426 - val_acc: 0.3795\n",
            "Epoch 34/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6580 - acc: 0.4058 - val_loss: 1.7356 - val_acc: 0.3823\n",
            "Epoch 35/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6550 - acc: 0.4043 - val_loss: 1.7134 - val_acc: 0.3867\n",
            "Epoch 36/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.6578 - acc: 0.4053 - val_loss: 1.7192 - val_acc: 0.3809\n",
            "Epoch 37/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6550 - acc: 0.4052 - val_loss: 1.9328 - val_acc: 0.3287\n",
            "Epoch 38/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6604 - acc: 0.4012 - val_loss: 1.7676 - val_acc: 0.3791\n",
            "Epoch 39/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6515 - acc: 0.4052 - val_loss: 1.7316 - val_acc: 0.3755\n",
            "Epoch 40/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6510 - acc: 0.4057 - val_loss: 1.7286 - val_acc: 0.3839\n",
            "Epoch 41/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.6482 - acc: 0.4075 - val_loss: 1.7213 - val_acc: 0.3777\n",
            "Epoch 42/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6445 - acc: 0.4069 - val_loss: 1.7184 - val_acc: 0.3831\n",
            "Epoch 43/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6459 - acc: 0.4080 - val_loss: 1.7021 - val_acc: 0.3923\n",
            "Epoch 44/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6477 - acc: 0.4076 - val_loss: 1.7214 - val_acc: 0.3856\n",
            "Epoch 45/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6452 - acc: 0.4076 - val_loss: 1.7032 - val_acc: 0.3889\n",
            "Epoch 46/50\n",
            "42500/42500 [==============================] - 1s 33us/sample - loss: 1.6442 - acc: 0.4077 - val_loss: 1.7519 - val_acc: 0.3761\n",
            "Epoch 47/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.6410 - acc: 0.4113 - val_loss: 1.7117 - val_acc: 0.3800\n",
            "Epoch 48/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.6419 - acc: 0.4088 - val_loss: 1.7050 - val_acc: 0.3877\n",
            "Epoch 49/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.6402 - acc: 0.4084 - val_loss: 1.7063 - val_acc: 0.3841\n",
            "Epoch 50/50\n",
            "42500/42500 [==============================] - 1s 34us/sample - loss: 1.6370 - acc: 0.4110 - val_loss: 1.7126 - val_acc: 0.3867\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7f194edf6c50>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 45
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "14Fju4pe0Ln2",
        "colab_type": "code",
        "outputId": "d96423bb-7c7f-4dd6-ad0d-22f5480eda7a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "source": [
        "AloneModel = load_model(\"alone.h5\")\n",
        "AloneModel.evaluate(x_test,y_test)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "10000/10000 [==============================] - 1s 66us/sample - loss: 1.6973 - acc: 0.3934\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[1.6973005071640015, 0.3934]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 46
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "P91kVQ-ORtzh",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EW1IbWlkRvQv",
        "colab_type": "text"
      },
      "source": [
        "# References\n",
        "[Nervanasystem github's\n",
        "](https://nervanasystems.github.io/distiller/knowledge_distillation.html)\n",
        "\n",
        "[Hinton et. al. - \n",
        "Distilling the Knowledge in a Neural Network](https://arxiv.org/abs/1503.02531)\n",
        "\n",
        "[Seyed-Iman Mirzadeh et. al. - Improved Knowledge Distillation via Teacher Assistant:Bridging the Gap Between Student and Teacher](https://arxiv.org/abs/1902.03393)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uDQb_wAfRwat",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}