keras-multilayer.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "keras-multilayer.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/hikmatfarhat-ndu/CSC645/blob/master/keras-multilayer.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Wxn309X9dUHT",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "import random\n",
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "from tensorflow.keras import layers\n",
        "from tensorflow.keras.models import Sequential\n",
        "from tensorflow.keras.layers import Dense, Conv2D, MaxPooling2D, Dropout, Flatten"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HlppWpptdZwA",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#from google.colab import drive\n",
        "#drive.mount('/content/drive')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-R5raOpLdkRs",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def load_dataset():\n",
        "  # tensorflow assumes the input and output are row vectors\n",
        "  # where as in our implementation we use them as column vectors\n",
        "    tr,te=keras.datasets.mnist.load_data()\n",
        "    X=tr[0].astype(\"float32\")/255\n",
        "    Y=tr[1]\n",
        "    #X=X.T\n",
        "    Y=Y.reshape(1,len(Y))\n",
        "    V=np.zeros((10,Y.shape[1]))\n",
        "    for j in range(Y.shape[1]):\n",
        "        V[Y[0,j],j]=1\n",
        "        \n",
        "    Y=V.astype(\"float32\")\n",
        "    test_data=te[0].astype(\"float32\")/255\n",
        "    test_labels=te[1].astype(\"float32\")\n",
        "    test_labels=test_labels.reshape(1,len(test_labels))\n",
        "    return X,Y.T,test_data,test_labels.T\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gvEmfFo74Gqe",
        "colab_type": "text"
      },
      "source": [
        "## Create Keras Model\n",
        "\n",
        "All the models we use are sequential, i.e. the output of one is fed into the input of the next. The \"add\" operation is akind to the push operation on a stack. Below we first add the first layer that takes input of certain shape."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "I2TO1rtdfxFj",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "model=Sequential()\n",
        "model.add(Dense(128,activation=\"sigmoid\",name=\"dense_1\",input_shape=(784,)))\n",
        "model.add(Dense(64,activation=\"sigmoid\"))\n",
        "model.add(Dense(10,activation=\"sigmoid\"))\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lZ-fcE1gf6Di",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "X,Y,test_data,test_labels = load_dataset()\n",
        "\n",
        "X=X.reshape(60000,784)\n",
        "test_data=test_data.reshape(10000,784)\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oO72TEvcgNmK",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "model.compile(\n",
        "    optimizer=keras.optimizers.RMSprop(),  # Optimizer\n",
        "    # Loss function to minimize\n",
        "    #loss=keras.losses.SparseCategoricalCrossentropy(),\n",
        "    loss=keras.losses.CategoricalCrossentropy(),\n",
        "    # List of metrics to monitor\n",
        "    #metrics=[keras.metrics.SparseCategoricalAccuracy()],\n",
        "\n",
        "    metrics=[keras.metrics.CategoricalAccuracy()],\n",
        ")\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "O3fi-5KogPCp",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "print(\"Fit model on training data\")\n",
        "history = model.fit(\n",
        "    X,\n",
        "    Y,\n",
        "    batch_size=64,\n",
        "    epochs=10,\n",
        "   \n",
        ")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BKe136ej359z",
        "colab_type": "text"
      },
      "source": [
        "## Evaluate on test data\n",
        "\n",
        "Note that the test_data is **not** in one_hot encoding."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "k9pqSbriuiXb",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#model.evaluate(test_data, tf.one_hot(test_labels,depth=10))\n",
        "\n",
        "test_labels_one_hot=tf.one_hot(test_labels.flatten(),depth=10)\n",
        "model.evaluate(test_data,test_labels_one_hot)"
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}