neural_network.py.ipynb

neural_network.py.ipynb

{
  "cells": [
    {
      "metadata": {
        "ExecuteTime": {
          "start_time": "2019-09-09T21:54:32.444Z"
        },
        "trusted": true
      },
      "cell_type": "code",
      "source": "import numpy as np\nimport tensorflow as tf\nfrom utils import get_mnist_data\n\n\ndef single_layer_net(x):\n\n    input_dim       = 784\n    n_classes       = 10\n\n    with tf.variable_scope('single_layer_net'):\n\n        W = tf.Variable(initial_value=tf.random_normal(shape=[input_dim, n_classes]), name='weights')\n        b = tf.Variable(initial_value=tf.zeros(shape=[n_classes]), name='biases')\n\n        logits = tf.matmul(x, W) + b\n\n        y = tf.nn.softmax(logits)\n\n        return y\n\n\ndef multi_layer_net(x):\n\n    input_dim       = 784\n    hidden_dim      = 100\n    n_classes       = 10\n\n    with tf.variable_scope('multi_layer_net'):\n\n        W_1 = tf.Variable(initial_value=tf.random_normal(shape=[input_dim, hidden_dim]), name='l1_weights')\n        b_1 = tf.Variable(initial_value=tf.zeros(shape=[hidden_dim]), name='l1_biases')\n\n        hidden_1 = tf.nn.relu(tf.matmul(x, W_1) + b_1)\n\n        W_2 = tf.Variable(initial_value=tf.random_normal(shape=[hidden_dim, n_classes]), name='l2_weights')\n        b_2 = tf.Variable(initial_value=tf.zeros(shape=[n_classes]), name='l2_biases')\n\n        logits = tf.matmul(hidden_1, W_2) + b_2\n\n        y = tf.nn.softmax(logits)\n\n        return y\n\n\nif __name__ == '__main__':\n\n    # Load MNIST data\n    mnist = get_mnist_data('/tmp/mnist', verbose=True)\n\n    # Placeholders\n    x = tf.placeholder(dtype=tf.float32, shape=[None, 784])     # input placeholder\n\n    # Placeholder for targets\n    targets = tf.placeholder(dtype=tf.float32, shape=[None, 10])\n\n    # Define model output\n    y = multi_layer_net(x)\n\n    # Define loss function\n    loss = tf.reduce_mean(-tf.reduce_sum(targets * tf.log(y + np.finfo('float32').eps), axis=1))\n\n    # Define train step\n    train_step = tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss)\n\n    init_op = tf.global_variables_initializer()\n\n    # Define metrics\n    correct_predictions = tf.equal(tf.argmax(y, axis=1), tf.argmax(targets, axis=1))\n    accuracy = tf.reduce_mean(tf.cast(correct_predictions, tf.float32))\n\n    with tf.Session() as sess:\n\n        # Initialize all variables\n        sess.run(init_op)\n\n        # Training parameters\n        training_epochs = 100\n        batch_size      = 128\n\n        # Number of batches to process to see whole dataset\n        batches_each_epoch = mnist.train.num_examples // batch_size\n\n        for epoch in range(training_epochs):\n\n            # During training measure accuracy on validation set to have an idea of what's happening\n            val_accuracy = sess.run(fetches=accuracy,\n                                    feed_dict={x: mnist.validation.images, targets: mnist.validation.labels})\n            print('Epoch: {:06d} - VAL accuracy: {:.03f}'.format(epoch, val_accuracy))\n\n            for _ in range(batches_each_epoch):\n\n                # Load a batch of training data\n                x_batch, target_batch = mnist.train.next_batch(batch_size)\n\n                # Actually run one training step here\n                sess.run(fetches=[train_step],\n                         feed_dict={x: x_batch, targets: target_batch})\n\n        # Eventually evaluate on whole test set when training ends\n        test_accuracy = sess.run(fetches=accuracy,\n                                 feed_dict={x: mnist.test.images, targets: mnist.test.labels})\n        print('*' * 50)\n        print('Training ended. TEST accuracy: {:.03f}'.format(test_accuracy))",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "",
      "execution_count": null,
      "outputs": []
    }
  ],
  "metadata": {
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3",
      "language": "python"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.7.3"
    },
    "latex_envs": {
      "eqNumInitial": 1,
      "eqLabelWithNumbers": true,
      "current_citInitial": 1,
      "cite_by": "apalike",
      "bibliofile": "biblio.bib",
      "LaTeX_envs_menu_present": true,
      "labels_anchors": false,
      "latex_user_defs": false,
      "user_envs_cfg": false,
      "report_style_numbering": false,
      "autoclose": false,
      "autocomplete": true,
      "hotkeys": {
        "equation": "Ctrl-E",
        "itemize": "Ctrl-I"
      }
    },
    "nbTranslate": {
      "hotkey": "alt-t",
      "sourceLang": "en",
      "targetLang": "fr",
      "displayLangs": [
        "*"
      ],
      "langInMainMenu": true,
      "useGoogleTranslate": true
    },
    "toc": {
      "nav_menu": {},
      "number_sections": true,
      "sideBar": true,
      "skip_h1_title": false,
      "base_numbering": 1,
      "title_cell": "Table of Contents",
      "title_sidebar": "Contents",
      "toc_cell": false,
      "toc_position": {},
      "toc_section_display": true,
      "toc_window_display": false
    },
    "gist": {
      "id": "",
      "data": {
        "description": "neural_network.py.ipynb",
        "public": true
      }
    }
  },
  "nbformat": 4,
  "nbformat_minor": 2
}