{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# Neural Network with Eager API\n", "\n", "Build a 2-hidden layers fully connected neural network (a.k.a multilayer perceptron) with TensorFlow's Eager API.\n", "\n", "This example is using some of TensorFlow higher-level wrappers (tf.estimators, tf.layers, tf.metrics, ...), you can check 'neural_network_raw' example for a raw, and more detailed TensorFlow implementation.\n", "\n", "- Author: Aymeric Damien\n", "- Project: https://github.com/aymericdamien/TensorFlow-Examples/" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Neural Network Overview\n", "\n", "\"nn\"\n", "\n", "## MNIST Dataset Overview\n", "\n", "This example is using MNIST handwritten digits. The dataset contains 60,000 examples for training and 10,000 examples for testing. The digits have been size-normalized and centered in a fixed-size image (28x28 pixels) with values from 0 to 1. For simplicity, each image has been flattened and converted to a 1-D numpy array of 784 features (28*28).\n", "\n", "![MNIST Dataset](http://neuralnetworksanddeeplearning.com/images/mnist_100_digits.png)\n", "\n", "More info: http://yann.lecun.com/exdb/mnist/" ] }, { "cell_type": "code", "execution_count": 1, "metadata": { "collapsed": true }, "outputs": [], "source": [ "from __future__ import print_function\n", "\n", "import tensorflow as tf" ] }, { "cell_type": "code", "execution_count": 2, "metadata": { "collapsed": true }, "outputs": [], "source": [ "# Set Eager API\n", "tf.enable_eager_execution()\n", "tfe = tf.contrib.eager" ] }, { "cell_type": "code", "execution_count": 3, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Extracting /tmp/data/train-images-idx3-ubyte.gz\n", "Extracting /tmp/data/train-labels-idx1-ubyte.gz\n", "Extracting /tmp/data/t10k-images-idx3-ubyte.gz\n", "Extracting /tmp/data/t10k-labels-idx1-ubyte.gz\n" ] } ], "source": [ "# Import MNIST data\n", "from tensorflow.examples.tutorials.mnist import input_data\n", "mnist = input_data.read_data_sets(\"/tmp/data/\", one_hot=False)" ] }, { "cell_type": "code", "execution_count": 4, "metadata": { "collapsed": true }, "outputs": [], "source": [ "# Parameters\n", "learning_rate = 0.001\n", "num_steps = 1000\n", "batch_size = 128\n", "display_step = 100\n", "\n", "# Network Parameters\n", "n_hidden_1 = 256 # 1st layer number of neurons\n", "n_hidden_2 = 256 # 2nd layer number of neurons\n", "num_input = 784 # MNIST data input (img shape: 28*28)\n", "num_classes = 10 # MNIST total classes (0-9 digits)" ] }, { "cell_type": "code", "execution_count": 5, "metadata": { "collapsed": true }, "outputs": [], "source": [ "# Using TF Dataset to split data into batches\n", "dataset = tf.data.Dataset.from_tensor_slices(\n", " (mnist.train.images, mnist.train.labels))\n", "dataset = dataset.repeat().batch(batch_size).prefetch(batch_size)\n", "dataset_iter = tfe.Iterator(dataset)" ] }, { "cell_type": "code", "execution_count": 6, "metadata": { "collapsed": true }, "outputs": [], "source": [ "# Define the neural network. To use eager API and tf.layers API together,\n", "# we must instantiate a tfe.Network class as follow:\n", "class NeuralNet(tfe.Network):\n", " def __init__(self):\n", " # Define each layer\n", " super(NeuralNet, self).__init__()\n", " # Hidden fully connected layer with 256 neurons\n", " self.layer1 = self.track_layer(\n", " tf.layers.Dense(n_hidden_1, activation=tf.nn.relu))\n", " # Hidden fully connected layer with 256 neurons\n", " self.layer2 = self.track_layer(\n", " tf.layers.Dense(n_hidden_2, activation=tf.nn.relu))\n", " # Output fully connected layer with a neuron for each class\n", " self.out_layer = self.track_layer(tf.layers.Dense(num_classes))\n", "\n", " def call(self, x):\n", " x = self.layer1(x)\n", " x = self.layer2(x)\n", " return self.out_layer(x)\n", "\n", "\n", "neural_net = NeuralNet()" ] }, { "cell_type": "code", "execution_count": 7, "metadata": { "collapsed": true }, "outputs": [], "source": [ "# Cross-Entropy loss function\n", "def loss_fn(inference_fn, inputs, labels):\n", " # Using sparse_softmax cross entropy\n", " return tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(\n", " logits=inference_fn(inputs), labels=labels))\n", "\n", "\n", "# Calculate accuracy\n", "def accuracy_fn(inference_fn, inputs, labels):\n", " prediction = tf.nn.softmax(inference_fn(inputs))\n", " correct_pred = tf.equal(tf.argmax(prediction, 1), labels)\n", " return tf.reduce_mean(tf.cast(correct_pred, tf.float32))\n", "\n", "\n", "# SGD Optimizer\n", "optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)\n", "\n", "# Compute gradients\n", "grad = tfe.implicit_gradients(loss_fn)" ] }, { "cell_type": "code", "execution_count": 8, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Initial loss= 2.340397596\n", "Step: 0001 loss= 2.340397596 accuracy= 0.0703\n", "Step: 0100 loss= 0.586046159 accuracy= 0.8305\n", "Step: 0200 loss= 0.253318846 accuracy= 0.9282\n", "Step: 0300 loss= 0.214748293 accuracy= 0.9377\n", "Step: 0400 loss= 0.180644721 accuracy= 0.9466\n", "Step: 0500 loss= 0.137285724 accuracy= 0.9591\n", "Step: 0600 loss= 0.119845696 accuracy= 0.9636\n", "Step: 0700 loss= 0.113618039 accuracy= 0.9665\n", "Step: 0800 loss= 0.109642141 accuracy= 0.9676\n", "Step: 0900 loss= 0.085067607 accuracy= 0.9746\n", "Step: 1000 loss= 0.079819344 accuracy= 0.9754\n" ] } ], "source": [ "# Training\n", "average_loss = 0.\n", "average_acc = 0.\n", "for step in range(num_steps):\n", "\n", " # Iterate through the dataset\n", " d = dataset_iter.next()\n", " \n", " # Images\n", " x_batch = d[0]\n", " # Labels\n", " y_batch = tf.cast(d[1], dtype=tf.int64)\n", "\n", " # Compute the batch loss\n", " batch_loss = loss_fn(neural_net, x_batch, y_batch)\n", " average_loss += batch_loss\n", " # Compute the batch accuracy\n", " batch_accuracy = accuracy_fn(neural_net, x_batch, y_batch)\n", " average_acc += batch_accuracy\n", "\n", " if step == 0:\n", " # Display the initial cost, before optimizing\n", " print(\"Initial loss= {:.9f}\".format(average_loss))\n", "\n", " # Update the variables following gradients info\n", " optimizer.apply_gradients(grad(neural_net, x_batch, y_batch))\n", "\n", " # Display info\n", " if (step + 1) % display_step == 0 or step == 0:\n", " if step > 0:\n", " average_loss /= display_step\n", " average_acc /= display_step\n", " print(\"Step:\", '%04d' % (step + 1), \" loss=\",\n", " \"{:.9f}\".format(average_loss), \" accuracy=\",\n", " \"{:.4f}\".format(average_acc))\n", " average_loss = 0.\n", " average_acc = 0." ] }, { "cell_type": "code", "execution_count": 9, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Testset Accuracy: 0.9719\n" ] } ], "source": [ "# Evaluate model on the test image set\n", "testX = mnist.test.images\n", "testY = mnist.test.labels\n", "\n", "test_acc = accuracy_fn(neural_net, testX, testY)\n", "print(\"Testset Accuracy: {:.4f}\".format(test_acc))" ] } ], "metadata": { "kernelspec": { "display_name": "Python 2", "language": "python", "name": "python2" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 2 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython2", "version": "2.7.14" } }, "nbformat": 4, "nbformat_minor": 1 }