This file is located at uTensor/utensor-mnist-demo

deep_mlp.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-04-16T14:29:56.593505Z",
     "start_time": "2018-04-16T14:29:52.879458Z"
    }
   },
   "outputs": [],
   "source": [
    "# This script is based on:\n",
    "# https://www.tensorflow.org/get_started/mnist/pros\n",
    "\n",
    "import sys\n",
    "import tensorflow as tf\n",
    "from tensorflow.examples.tutorials.mnist import input_data\n",
    "from tensorflow.python.framework import graph_util as gu\n",
    "from tensorflow.python.framework.graph_util import remove_training_nodes\n",
    "from tensorflow.tools.graph_transforms import TransformGraph"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Import training data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-04-16T14:29:57.272673Z",
     "start_time": "2018-04-16T14:29:56.595420Z"
    }
   },
   "outputs": [],
   "source": [
    "mnist = input_data.read_data_sets(\"mnist_data/\", one_hot=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Define Tensorflow Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-04-16T14:29:57.278092Z",
     "start_time": "2018-04-16T14:29:57.274735Z"
    }
   },
   "outputs": [],
   "source": [
    "batch_size = 50"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Fully connected 2 layer NN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-04-16T14:29:57.288994Z",
     "start_time": "2018-04-16T14:29:57.280660Z"
    }
   },
   "outputs": [],
   "source": [
    "def deepnn(x):\n",
    "  W_fc1 = weight_variable([784, 128], name='W_fc1')\n",
    "  b_fc1 = bias_variable([128], name='b_fc1')\n",
    "  a_fc1 = tf.add(tf.matmul(x, W_fc1), b_fc1, name=\"zscore\")\n",
    "  h_fc1 = tf.nn.relu(a_fc1)\n",
    "\n",
    "  W_fc2 = weight_variable([128, 64], name='W_fc2')\n",
    "  b_fc2 = bias_variable([64], name='b_fc2')\n",
    "  a_fc2 = tf.add(tf.matmul(h_fc1, W_fc2), b_fc2, name=\"zscore\")\n",
    "  h_fc2 = tf.nn.relu(a_fc2)\n",
    "\n",
    "  W_fc3 = weight_variable([64, 10], name='W_fc3')\n",
    "  b_fc3 = bias_variable([10], name='b_fc3')\n",
    "  logits = tf.add(tf.matmul(h_fc2, W_fc3), b_fc3, name=\"logits\")\n",
    "  y_pred = tf.argmax(logits, 1, name='y_pred')\n",
    "\n",
    "  return y_pred, logits\n",
    "\n",
    "\n",
    "def weight_variable(shape, name):\n",
    "  \"\"\"weight_variable generates a weight variable of a given shape.\"\"\"\n",
    "  initial = tf.truncated_normal(shape, stddev=0.1)\n",
    "  return tf.Variable(initial, name)\n",
    "\n",
    "\n",
    "def bias_variable(shape, name):\n",
    "  \"\"\"bias_variable generates a bias variable of a given shape.\"\"\"\n",
    "  initial = tf.constant(0.1, shape=shape)\n",
    "  return tf.Variable(initial, name)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Specify inputs, outputs, and a cost function"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-04-16T14:29:57.440031Z",
     "start_time": "2018-04-16T14:29:57.291656Z"
    }
   },
   "outputs": [],
   "source": [
    "# Reset default graph\n",
    "tf.reset_default_graph()\n",
    "\n",
    "# Create the model\n",
    "x = tf.placeholder(tf.float32, [None, 784], name=\"x\")\n",
    "\n",
    "# Define loss and optimizer\n",
    "y_ = tf.placeholder(tf.float32, [None, 10], name=\"y\")\n",
    "\n",
    "# Build the graph for the deep net\n",
    "y_pred, logits = deepnn(x)\n",
    "\n",
    "with tf.name_scope(\"Loss\"):\n",
    "    cross_entropy = tf.nn.softmax_cross_entropy_with_logits_v2(labels=y_, \n",
    "                                                               logits=logits)\n",
    "    loss = tf.reduce_mean(cross_entropy, name=\"cross_entropy_loss\")\n",
    "train_step = tf.train.AdamOptimizer(1e-4).minimize(loss, name=\"train_step\")\n",
    "  \n",
    "# Here we specify the output as \"Prediction/y_pred\", this will be important later\n",
    "with tf.name_scope(\"Prediction\"): \n",
    "    correct_prediction = tf.equal(y_pred, \n",
    "                                  tf.argmax(y_, 1))\n",
    "    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32), name=\"accuracy\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Train the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-04-16T14:30:24.491091Z",
     "start_time": "2018-04-16T14:29:57.574591Z"
    }
   },
   "outputs": [],
   "source": [
    "sess = tf.Session()\n",
    "# Initialize the variables (i.e. assign their default value)\n",
    "sess.run(tf.global_variables_initializer())\n",
    "saver = tf.train.Saver()\n",
    "\n",
    "for i in range(1, 20001):\n",
    "  batch_images, batch_labels = mnist.train.next_batch(batch_size)\n",
    "  feed_dict = {x: batch_images, y_: batch_labels}\n",
    "  if i % 1000 == 0:\n",
    "    train_accuracy = sess.run(accuracy, feed_dict=feed_dict)\n",
    "    print('step %d, training accuracy %g' % (i, train_accuracy))\n",
    "  sess.run(train_step, feed_dict=feed_dict)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## What is the final accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-04-16T14:30:24.560096Z",
     "start_time": "2018-04-16T14:30:24.493809Z"
    }
   },
   "outputs": [],
   "source": [
    "print('test accuracy %g' % sess.run(accuracy, feed_dict={x: mnist.test.images, y_: mnist.test.labels}))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Freeze the graph"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-04-16T14:30:24.614593Z",
     "start_time": "2018-04-16T14:30:24.564158Z"
    }
   },
   "outputs": [],
   "source": [
    "saver.save(sess, \"./chkps/mnist_model\")\n",
    "out_nodes = [y_pred.op.name]\n",
    "print(out_nodes)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Remove unnecessary training nodes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-04-16T14:30:36.629747Z",
     "start_time": "2018-04-16T14:30:36.606952Z"
    }
   },
   "outputs": [],
   "source": [
    "sub_graph_def = remove_training_nodes(sess.graph_def)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Freeze Constants"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-04-16T14:30:37.568211Z",
     "start_time": "2018-04-16T14:30:37.552980Z"
    }
   },
   "outputs": [],
   "source": [
    "sub_graph_def = gu.convert_variables_to_constants(sess, sub_graph_def, out_nodes)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Save the graph to PB file"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-04-16T14:30:46.068240Z",
     "start_time": "2018-04-16T14:30:46.061283Z"
    }
   },
   "outputs": [],
   "source": [
    "graph_path = tf.train.write_graph(sub_graph_def,\n",
    "                                  \"./mnist_model\",\n",
    "                                  \"deep_mlp.pb\",\n",
    "                                  as_text=False)\n",
    "\n",
    "print('written graph to: %s' % graph_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-04-16T14:31:10.946225Z",
     "start_time": "2018-04-16T14:31:10.942746Z"
    }
   },
   "outputs": [],
   "source": [
    "# close session\n",
    "sess.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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"
  },
  "toc": {
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": "block",
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}