TensorFlow学习笔记（二）---MNIST代码分析

TensorFlow学习笔记（二）---MNIST代码分析

1、mnist_softmax.py全部 代码如下：

"""A very simple MNIST classifier.
See extensive documentation at
http://tensorflow.org/tutorials/mnist/beginners/index.md
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import argparse

# Import data
from tensorflow.examples.tutorials.mnist import input_data

import tensorflow as tf

FLAGS = None



def main(_):
  mnist = input_data.read_data_sets(FLAGS.data_dir, one_hot=True)

  # Create the model
  x = tf.placeholder(tf.float32, [None, 784])
  W = tf.Variable(tf.zeros([784, 10]))
  b = tf.Variable(tf.zeros([10]))
  y = tf.matmul(x, W) + b

  # Define loss and optimizer
  y_ = tf.placeholder(tf.float32, [None, 10])

  # The raw formulation of cross-entropy,
  #
  #   tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(tf.softmax(y)),
  #                                 reduction_indices=[1]))
  #
  # can be numerically unstable.
  #
  # So here we use tf.nn.softmax_cross_entropy_with_logits on the raw
  # outputs of 'y', and then average across the batch.
  cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(y, y_))
  train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)

  sess = tf.InteractiveSession()
  # Train
  tf.initialize_all_variables().run()
  for _ in range(1000):
    batch_xs, batch_ys = mnist.train.next_batch(100)
    sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})

  # Test trained model
  correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
  accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
  print(sess.run(accuracy, feed_dict={x: mnist.test.images,
                                      y_: mnist.test.labels}))

if __name__ == '__main__':
  parser = argparse.ArgumentParser()
  parser.add_argument('--data_dir', type=str, default='/tmp/data',
                      help='Directory for storing data')
  FLAGS = parser.parse_args()
  tf.app.run()

2、mnist_with_summaries.py全部代码如下：

"""A simple MNIST classifier which displays summaries in TensorBoard.
 This is an unimpressive MNIST model, but it is a good example of using
tf.name_scope to make a graph legible in the TensorBoard graph explorer, and of
naming summary tags so that they are grouped meaningfully in TensorBoard.
It demonstrates the functionality of every TensorBoard dashboard.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import argparse

import tensorflow as tf

from tensorflow.examples.tutorials.mnist import input_data

FLAGS = None



def train():
  # Import data
  mnist = input_data.read_data_sets(FLAGS.data_dir,
                                    one_hot=True,
                                    fake_data=FLAGS.fake_data)

  sess = tf.InteractiveSession()

  # Create a multilayer model.

  # Input placeholders
  with tf.name_scope('input'):
    x = tf.placeholder(tf.float32, [None, 784], name='x-input')
    y_ = tf.placeholder(tf.float32, [None, 10], name='y-input')

  with tf.name_scope('input_reshape'):
    image_shaped_input = tf.reshape(x, [-1, 28, 28, 1])
    tf.image_summary('input', image_shaped_input, 10)

  # We can't initialize these variables to 0 - the network will get stuck.
  def weight_variable(shape):
    """Create a weight variable with appropriate initialization."""
    initial = tf.truncated_normal(shape, stddev=0.1)
    return tf.Variable(initial)

  def bias_variable(shape):
    """Create a bias variable with appropriate initialization."""
    initial = tf.constant(0.1, shape=shape)
    return tf.Variable(initial)

  def variable_summaries(var, name):
    """Attach a lot of summaries to a Tensor."""
    with tf.name_scope('summaries'):
      mean = tf.reduce_mean(var)
      tf.scalar_summary('mean/' + name, mean)
      with tf.name_scope('stddev'):
        stddev = tf.sqrt(tf.reduce_mean(tf.square(var - mean)))
      tf.scalar_summary('stddev/' + name, stddev)
      tf.scalar_summary('max/' + name, tf.reduce_max(var))
      tf.scalar_summary('min/' + name, tf.reduce_min(var))
      tf.histogram_summary(name, var)

  def nn_layer(input_tensor, input_dim, output_dim, layer_name, act=tf.nn.relu):
    """Reusable code for making a simple neural net layer.
    It does a matrix multiply, bias add, and then uses relu to nonlinearize.
    It also sets up name scoping so that the resultant graph is easy to read,
    and adds a number of summary ops.
    """
    # Adding a name scope ensures logical grouping of the layers in the graph.
    with tf.name_scope(layer_name):
      # This Variable will hold the state of the weights for the layer
      with tf.name_scope('weights'):
        weights = weight_variable([input_dim, output_dim])
        variable_summaries(weights, layer_name + '/weights')
      with tf.name_scope('biases'):
        biases = bias_variable([output_dim])
        variable_summaries(biases, layer_name + '/biases')
      with tf.name_scope('Wx_plus_b'):
        preactivate = tf.matmul(input_tensor, weights) + biases
        tf.histogram_summary(layer_name + '/pre_activations', preactivate)
      activations = act(preactivate, name='activation')
      tf.histogram_summary(layer_name + '/activations', activations)
      return activations

  hidden1 = nn_layer(x, 784, 500, 'layer1')

  with tf.name_scope('dropout'):
    keep_prob = tf.placeholder(tf.float32)
    tf.scalar_summary('dropout_keep_probability', keep_prob)
    dropped = tf.nn.dropout(hidden1, keep_prob)

  # Do not apply softmax activation yet, see below.
  y = nn_layer(dropped, 500, 10, 'layer2', act=tf.identity)

  with tf.name_scope('cross_entropy'):
    # The raw formulation of cross-entropy,
    #
    # tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(tf.softmax(y)),
    #                               reduction_indices=[1]))
    #
    # can be numerically unstable.
    #
    # So here we use tf.nn.softmax_cross_entropy_with_logits on the
    # raw outputs of the nn_layer above, and then average across
    # the batch.
    diff = tf.nn.softmax_cross_entropy_with_logits(y, y_)
    with tf.name_scope('total'):
      cross_entropy = tf.reduce_mean(diff)
    tf.scalar_summary('cross entropy', cross_entropy)

  with tf.name_scope('train'):
    train_step = tf.train.AdamOptimizer(FLAGS.learning_rate).minimize(
        cross_entropy)

  with tf.name_scope('accuracy'):
    with tf.name_scope('correct_prediction'):
      correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
    with tf.name_scope('accuracy'):
      accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
    tf.scalar_summary('accuracy', accuracy)

  # Merge all the summaries and write them out to /tmp/mnist_logs (by default)
  merged = tf.merge_all_summaries()
  train_writer = tf.train.SummaryWriter(FLAGS.summaries_dir + '/train',
                                        sess.graph)
  test_writer = tf.train.SummaryWriter(FLAGS.summaries_dir + '/test')
  tf.initialize_all_variables().run()

  # Train the model, and also write summaries.
  # Every 10th step, measure test-set accuracy, and write test summaries
  # All other steps, run train_step on training data, & add training summaries

  def feed_dict(train):
    """Make a TensorFlow feed_dict: maps data onto Tensor placeholders."""
    if train or FLAGS.fake_data:
      xs, ys = mnist.train.next_batch(100, fake_data=FLAGS.fake_data)
      k = FLAGS.dropout
    else:
      xs, ys = mnist.test.images, mnist.test.labels
      k = 1.0
    return {x: xs, y_: ys, keep_prob: k}

  for i in range(FLAGS.max_steps):
    if i % 10 == 0:  # Record summaries and test-set accuracy
      summary, acc = sess.run([merged, accuracy], feed_dict=feed_dict(False))
      test_writer.add_summary(summary, i)
      print('Accuracy at step %s: %s' % (i, acc))
    else:  # Record train set summaries, and train
      if i % 100 == 99:  # Record execution stats
        run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
        run_metadata = tf.RunMetadata()
        summary, _ = sess.run([merged, train_step],
                              feed_dict=feed_dict(True),
                              options=run_options,
                              run_metadata=run_metadata)
        train_writer.add_run_metadata(run_metadata, 'step%03d' % i)
        train_writer.add_summary(summary, i)
        print('Adding run metadata for', i)
      else:  # Record a summary
        summary, _ = sess.run([merged, train_step], feed_dict=feed_dict(True))
        train_writer.add_summary(summary, i)
  train_writer.close()
  test_writer.close()



def main(_):
  if tf.gfile.Exists(FLAGS.summaries_dir):
    tf.gfile.DeleteRecursively(FLAGS.summaries_dir)
  tf.gfile.MakeDirs(FLAGS.summaries_dir)
  train()



if __name__ == '__main__':
  parser = argparse.ArgumentParser()
  parser.add_argument('--fake_data', nargs='?', const=True, type=bool,
                      default=False,
                      help='If true, uses fake data for unit testing.')
  parser.add_argument('--max_steps', type=int, default=1000,
                      help='Number of steps to run trainer.')
  parser.add_argument('--learning_rate', type=float, default=0.001,
                      help='Initial learning rate')
  parser.add_argument('--dropout', type=float, default=0.9,
                      help='Keep probability for training dropout.')
  parser.add_argument('--data_dir', type=str, default='/tmp/data',
                      help='Directory for storing data')
  parser.add_argument('--summaries_dir', type=str, default='/tmp/mnist_logs',
                      help='Summaries directory')
  FLAGS = parser.parse_args()
  tf.app.run()

（待续）