TensorBoard：可视化学习

def variable_summaries(var):

  """Attach a lot of summaries to a Tensor (for TensorBoard visualization)."""

  with tf.name_scope('summaries'):

    mean = tf.reduce_mean(var)

    tf.summary.scalar('mean', mean)

    with tf.name_scope('stddev'):

      stddev = tf.sqrt(tf.reduce_mean(tf.square(var - mean)))

    tf.summary.scalar('stddev', stddev)

    tf.summary.scalar('max', tf.reduce_max(var))

    tf.summary.scalar('min', tf.reduce_min(var))

    tf.summary.histogram('histogram', 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)

    with tf.name_scope('biases'):

      biases = bias_variable([output_dim])

      variable_summaries(biases)

    with tf.name_scope('Wx_plus_b'):

      preactivate = tf.matmul(input_tensor, weights) + biases

      tf.summary.histogram('pre_activations', preactivate)

    activations = act(preactivate, name='activation')

    tf.summary.histogram('activations', activations)

    return activations



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



with tf.name_scope('dropout'):

  keep_prob = tf.placeholder(tf.float32)

  tf.summary.scalar('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.losses.sparse_softmax_cross_entropy on the

  # raw logit outputs of the nn_layer above.

  with tf.name_scope('total'):

    cross_entropy = tf.losses.sparse_softmax_cross_entropy(labels=y_, logits=y)

tf.summary.scalar('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.summary.scalar('accuracy', accuracy)



# Merge all the summaries and write them out to /tmp/mnist_logs (by default)

merged = tf.summary.merge_all()

train_writer = tf.summary.FileWriter(FLAGS.summaries_dir + '/train',

                                      sess.graph)

test_writer = tf.summary.FileWriter(FLAGS.summaries_dir + '/test')

tf.global_variables_initializer().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

    summary, _ = sess.run([merged, train_step], feed_dict=feed_dict(True))

    train_writer.add_summary(summary, i)