TensorFlow-3-TensorBoard: Visualizing Learning

模型的保存和载入:

模型的保存:saver.save 保存各种构建模型graph的操作(矩阵相乘，sigmoid等等....)

saver = tf.train.Saver() # 生成saver

with tf.Session() as sess:

     sess.run(tf.global_variables_initializer()) # 先对模型初始化

     # 然后将数据丢入模型进行训练blablabla

     # 训练完以后，使用saver.save 来保存

     saver.save(sess, "save_path/file_name") #file_name如果不存在的话，会自动创建

模型载入:saver.restore

saver = tf.train.Saver()

with tf.Session() as sess:

#参数可以进行初始化，也可不进行初始化。即使初始化了，初始化的值也会被restore的值给覆盖 sess.run(tf.global_variables_initializer())

saver.restore(sess, "save_path/file_name") #会将已经保存的变量值resotre到 变量中。

TensorBoard的使用

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

记录 Learning rate, loss 随时间的变化, 并可以给标签,像'learning rate' 或者'loss function'.

tf.summary.histogram

像查看activation的分布,脱离某个特殊的层,或者gradients或者weights的变化

tf.summary.merge_all

结合所有记录的信息使用

tf.summary.FileWriter.

将summary的信息写到磁盘上.

可选择的传一个Graph到FileWriter中. 如果传入了, TensorBoard将会可视化你的图.

使用顺序:

1. tf.summary.scalar和tf.summary.histogram收集信息
2. 收集过后 tf.summary.merge_all
3. 初始化FileWriter
4. Session去run结合后的summary
5. FileWirter.add_summary(summary, i),添加summary

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))

# 1

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

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

# 2

merged = tf.summary.merge_all()

# 3

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

......

for i in range(FLAGS.max_steps):

     if i % 10 == 0: # Record summaries and test-set accuracy

          # 4

          summary, acc = sess.run([merged, accuracy], feed_dict=feed_dict(False))

          # 5

    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)

添加变量,dropout,添加层的写法

官方文档中提供的

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)

自己跳的坑

如果想要获得合并的信息,必须在run的时候,传输入进去,否则会报错:

InvalidArgumentError: You must feed a value for placeholder tensor 'Placeholder' with dtype float and shape [1000,625]

merged = tf.summary.merge_all()

for i in range(500):

  # 7. here to determine the keeping probability

  sess.run(train_step, feed_dict={xs: X_train, ys: y_train, keep_prob: 0.5}) # 解决问题

  if i % 50 == 0:

  # record loss

  train_result = sess.run(merged, feed_dict={xs: X_train, ys: y_train, keep_prob: 1}) # record result, so don't dropout

  test_result = sess.run(merged, feed_dict={xs: X_test, ys: y_test, keep_prob: 1})

  train_writer.add_summary(train_result, i)

  test_writer.add_summary(test_result, i)

当只获取一个tf.summary.scalar时,直接run就可以,可以不传输入:

dev_summary = tf.summary.scalar('dev_accuracy', accuracy)

dev_summary = session.run(dev_summary)

writer.add_summary(dev_summary,epoches)

writer.flush()

查看 TensorBoard的写法

writer = tf.summary.FileWriter("D://WorkSpace//DP_workspace//TensorflowTest//log",tf.get_default_graph()) 

进入命令行:

tensorboard --logdir=D://WorkSpace//DP_workspace//TensorflowTest//log # ！！must write in this way

进入chrome,输入:

http://localhost:6006/