本篇博客主要介紹下 tensorboard 的使用方法，tensorboard 是 tensorflow 中一個(gè)可視化訓(xùn)練過程中數(shù)據(jù)的工具，它不需要單獨(dú)安裝，tensorflow 安裝過程中已經(jīng)將其裝好了，它可以通過tensorflow程序運(yùn)行過程中產(chǎn)生的日志文件可視化tensorflow程序的運(yùn)行狀態(tài)，它和tensorflow程序跑在不同的進(jìn)程。下面基于官方的例子源碼來講解 mnist_with_summaries.py

編碼階段

1.添加關(guān)心的tensor或者Variable變量到tensorboard中

tf.summary.image 添加需要觀察的圖片信息

  with tf.name_scope('input_reshape'):#使用命名空間，將一些節(jié)點(diǎn)信息統(tǒng)一在一起，使計(jì)算圖看起來整潔
    image_shaped_input = tf.reshape(x, [-1, 28, 28, 1])
    tf.summary.image('input', image_shaped_input, 10) # 參數(shù)：name、tensor、max_outputs
    #max_outputs默認(rèn)是3，我們這里讓其多顯示幾張就寫成了10
    #使用命名空間后，image的名字類似：input_reshape/input/xxxxx  

tf.summary.scalar 添加需要觀察的變量信息

  # 定義一個(gè)對(duì)Variable變量（這里有weight和bias）的命名空間公共方法，并計(jì)算他們的mean、stddev
  # max、min、histogram等值并收集在Tensorboard中供用戶查看
  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) #參數(shù) ：name, values
      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)) #名字類似：xxx/summaries/max
      tf.summary.scalar('min', tf.reduce_min(var))
      tf.summary.histogram('histogram', var)

tf.summary.histogram 添加對(duì)變量或者tensor取值范圍的直方圖信息

      with tf.name_scope('Wx_plus_b'):
        preactivate = tf.matmul(input_tensor, weights) + biases
        tf.summary.histogram('pre_activations', preactivate) #參數(shù) ：name, values

2.匯總所有操作節(jié)點(diǎn)，并通過FileWriter創(chuàng)建保存運(yùn)行過程中信息的文件

tf.summary.merge_all 匯總所有節(jié)點(diǎn)操作，并定義兩個(gè)文件記錄器FileWriter

  #匯總所有操作，并定義兩個(gè)文件記錄器FileWriter
  merged = tf.summary.merge_all()
  train_writer = tf.summary.FileWriter(FLAGS.log_dir + '/train', sess.graph) #添加整個(gè)計(jì)算圖
  test_writer = tf.summary.FileWriter(FLAGS.log_dir + '/test')

3.訓(xùn)練或者測(cè)試過程中運(yùn)行匯總的節(jié)點(diǎn)merged，會(huì)產(chǎn)生運(yùn)行信息并將這些信息寫入上一步中創(chuàng)建的文件當(dāng)中

train_writer.add_summary 往文件中寫入信息

        #記錄訓(xùn)練時(shí)運(yùn)算時(shí)間和內(nèi)存占用情況
        run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE) #設(shè)置trace_level
        run_metadata = tf.RunMetadata() #定義tensorflow運(yùn)行元信息
        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)#添加訓(xùn)練元信息
        train_writer.add_summary(summary, i)

完整的代碼如下：

# coding=UTF-8
import argparse
import os
import sys

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,
                                    fake_data=FLAGS.fake_data)

  # 默認(rèn)的session，可以先構(gòu)建session后定義操作，如果使用tf.Session()需要在啟動(dòng)session之前構(gòu)建整個(gè)計(jì)算圖，
  # 然后啟動(dòng)該計(jì)算圖。它還可以直接在不聲明session的條件下直接使用run()，eval()
  sess = tf.InteractiveSession()


  # Create a multilayer model.

  # Input placeholders
  with tf.name_scope('input'): #使用命名空間，將一些節(jié)點(diǎn)信息統(tǒng)一在一起，使計(jì)算圖看起來整潔
    x = tf.placeholder(tf.float32, [None, 784], name='x-input')
    y_ = tf.placeholder(tf.int64, [None], name='y-input')

  with tf.name_scope('input_reshape'):
    image_shaped_input = tf.reshape(x, [-1, 28, 28, 1])
    tf.summary.image('input', image_shaped_input, 10) #使用命名空間后，image的名字類似：input_reshape/input/xxxxx

  # We can't initialize these variables to 0 - the network will get stuck.
  #模型參數(shù)初始化
  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)

  # 定義一個(gè)對(duì)Variable變量（這里有weight和bias）的命名空間公共方法，并計(jì)算他們的mean、stddev
  # max、min、histogram等值并收集在Tensorboard中供用戶查看
  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)) #名字類似：xxx/summaries/max
      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.
    """
    # 定義一個(gè)MLP多層神經(jīng)網(wǎng)絡(luò)來訓(xùn)練數(shù)據(jù)，包括：初始化weight和bias、做一個(gè)矩陣相乘再加上一個(gè)偏置項(xiàng)，然后經(jīng)過一個(gè)非線性
    #激活函數(shù)
    # 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') #使用前面定義的網(wǎng)絡(luò)

  #使用dropout
  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.
  # 這里激活函數(shù)用的是全等映射，即直接將輸入復(fù)制給輸出
  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, and then average across
    # the batch.
    with tf.name_scope('total'):
      #計(jì)算softmax和交叉熵
      cross_entropy = tf.losses.sparse_softmax_cross_entropy(
          labels=y_, logits=y)
  tf.summary.scalar('cross_entropy', cross_entropy)

  #使用Adam優(yōu)化器對(duì)損失進(jìn)行優(yōu)化
  with tf.name_scope('train'):
    train_step = tf.train.AdamOptimizer(FLAGS.learning_rate).minimize(
        cross_entropy)

  #統(tǒng)計(jì)正確率
  with tf.name_scope('accuracy'):
    with tf.name_scope('correct_prediction'):
      correct_prediction = tf.equal(tf.argmax(y, 1), y_)
    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
  # ./logs/(by default)
  #匯總所有操作，并定義兩個(gè)文件記錄器FileWriter
  merged = tf.summary.merge_all()
  train_writer = tf.summary.FileWriter(FLAGS.log_dir + '/train', sess.graph)
  test_writer = tf.summary.FileWriter(FLAGS.log_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
  #定義一個(gè)feed_dict函數(shù)來確定要訓(xùn)練數(shù)據(jù)還是測(cè)試數(shù)據(jù)
  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
        #記錄訓(xùn)練時(shí)運(yùn)算時(shí)間和內(nèi)存占用情況
        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)#添加訓(xùn)練元信息
        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() #記得關(guān)閉
  test_writer.close()


def main(_):
  if tf.gfile.Exists(FLAGS.log_dir):#文件存在就刪除，重新訓(xùn)練生成
    tf.gfile.DeleteRecursively(FLAGS.log_dir)
  tf.gfile.MakeDirs(FLAGS.log_dir)
  train()


if __name__ == '__main__':
  parser = argparse.ArgumentParser() #命令行參數(shù)解析，沒有默認(rèn)值就提示用戶輸入
  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="./mnist_data",
      help='Directory for storing input data')
  parser.add_argument(
      '--log_dir',
      type=str,
      default="./logs",
      help='Summaries log directory')
  FLAGS, unparsed = parser.parse_known_args()
  tf.app.run(main=main, argv=[sys.argv[0]] + unparsed)

TensorBoard 可視化文件生成

上面代碼運(yùn)行完成后會(huì)在train_writer和test_writer指定目錄下生成類似"events.out.tfevents.1513910245.N22411D1"這種的文件，然后通過命令行輸入命令

tensorboard --logdir=path/to/log-directory #注意這里只需要指定到生成文件的上一級(jí)目錄就可以了

會(huì)有如下提示：

F:\>tensorboard --logdir=./log
TensorBoard 0.4.0rc3 at http://N22411D1:6006 (Press CTRL+C to quit)

最后我們通過將" http://N22411D1:6006"輸入谷歌或者火狐瀏覽器就可以了。

TensorBoard 可視化文件分析

請(qǐng)放大查看原圖，圖中有注釋說明。

SCALARS

統(tǒng)計(jì)一些準(zhǔn)確率、損失函數(shù)、weight等單個(gè)值的變化趨勢(shì)

tensorboard_summary_scalars.PNG

IMAGES

顯示你指定的一些圖片信息

tensorboard_image.PNG

GRAPHS

顯示你定義的整個(gè)計(jì)算圖，包括計(jì)算圖里面每個(gè)節(jié)點(diǎn)的詳細(xì)信息，比如輸入輸出的shape是多少，內(nèi)存占用，計(jì)算時(shí)間占用，節(jié)點(diǎn)名稱等等

tensorflow_graphs.PNG

DISTRIBUTIONS

顯示你指定的一些模型參數(shù)隨著迭代次數(shù)增加的變化趨勢(shì)

tensorboard_distributions.PNG

HISTOGRAMS

顯示你指定的一些模型參數(shù)隨著迭代次數(shù)增加的變化趨勢(shì)

tensorboard_histograms.PNG

參考：《TensorFlow實(shí)戰(zhàn)》