tensorflow的用法：Tensorflow Debugger的使用

参考https://blog.csdn.net/u010312436/article/details/78656723/以及《TensorFlow技术解析与实践》这本书。
参考的这篇博客写得很详细，那我这里就写简单点以一个错误运行的MNSIT训练集为例，看看如何通过Tensorflow Debugger来找到出错的地方，并改正。文件debug_mnist.py代码如下：

#!/usr/bin/env python
# -*- coding: utf-8 -*-
# author：袁阳平 yyp 
# time:2019/9/5
# 写代码==我开心


# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Demo of the tfdbg curses CLI: Locating the source of bad numerical values.
The neural network in this demo is larged based on the tutorial at:
  tensorflow/examples/tutorials/mnist/mnist_with_summaries.py
But modifications are made so that problematic numerical values (infs and nans)
appear in nodes of the graph during training.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import argparse
import sys
import tempfile

import tensorflow

from tensorflow.examples.tutorials.mnist import input_data
from tensorflow.python import debug as tf_debug



tf = tensorflow.compat.v1


IMAGE_SIZE = 28
HIDDEN_SIZE = 500
NUM_LABELS = 10
RAND_SEED = 42


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

  def feed_dict(train):
    if train or FLAGS.fake_data:
      xs, ys = mnist.train.next_batch(FLAGS.train_batch_size,
                                      fake_data=FLAGS.fake_data)
    else:
      xs, ys = mnist.test.images, mnist.test.labels

    return {x: xs, y_: ys}

  sess = tf.InteractiveSession()

  # Create the MNIST neural network graph.

  # Input placeholders.
  with tf.name_scope("input"):
    x = tf.placeholder(
        tf.float32, [None, IMAGE_SIZE * IMAGE_SIZE], name="x-input")
    y_ = tf.placeholder(tf.float32, [None, NUM_LABELS], name="y-input")

  def weight_variable(shape):
    """Create a weight variable with appropriate initialization."""
    initial = tf.truncated_normal(shape, stddev=0.1, seed=RAND_SEED)
    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 nn_layer(input_tensor, input_dim, output_dim, layer_name, act=tf.nn.relu):
    """Reusable code for making a simple neural net layer."""
    # 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])
      with tf.name_scope("biases"):
        biases = bias_variable([output_dim])
      with tf.name_scope("Wx_plus_b"):
        preactivate = tf.matmul(input_tensor, weights) + biases

      activations = act(preactivate)
      return activations

  hidden = nn_layer(x, IMAGE_SIZE**2, HIDDEN_SIZE, "hidden")
  logits = nn_layer(hidden, HIDDEN_SIZE, NUM_LABELS, "output", tf.identity)
  y = tf.nn.softmax(logits)

  with tf.name_scope("cross_entropy"):
    # The following line is the culprit of the bad numerical values that appear
    # during training of this graph. Log of zero gives inf, which is first seen
    # in the intermediate tensor "cross_entropy/Log:0" during the 4th run()
    # call. A multiplication of the inf values with zeros leads to nans,
    # which is first in "cross_entropy/mul:0".
    #
    # You can use the built-in, numerically-stable implementation to fix this
    # issue:
    #   diff = tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=logits)

    diff = -(y_ * tf.log(y))
    with tf.name_scope("total"):
      cross_entropy = tf.reduce_mean(diff)

  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))

  sess.run(tf.global_variables_initializer())

  if FLAGS.debug and FLAGS.tensorboard_debug_address:
    raise ValueError(
        "The --debug and --tensorboard_debug_address flags are mutually "
        "exclusive.")
  if FLAGS.debug:
    # config_file_path = (tempfile.mktemp(".tfdbg_config")
    #                     if FLAGS.use_random_config_path else None)
    sess = tf_debug.LocalCLIDebugWrapperSession(
        sess,
        ui_type=FLAGS.ui_type)
    sess.add_tensor_filter("has_inf_or_nan", tf_debug.has_inf_or_nan)

  elif FLAGS.tensorboard_debug_address:
    sess = tf_debug.TensorBoardDebugWrapperSession(
        sess, FLAGS.tensorboard_debug_address)

  # Add this point, sess is a debug wrapper around the actual Session if
  # FLAGS.debug is true. In that case, calling run() will launch the CLI.
  for i in range(FLAGS.max_steps):
    acc = sess.run(accuracy, feed_dict=feed_dict(False))
    print("Accuracy at step %d: %s" % (i, acc))

    sess.run(train_step, feed_dict=feed_dict(True))


if __name__ == "__main__":
  parser = argparse.ArgumentParser()
  parser.register("type", "bool", lambda v: v.lower() == "true")
  parser.add_argument(
      "--max_steps",
      type=int,
      default=100,
      help="Number of steps to run trainer.")
  parser.add_argument(
      "--train_batch_size",
      type=int,
      default=100,
      help="Batch size used during training.")
  parser.add_argument(
      "--learning_rate",
      type=float,
      default=0.025,
      help="Initial learning rate.")
  parser.add_argument(
      "--data_dir",
      type=str,
      default="/tmp/mnist_data",
      help="Directory for storing data")
  parser.add_argument(
      "--ui_type",
      type=str,
      default="curses",
      help="Command-line user interface type (curses | readline)")
  parser.add_argument(
      "--fake_data",
      type="bool",
      nargs="?",
      const=True,
      default=False,
      help="Use fake MNIST data for unit testing")
  parser.add_argument(
      "--debug",
      type="bool",
      nargs="?",
      const=True,
      default=False,
      help="Use debugger to track down bad values during training. "
      "Mutually exclusive with the --tensorboard_debug_address flag.")
  parser.add_argument(
      "--tensorboard_debug_address",
      type=str,
      default=None,
      help="Connect to the TensorBoard Debugger Plugin backend specified by "
      "the gRPC address (e.g., localhost:1234). Mutually exclusive with the "
      "--debug flag.")
  parser.add_argument(
      "--use_random_config_path",
      type="bool",
      nargs="?",
      const=True,
      default=False,
      help="""If set, set config file path to a random file in the temporary
      directory.""")
  FLAGS, unparsed = parser.parse_known_args()
  with tf.Graph().as_default():
    tf.app.run(main=main, argv=[sys.argv[0]] + unparsed)

我们先不加调试器，直接执行，如下：

python mnist.py

经过10次训练，结果如下：

可以看出，准确率在第一次训练时有所上升，后面一直保持在较低的水平。
下面我们就用Tensorflow Debugger来尝试调试。仅在原有的文件中加入下面3行代码。

from tensorflow.python import debug as tf_debug  ## 导入debug模块
sess = tf_debug.LocalCLIDebugWrapperSession( sess, ui_type=FLAGS.ui_type)
sess.add_tensor_filter("has_inf_or_nan", tf_debug.has_inf_or_nan)  ## 这两行代码放在sess.run(tf.global_variables_initializer())之后，

通过添加的代码可以看出，我们为张量值注册了一个过滤器has_inf_or_nan，它能够判断出图的任何中间张量中是否有nan或inf值。
现在我们就开启调试模式（debug），来找出准确率无法提高的原因。首先执行：

python debug_mnist.py --debug=True

这时就进入了Debugger界面，如下图：

这就是我们说的运行开始的UI（run-start UI），在tfdbg>后面可以输入交互式的命令，如run（或者缩写r），就可以进入运行结束后UI（run-en UI），如下图：

从上图可以看出，数值没有异常。然后，可以通过如下命令连续运行10次。

tfdbg>  run -t 10

可以使用下面的命令直到找到在图形中的第一个nan或者inf值（这类似于调试中的打断点）：

tfdbg>  run -f has_inf_or_nan

结果如下：

根据上图，仔细看一下Softmax:0这个节点，方法如下：

tfdbg> pt Softmax:0

结果如下：

用ni命令的-t标志进行追溯，方法如下：

ni -t Sotfmax:0

追溯结果如下图所示

可以看出104行中y = tf.nn.softmax(logits)这里的问题，logits的值也是有问题。
查看cross_entropy/Log:0这个节点，方法如下：

tfdbg> pt Softmax:0

结果如下：

用ni命令的-t标志进行追溯，方法如下：

ni -t corss_entropy/Log

追溯结果如下：

从追溯结果可以看出，debug_mnist.py的117行是罪魁祸首

diff = -(y_ * tf.log(y))

而上述代码中的y正是由前面y = tf.nn.softmax(logits)产生，正是前面产生的值出了问题所以就导致了现在的问题，那么就要对y值进行裁剪，如下：

diff = -(y_ * tf.clip_by_value(y, 1e-8, 1.0))

对修改之后的代码运行10次，结果如下：

可以看出准确率不再在低值徘徊，恢复正常，虽然准确率也是很低，但是是因为比如训练次数较少。

下面简单讨论下这个这个问题，run -f has_inf_or_nan命令中的has_inf_or_nan的由来，可不可以用别的参数。
至于由来，是由sess.add_tensor_filter(“has_inf_or_nan”, tf_debug.has_inf_or_nan)中的"has_inf_or_nan"得来的。
可以用别的参数，那就是自己构建张量filter，以下面代码为例，可以查找值为0的张量：

def my_filter_callable(datum, tensor):
  # A filter that detects zero-valued scalars.
  return len(tensor.shape) == 0 and tensor == 0.0
 
sess.add_tensor_filter('my_filter', my_filter_callable)

后面使用的时候，用下列命令即可：

run -f my_filter