tensorflow的用法：Tensorflow Debugger的使用

╰半夏微凉° 2024-04-18 11:30 60阅读 0赞

参考[https://blog.csdn.net/u010312436/article/details/78656723/][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：  
![在这里插入图片描述][20190906090207205.png]  
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：  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpZGljaGVuZ2ZvMDQxMg_size_16_color_FFFFFF_t_70]  
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：  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpZGljaGVuZ2ZvMDQxMg_size_16_color_FFFFFF_t_70 1]  
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：  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpZGljaGVuZ2ZvMDQxMg_size_16_color_FFFFFF_t_70 2]  
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：  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpZGljaGVuZ2ZvMDQxMg_size_16_color_FFFFFF_t_70 3]  
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  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpZGljaGVuZ2ZvMDQxMg_size_16_color_FFFFFF_t_70 4]  
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：

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpZGljaGVuZ2ZvMDQxMg_size_16_color_FFFFFF_t_70 5]  
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：  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpZGljaGVuZ2ZvMDQxMg_size_16_color_FFFFFF_t_70 6]  
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：  
![在这里插入图片描述][20190906095451839.png]  
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

[https_blog.csdn.net_u010312436_article_details_78656723]: https://blog.csdn.net/u010312436/article/details/78656723/
[20190906090207205.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/18/fc7d064036dd4796ab659e91c2c4921c.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpZGljaGVuZ2ZvMDQxMg_size_16_color_FFFFFF_t_70]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/18/b5161764ed5341c1bb7f41e5e6fd26b2.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpZGljaGVuZ2ZvMDQxMg_size_16_color_FFFFFF_t_70 1]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/18/8fa5b9932dd342029609792c5fc50912.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpZGljaGVuZ2ZvMDQxMg_size_16_color_FFFFFF_t_70 2]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/18/674af5bd34d943809a9992afa8df7ba6.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpZGljaGVuZ2ZvMDQxMg_size_16_color_FFFFFF_t_70 3]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/18/2b8f2192bc984fb6804bbee24fc9e9dd.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpZGljaGVuZ2ZvMDQxMg_size_16_color_FFFFFF_t_70 4]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/18/4f3ccebb71d04cabb955b9d75ef561da.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpZGljaGVuZ2ZvMDQxMg_size_16_color_FFFFFF_t_70 5]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/18/6dcaf5409635433e9b42458dcf77f746.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpZGljaGVuZ2ZvMDQxMg_size_16_color_FFFFFF_t_70 6]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/18/c01bb25621474ec887fde008ff7f5b5b.png
[20190906095451839.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/18/89ee78b4d5d84a3d8aaea063b6f6b814.png