tensorboard --debug显示无信息

Question

我有Win7 x64，TensorFlow（GPU）1.3.0，TensorBoard 0.1.5，Python 3.6，CUDA 8.0。 Tensorboard运行良好，但是当我尝试--debug时，tensorboard表示没有额外的线/输出 这是对的吗？

D:>tensorboard --logdir=tr:C:\Users\Martin\tmp.tf\04_hist --debug

注意：“tr:”是Win的解决方法。

Answer 1

默认情况下，我在Windows操作系统中使用--debug选项，仅在出现警告时才会看到消息

(3.5-n) C:\soft\Miniconda3>tensorboard --logdi=test:D:\Projects\logs\2 --debug
TensorBoard 0.1.7 at http://PDESK:6006 (Press CTRL+C to quit)
W1013 13:42:00.471421 Reloader tf_logging.py:86] Found more than one graph event per run, or there was a metagraph containing a graph_def, as well as one or more graph events.  Overwriting the graph with the newest event.
W1013 13:42:00.510424 Reloader tf_logging.py:86] Found more than one graph event per run, or there was a metagraph containing a graph_def, as well as one or more graph events.  Overwriting the graph with the newest event.

Answer 2

大概-您想在tensorflow中使用tf_debug。

因此您想将--debug传递给运行tensorflow python代码的python命令。 python mnist.py

https://github.com/keisuke-nakata/keras-tfdbg-tutorial

对于Keras

from keras import backend as K


def set_debugger_session():
    sess = K.get_session()
    sess = tf_debug.LocalCLIDebugWrapperSession(sess)
    sess.add_tensor_filter('has_inf_or_nan', has_inf_or_nan)
    K.set_session(sess)

或查找 TensorBoardDebugWrapperSession -

sess = tf_debug.TensorBoardDebugWrapperSession(sess, "127.0.0.0:6006")

python debug_minst.py --debug触发tfdbg解释器。

因为您提到了tensorboard-您可能对tensorboard调试插件感兴趣。我还没有解决这个问题-但是，如果您无法解决这个问题-请在此处询问另一个问题。