Tensorboard keras debugger feature

[JonathanCMitchell]

I am unable to use the latest Tensorboard debugging feature with keras. When following the documentation posted here and elsewhere I was instructed to insert the code snippet below:

import keras
from tensorflow.python import debug as tf_debug
keras.backend.set_session(
    tf_debug.LocalCLIDebugWrapperSession(tf.Session()), "<USER>:7000")


<Model fit generator stuff here>

Then I run the command tensorboard --logdir=/logs/ --reload_interval=1 --debugger_port 7000 in a terminal window.

Then I run the program containing the model file and the fit_generator and I get this error

2018-04-02 15:00:01.435906: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1120] Creating TensorFlow device (/device:GPU:0) -> (device: 0, name: GeForce GTX 1060, pci bus id: 0000:01:00.0, compute capability: 6.1)
Traceback (most recent call last):
  File "coco.py", line 505, in <module>
    layers='heads')
  File "/home/jon/Documents/Mask_RCNN/model.py", line 2296, in train
    tf_debug.LocalCLIDebugWrapperSession(tf.Session()), "jon:7000")
TypeError: set_session() takes 1 positional argument but 2 were given

When I simply ommit the second argument to tf_debug.LocalCLIDebugWrapperSession(), I get the terminal output of tfdgb and it doesn't forward the port to tensorboard.

Is this a versioning issue?
==== Versions ===
Tensorboard v 1.7
Tensorflow v 1.4.0
Keras v 2.4.1
Python3.5
Ubuntu 16.04

When I look at the keras.callbacks.py file it is clear that the function set_session only contains one argument. You can see that here

[caisq]

@JonathanCMitchell LocalCLIDebugWrapperSession you are using is for the command-line interface, not the graphical debugger plugin. Can you try using TensorBoardDebugWrapperSession instead?

[nfelt]

@caisq Hmm should the example snippet be updated then?

tensorboard/tensorboard/plugins/debugger/tf_debugger_dashboard/tf-debugger-initial-dialog.html

Lines 74 to 75 in 998adf2

	keras.backend.set_session(
	tf_debug.LocalCLIDebugWrapperSession(tf.Session()), "[[_host]]:[[_port]]")