Keras, Tensorflow: How to set breakpoint (debug) in custom layer when evaluating?

Question

I just want to do some numerical validation inside the custom layer.

Suppose we have a very simple custom layer:

class test_layer(keras.layers.Layer         


        

Answer 1

1. Yes. The call() method is only used to build the computational graph.

2. As to the debug. I prefer using TFDBG, which is a recommended debugging tool for tensorflow, although it doesn't provide break point functions.

For Keras, you can add these line to your script to use TFDBG

import tf.keras.backend as K
from tensorflow.python import debug as tf_debug
sess = K.get_session()
sess = tf_debug.LocalCLIDebugWrapperSession(sess)
K.set_session(sess)