the coder of batch normalization in keras about nmt-keras

[lvapeab]

Hi,
Yes, batch norm = 1 is the transform from the paper.

For debugging the code, maybe you should take a look to the Theano debugging help. I find useful the printing option. You can access to it with the Keras backend function.
E.g.:

x = K.printing('x=')(x) # x is a tensor whose value you want to print (when it is computed)

[hanyaqian]

And i find it is difficult to debug and check the intermediate results, do you have any good ideas?

Thanks!

[hanyaqian]

Thanks for your reply.