Problems with coverage penalty

[yftah89]

Hi :)
When I'm changing the coverage to true :
COVERAGE_PENALTY = True # Apply source coverage penalty
COVERAGE_NORM_FACTOR = 0.2 # Coverage penalty factor

I get the following error on evaluation time :
nmt-keras/src/keras-wrapper/keras_wrapper/cnn_model.py", line 2140, in predictBeamSearchNet
length_penalties = [1.0 for _ in len(samples)]
TypeError: 'int' object is not iterable

I don't know if its matter but the length penalty is still set to False.
I think that the same problem will return on line 2272 in the same file:
coverage_penalties = [0.0 for _ in len(samples)]

Thanks again for all of your help, I'm really appreciating it .

[lvapeab]

Hello @yftah89 ,

thank you for noticing it. I've fixed that bug (lvapeab/multimodal_keras_wrapper@3b2cc5f). You should update the Multimodal Keras Wrapper.

Cheers!

[yftah89]

Thanks :)
working just fine now.