Still getting blanks in translate.py output about opennmt-py HOT 6 CLOSED

are you working on a translation task or something else ?
I am mostly working on the transformer model, but I was also able to train and translate with a LSTM one.

from opennmt-py.

It's a "translation" task, though I'm using the model for classification (by translating a sentence to a category of 1 or 0), rather than translation. It experimental, but it works pretty well in our use case.

from opennmt-py.

are you saying the target is always 1 or 0 for an input=a sentence ?
and you mean you have empty output at times ?
did it ever work 100% with an output being never empty (with openNMT-py or another toolkit) ?

from opennmt-py.

are you saying the target is always 1 or 0 for an input=a sentence ?

That is correct.

and you mean you have empty output at times ?

That's also correct.

did it ever work 100% with an output being never empty (with openNMT-py or another toolkit) ?

I also had this problem when running this particular data model through openNMT-py version 0.3.1.post2.
When run with a different data model (when actually translating a sentence to another sentence), I did not have this issue when running a previous version of openNMT-py, but I'm not sure which version we were using in that case (most likely version 0.3.1.post2 or older).
I have not evaluated the latest version of openNMT-py on a sentence-to-sentence model to determine if it also results in blank lines in the output of running translate.py.

from opennmt-py.

you could try 2 things:

1- beam_size 1: greedy search may help in your case
2- translate with -min_length 1

from opennmt-py.

closing this, reopen if needed.

from opennmt-py.