padding mode, compute mask, more simple batch indexing

[pinesnow72]

For the issues below I raised and you replied, I opened this pull request.

19 #20 #21

[ydovzhenko]

Awesome! I'll get back to you in a bit; need to figure out why the tests aren't running.

[ydovzhenko]

Could you add the following to the PR:

• version bump to 0.1.1 here: https://github.com/kensho-technologies/bubs/blob/43d911d62de3af61ab629bebfa1c446e5bc0def9/bubs/__init__.py#L2
• modify current unit tests to account for the new batch shape, make sure they pass locally
• add a new unit test to check for correct behavior with both prepad=False and prepad=True

I'm working out the issue with new build not being triggered meanwhile - we can work in parallel on these.

[pinesnow72]

Could you add the following to the PR:

• version bump to 0.1.1 here: https://github.com/kensho-technologies/bubs/blob/43d911d62de3af61ab629bebfa1c446e5bc0def9/bubs/__init__.py#L2
• modify current unit tests to account for the new batch shape, make sure they pass locally
• add a new unit test to check for correct behavior with both prepad=False and prepad=True

I'm working out the issue with new build not being triggered meanwhile - we can work in parallel on these.

Ok, I forgot to modify unit tests and will modify them soon. I will add some code snippet to some existing unit tests for checking prepad=False and prepad=True In addition, I will remove char_to_int.py. Instead of it, I will add code (methods) to extract char_to_int dictionary (charmap) from Flair embeddings and store it into a pickle. And then that pickle would be loaded and passed to InputEncoder.