Extra dependances

[ayasyrev]

• [* ] I've checked contribution guide.

Add "extra" options for pip install.

• tests
• optional
• loggers
• nlp

If OK, add info to documentation.
Now we can move "tensorboard" to loggers and create cv requrements file

[ayasyrev]

Now "extra" is: "optional", "nlp", "full" and "tests".
Later will be "cv", "pipelines", "audio".

[AlekseySh]

I suggest some changes. If we go this way, we will only have "full" and "nlp" pip tags for now. We will add "audio" and "cv" later, but not "optional" and "tests"

Also, let's update installation instruction. Please, check contribution guide to understand how to change something in Readme.

[AlekseySh]

@ayasyrev everything looks okay to me

let's finish with my comment above
and fix that numpy error, it should be simple

after that we are good to merge

[DaloroAT]

Do we need to add typings from transformers for NLP related modules (datasets, models)? Right now placeholders with Any type there. Or we can do it in separate PR.

[AlekseySh]

@DaloroAT

#596 (comment)

I've tried and failed in the previous PRs. Spent a few hours on that. Seems like I could not make work type hints when they created in if-else condition. Probably I can may it work... I dont'know

[DaloroAT]

I've tried and failed in the previous PRs. Spent a few hours on that. Seems like I could not make work type hints when they created in if-else condition. Probably I can may it work... I dont'know

Now you can import Tokenizers and models from transformers in codebase without if-else. It makes sense, because lib might be installed with [nlp] extra

[AlekseySh]

Now you can import Tokenizers and models from transformers in codebase without if-else. It makes sense, because lib might be installed with [nlp] extra

@DaloroAT
what if I installed library without nlp? I will get error when imports chain will be executed

[ayasyrev]

Return tensorboard to main requirements. It needs for short tests.