Mixail Petrochuk

Ideas

16

In order to easily create a new process, the configuration should we exported from the old process, and imported into the new process.

Update package name

3

Add this quote to the documentation: > Defaults are arguably the most important design decisions you'll ever make as a software developer. https://blog.codinghorror.com/the-power-of-defaults/ Rename this package to `Defaults`, it makes...

This error... > The decorator was not executed immediately before `%s` at (%s:%s); therefore, it\'s `HParams` may not have been injected. is triggered in a multithreaded environment despite `setprofile` not...

At the moment, the lazy loading features of HParams are not documented. It'd be nice to have basic documentation on them.

Add BPE Text Encoder

3

Tokenizer Reference: https://github.com/eladhoffer/seq2seq.pytorch/blob/master/seq2seq/tools/tokenizer.py

Make `check_files` more generic

1

Line: https://github.com/PetrochukM/PyTorch-NLP/blob/be1a08edbd898d30273b7062c9c16fcf7f788787/torchnlp/download.py#L126 Instead of `check_files`, we could check that particular path exist. That'll enable us to check for particular directories, as well.

Add GLUE datasets

7

GLUE datasets are standard for evaluating NLU tasks. > In pursuit of this objective, we introduce the General Language Understanding Evaluation benchmark > (GLUE), a tool for evaluating and analyzing...

Write blog post

1

## Expected Behavior PyTorch-NLP might be hard to understand if you are starting a NLP project, I expect a blog post to exist to help fresh faces. Example links: https://hn.algolia.com/?query=nlp&sort=byPopularity&prefix&page=0&dateRange=all&type=story...

`fork_rng_wrap` does not propagate function arguments `fork_rng_wrap`, and it should.

Simplify `Encoder`: Special Tokens, OOB, Batch Encoding

1

The `Encoder` is unnecessarily complex: - It has OOB that's hard to reason with - The `Encoder` implementation is coupled with special tokens - The `batch_encoding` isn't helpful We could...