Bert-extractive-summarizer

This release addresses the issue with custom BERT models. Thanks to hdatteln for addressing the pull request.

* Fixed issue: https://github.com/dmmiller612/bert-extractive-summarizer/issues/17, which was disallowing custom models.
* Fixed issue with random state and summarizations
* allowed option for passing in different languages.

What's New?

In this release, a new simple flask service is added to perform ad-hoc summarizations. This can be easily ran through docker using the provided makefile (or docker commands). Instructions are provided in the README.

On top of the service, the newest transformers library is also used now (instead of pytorch_transformers).

The main update for this release is the library moved to HuggingFace's transformers' library. Some changes do break backwards compatibility, but should be minor. A list of the full changes are below.

* Migrated to the new Transformers library
* Added xlnet, distilbert-base-uncased, and xlm networks
* vector_size is no longer required for the Bert Parent.
* Deprecated the SingleModel class in favor of Summarizer