Mark Neumann

bioul_to_spans doesn't handle mis-specified tag sequences

1

e.g `B-PER -> I-ORG -> L-MISC` is extracted as `(MISC, (0,2))`, despite being an invalid sequence. We want to make sure it extracts something that negatively affects the score, such...

I think this will be useful to test future releases before we release them.

https://docs.google.com/forms/d/e/1FAIpQLScS6BjlLLdF584RO0WlMwaSZP7iFRPyLdSxaPJuD54sfDtz8w/viewform?usp=sf_link

Refactor tests to not use global cache for models

3

We should move away from a functional test framework because it makes the "god-object-spacy-models" dangerous. Instead we should just load a single model for groups of related tests and re-use...

We have the ability to train an extremely strong dependency parser (via allennlp + the scientific version of elmo we have). We could generate a larger dataset of biomedical text...

Cog docs

2

Some web docs for Cog using [Mkdocs Material](https://squidfunk.github.io/mkdocs-material/). Currently the docs are kinda hard to discover, but this should help with that whilst allowing the docs to be: 1) kept...

Anonymised Variables should have consistent naming corresponding to their column

4

It's a little annoying that the anonymised variable names sometimes but not always correspond to the table/column name they come from. E.g in some datasets like academic, the variable name...

Fix for `self.embedding_dim` which is not a attribute of the module.

Got this mostly working locally, I just need to automate it a bit and tidy up some of the data fetching etc. Couple of things I want to try out...