SuffolkLITLab / FormFyxer

A tool for learning about and pre-processing forms
MIT License
11 stars 1 forks source link

Remove use of pickled (joblib) files for now #119

Closed nonprofittechy closed 11 months ago

nonprofittechy commented 11 months ago

pickling broke stuff when package versions changed; docassemble's packages depend on new versions and the pickled stuff was generated with an old version.

Once this is merged, we need to create associated PyPI package and set it as a dependency of the Weaver so installing the Weaver will use the fixed version of FormFyxer.

This is just a temporary fix. Long-term, we will:

  1. Figure out a simpler to update way to store the list of known field names, nouns for people, and rules for suffixes appended to people nouns. I suggest a YAML or JSON file.
  2. Switch to LLM-driven field normalization. This appears likely to perform a lot better. We can combine w/ the list of known field names.

Note the typing error isn't related to the changed code. I'll open a separate issue for that (#120). possibly was due to a change in mypy and not our code base?