sylva
sylva copied to clipboard
NLP helper (spaCy) lacks proper support for complex prompts in regards to residency
Sylva uses spaCy to help parse arbitrary text for possible residency info.
This helper module can be found at: https://github.com/ppfeister/sylva/blob/master/src/sylva/helpers/nlp.py
To reproduce using the included test suite, developers can either run pytest --runxfail (detailed) or pytest -rx (minimal).
Two examples of currently failing prompts from the test suite:
- I've lived in both Boston and Bremen before
- I've vacationed in Manchester, but lived in Bremen and Boston
Note that the second example is expected to not return Manchester, only Bremen and Boston.
General improvements to this module (accuracy, match rate...) are always welcome.