Closed percevalw closed 9 months ago
All modified and coverable lines are covered by tests :white_check_mark:
Comparison is base (
c918130
) 96.58% compared to head (8134c30
) 96.59%.
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
Description
Fixes #234
Since edsnlp 0.10.0, some of our components require optional dependencies (torch, transformers, etc). By default,
pip install edsnlp
does not install these, so importing one of these components (e.g.,eds.transformer
) will fail.It is not an issue when building a pipeline via
edsnlp.blank(...)
since we only load entry points when required, so a user that does not add these components will face no issue. However, as demonstrated by #234, when usingspacy.blank
orspacy.load
spacy will look into all spacy_factories entry point and try to load them all, and fail when encountering one of the components mentioned previously.To solve that, we split our components between two namespaces,
spacy_factories
and a newedsnlp_factories
namespace : now spacy will only look in the first namespace, while edsnlp will look in both.Checklist