Open tomlincr opened 1 year ago
Apologies, long day and thought I'd opened this PR on my fork to test coverage, CI etc.
Interesting, all passes locally.
Seems to be some variation in the embeddings generated by multiple fits when run by actions.
Will test shape matches instead for these offenders
:warning: Please install the to ensure uploads and comments are reliably processed by Codecov.
All modified and coverable lines are covered by tests :white_check_mark:
Project coverage is 97.54%. Comparing base (
d750b33
) to head (716a796
). Report is 31 commits behind head on master.
:exclamation: Your organization needs to install the Codecov GitHub app to enable full functionality.
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
I have tried to run the test suite of this pull request, but it is currently failing at the HOPE model test. I see that you are comparing the two embeddings - maybe there are numerical instabilities that lead to different results over different runs? I am not familiar with the internals of numpy & scipy that much.
Added
.fit_transform
method to all node embedding algorithms, primarily motivated by desire to usekarateclub
algorithms in ascikit-learn
pipeline.Adds:
y=None
argument, for scikit-learn compatibilityif y is not None
to allow passing e.g. node attributes through for a downstream task in the pipelineTests:
.get_embedding()