Categorical LF training stats need to be fixed!

snorkel-team / snorkel

A system for quickly generating training data with weak supervision

https://snorkel.org

Apache License 2.0

5.79k stars 859 forks source link

Categorical LF training stats need to be fixed! #800

Closed ajratner closed 6 years ago

kyrillp commented 6 years ago

Hello, do you mean csr_LabelMatrix.lf_stats(self, session, labels=None, est_accs=None) or GenerativeModel.learned_lf_stats(self)? I encountered the behavior that in the categorical setting I'm working in, csr_LabelMatrix.lf_stats gives values larger than 1.0 for coverage, overlaps and conflicts. GenerativeModel.learned_lf_stats seems to return reasonable fractions.

ajratner commented 6 years ago

@kyrillp yup- should be fixed by @regoldman's PR #804 just merged in!