facebookresearch / clara Goto Github PK

predict is not implemented in ClaraGibbs but would be very useful

In the example repo's example notebook, it is clear that when A > 1, the posterior of psi is only updated for labeler of index = 0, the rest of which does not get updated.

This seems to be because when self._init(ratings, labelers, true_ratings, scores) is called in ClaraGibbs.fit(), initializing assignment only works for the label of index = 0, the data field was not updated for other labeler indices.

I managed to replicate the same problems using my own dataset.

Simulator.py seems to be missing function generate_labeler_confusion_matrix, which is referred to in examples.ipynb.