scikit-learn-contrib/DESlib

For LCA, what if the samples with the class is not amongst the neighbour?

jayahm opened this issue · 3 comments

I'm wondering for LCA.

Since the local accuracy is estimated in respect of output classes assigned by a particular classifier, what if the local region of a test sample does not have the validation sample with that class label in the region?

Is this possible to happen?

If yes, how your implementation for LCA can handle this?

@Menelau

May I know which part of the library is doing so?

In the estimate_competence function line 180:

competences = np.ma.filled(competences_masked, 0)