clf.predict or clf.predict_proba ?

[tomaszmf]

The explainer._predict_down() method, in line 116 has the following:

target_yhat = self.clf.predict(observation)

But for classification this is always cast to 1.0 or 0.0 (for binary classification), thus making the contributions sum up to 0.0 or 1.0 and not to the predicted probability of the model.

Should the aforementioned line not be changed into:

target_yhat = self.clf.predict_proba(observation)

?

thanks.

[bondyra]

Yeah, probably. Sadly, I am not accountable for maintenance of this package anymore.