search

stanfordmlgroup / ngboost

Natural Gradient Boosting for Probabilistic Prediction
Apache License 2.0
1.65k stars 215 forks source link

NGBClassifier predict error #52

Closed juanramonua closed 4 years ago

juanramonua commented 4 years ago

Running your code from https://github.com/stanfordmlgroup/ngboost/blob/master/examples/classification.py I received the following error when I try

ngb.predict(X_test)

`--------------------------------------------------------------------------- AttributeError Traceback (most recent call last)

in () ----> 1 ngb.predict(X_test) /usr/local/lib/python3.6/dist-packages/ngboost/ngboost.py in predict(self, X) 133 def predict(self, X): 134 dist = self.pred_dist(X) --> 135 return list(dist.loc.flatten()) 136 137 def score(self, X, Y): AttributeError: 'NoneType' object has no attribute 'flatten' ----------------------------------------------------------------` `ngb.pred_dist(X_test)` works properly but the above error is also obtained when using the sklearn function with cross validation, "cross_validate()". Regards,
alejandroschuler commented 4 years ago

Good catch... I looked into it and this is happening because the way predict is implemented right now is a bit silly and only works for distributions that have a loc parameter.

The way to fix this, imo, is to define a predict method for all distributions that returns the conditional mean (i.e. the loc parameter for those we've implemented) for continuous distributions and the conditional mode (i.e. the most likely class) for discrete distributions. We can probably even use the underlying scipy distributions for this. I'll do that sometime soon.

In the meantime, you can easily get the predicted class by doing ngb.pred_dist(X_test).prob > 0.5.

alejandroschuler commented 4 years ago

Ok, I've fixed this issue with https://github.com/stanfordmlgroup/ngboost/commit/c05485187f1952b62797499ddbcc0b67618550aa. Please do play around with it and let me know if this now works for you and/or if you want other features.