On 02/19/2015 10:04 PM, Tim Head wrote: > > > Yes! Thought for me roc_auc_score complains if I pass the result of > predict_proba (wrong shape) but the output of decision_function() > works. Which makes sense. Presumably if you used only the last column > of the predict_proba() output it would also work.
Yes, that was what I meant. Sorry for being unclear. > > Not sure either, as soon as I read it I went: "Aahh, yes obviously Tim > you muppet". I do't think there is a good way for roc_auc_score() to > detect/guess that the user is being an idiot and passed a y_score > which contains only two distinct values. I thought about trying to detect what the number of unique values is, but Gael argued against it because it is not super robust and we can't forever hold people's hands ;) Having scorers as the default interface (that is what cross-validation uses) instead of the scoring functions, would kind of prevent what you did. Not sure that is a good idea, though. ------------------------------------------------------------------------------ Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server from Actuate! Instantly Supercharge Your Business Reports and Dashboards with Interactivity, Sharing, Native Excel Exports, App Integration & more Get technology previously reserved for billion-dollar corporations, FREE http://pubads.g.doubleclick.net/gampad/clk?id=190641631&iu=/4140/ostg.clktrk _______________________________________________ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
