Thanks, Gael, that's very useful information. I will do some hyperparameter tuning via GridSearch on the alpha and priors then for using roc_auc as scoring metric and see how it goes.
Best, Sebastian On Aug 31, 2014, at 5:10 PM, Gael Varoquaux <[email protected]> wrote: > On Sat, Aug 30, 2014 at 03:53:24PM -0400, Sebastian Raschka wrote: >> I was wondering if it somehow possible to define a loss function to the >> Naive Bayes classifier in scikit-learn. > > No. > >> For example, let's assume that we are interested in spam vs. ham >> classification. In this context, such a loss function would be useful >> to lower the False Positive rate (i.e., classifying ham as spam, which >> is "worse" than classifying spam as ham) > > You can shift the ratio of errors in one class vs the other by accessing > directly the output of 'predict_proba' and thresholding at a different > value than equal probability for each class. > > HTH, > > Gael > > ------------------------------------------------------------------------------ > Slashdot TV. > Video for Nerds. Stuff that matters. > http://tv.slashdot.org/ > _______________________________________________ > Scikit-learn-general mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general ------------------------------------------------------------------------------ Slashdot TV. Video for Nerds. Stuff that matters. http://tv.slashdot.org/ _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
