2013/12/13 Eustache DIEMERT <eusta...@diemert.fr>: > Mmmm... if calibration seems to be a good fit for sklearn, I'll try to > review the different existing approaches and see if it's difficult to > implement the most useful/popular one(s). > > Any hint on that ? is isotonic regression the most used form or should we > have a look on other well-known techniques ?
You are talking about calibration of the probability estimates of binary or multiclass classifiers right? If so, yes isotonic regression and Platt scaling are the 2 most well known methods to do this: http://machinelearning.wustl.edu/mlpapers/paper_files/icml2005_Niculescu-MizilC05.pdf There is also this paper mentioned by Mathieu in the PR to tackle the multiclass case: http://www.research.ibm.com/people/z/zadrozny/kdd2002-Transf.pdf -- Olivier http://twitter.com/ogrisel - http://github.com/ogrisel ------------------------------------------------------------------------------ Rapidly troubleshoot problems before they affect your business. Most IT organizations don't have a clear picture of how application performance affects their revenue. With AppDynamics, you get 100% visibility into your Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics Pro! http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk _______________________________________________ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
