On Mon, May 04, 2015 at 01:32:02PM +0200, federico vaggi wrote: > I think Gael makes a very strong argument, but I think the error should be as > explicit and informative as possible (for new users).
+1. Including suggesting the syntax X[:, np.newaxis], which is not trivial. G > On Fri, May 1, 2015 at 7:58 PM, Gael Varoquaux <gael.varoqu...@normalesup.org> > wrote: > I strongly advice raising an error. Very very very strongly. > Being lax about ambiguous inputs makes prototyping and interactive usage > easier: less typing, and the systems gets it right most of the time. > However, it makes production use and debugging complex code much harder. > Indeed, errors, that might not be related to a simple user error but > might be generated by a complex framework, do not lead to exceptions, but > to problems down the line. > We are not R. We require a bit more of typing, we don't have as many > shortcuts and magic syntax. But we can be used in production, on big > datasets. We can be used by people like Airbus to monitor failures of > part in planes [*], or by many others. > Yes beginners want things to 'just work', but in the long run, they are > thankful for a well-thought and strict specification. > Gaël > [*] > http://www.pyvideo.org/video/3519/ > scikit-learn-for-predictive-maintenance-at-airbus > On Fri, May 01, 2015 at 06:51:00PM +0100, Luca Puggini wrote: > > I vote for 3. > > On Fri, May 1, 2015 at 6:27 PM, Andreas Mueller <t3k...@gmail.com> > wrote: > > Hi all. > > A quick questions on future API. > > What should happen if a user passes an X with shape (N,), in other > words > > X.ndim == 1? > > This is unfortunately not really consistent in scikit-learn right > now. > > Three things are possible: > > 1) Raise an error > > 2) N = n_features, that is X contains a single sample > > 3) N = n_samples, that is X has a single feature > > I would think it should be N=n_samples. Gael thinks (iirc) we should > raise > > an error. > > In the code, we currently take N=n_features in predict, > decision_function, > > predict_proba and transform, basically everywhere. > > This is in part due to using ``check_array`` everywhere, which used > the > > backward-compatible (but odd) behavior of np.atleast2d. > > In ``fit``it looks like all estimators assume N=n_features, apart > from > > DictionaryLearning, MinMaxScaler, StandardScaler, which assume N= > n_samples. > > See https://github.com/scikit-learn/scikit-learn/pull/4511 for more > > discussion > > Obviously any change we make would mean a deprecation cycle, which > will > > mean warning in 0.17 and 0.18 when someone gives a 1-dim X that > we'll > > change something soon, and then actually change it in 0.19 (1.0?). > > Andy > > > > ------------------------------------------------------------------------------ > > One dashboard for servers and applications across > Physical-Virtual-Cloud > > Widest out-of-the-box monitoring support with 50+ applications > > Performance metrics, stats and reports that give you Actionable > Insights > > Deep dive visibility with transaction tracing using APM Insight. > > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > > _______________________________________________ > > Scikit-learn-general mailing list > > Scikit-learn-general@lists.sourceforge.net > > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > > ------------------------------------------------------------------------------ > > One dashboard for servers and applications across Physical-Virtual-Cloud > > Widest out-of-the-box monitoring support with 50+ applications > > Performance metrics, stats and reports that give you Actionable Insights > > Deep dive visibility with transaction tracing using APM Insight. > > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > > _______________________________________________ > > Scikit-learn-general mailing list > > Scikit-learn-general@lists.sourceforge.net > > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general -- Gael Varoquaux Researcher, INRIA Parietal NeuroSpin/CEA Saclay , Bat 145, 91191 Gif-sur-Yvette France Phone: ++ 33-1-69-08-79-68 http://gael-varoquaux.info http://twitter.com/GaelVaroquaux ------------------------------------------------------------------------------ One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y _______________________________________________ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
