----- Original message ----- > But then why classifiers only? And what stops of from doing this for > every kind of model? labels_ is part of the clustering API, but we > don't expose that through a Pipeline.
And coefs_ for linear models... > (Should we expose all fitted parameters of the final step of the > pipeline?) No, because on some of these we need to do an inverse_transform. In general I am against trying to do this for the user: we won't be able to do this well, and thus we'll spend our time closing issues on this. The question is: can we come up for an API in the pipeline that simplifies this problem? For instance a list of attributes that must transfered and a second list that must be inverse_transformed. For now, I am not convinced by my own suggestion above. For the OP's problem the simplest solution is to rely on the estimator attribute of the pipeline. Maybe we should try to make this attribute a bit more general, if there are some similar usecases with other meta-estimator, and rename it to fitted_estimator_, to collide with the best_estimator_ of GridSearch. Anyhow, as you see I am trying to find a bigger picture... (And I should listen to the talk rather than write emails) G ------------------------------------------------------------------------------ Everyone hates slow websites. So do we. Make your web apps faster with AppDynamics Download AppDynamics Lite for free today: http://p.sf.net/sfu/appdyn_d2d_feb _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
