@gael
> I think that the answer to this problem is to give to the REFCV as an
'estimator' parameter an GridSearchCV object that tunes the SVC. Would
> that make sens?
>
Yes, that makes sense, and I tried that. But any RFECV expect every
estimator to have a .coef_ attribute (it recursively eliminates features
based on their coefficients). The problem is that GridSearchCV does not
have the .coef_ -- it's true the GridSearch.best_estimator has a .coef_
attribute, but RFECV doesn't know that. Perhaps RFECV can check to see
whether it's estimator is derived from the GridSearchCV class, and if so,
instead use `estimator.best_estimatro_.coef`.
------------------------------------------------------------------------------
This SF email is sponsosred by:
Try Windows Azure free for 90 days Click Here
http://p.sf.net/sfu/sfd2d-msazure
_______________________________________________
Scikit-learn-general mailing list
Scikit-learn-general@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
