Like Emanuel said, the full grid is already available. If someone really cares about their parameter selection, they should have a look and decide themselves.
I think 'best_estimator' should be chosen in the easiest way possible. I agree with David, that this should not happen in practice and so I think it is of little relevance. I disagree on returning a list, though, since that would make the usage a bit more awkward and will be of use only in some degenerate cases. I am a bit against random choice as that would make the GridSearch have a random_state, which is a bit unexpected. If there really was a case where there were multiple optimum settings on the training set - say k in KNN, then the result on the test-set would be non-deterministic. That might be a bit unexpected, if you are using a deterministic classifier like KNN. On the other hand, I really think this is of little practical relevance. And if you care so little about your results that you don't look at what your grid search gave you, then you shouldn't care about which of the 'optimum' parameters you got. just my .02 euro ------------------------------------------------------------------------------ Virtualization & Cloud Management Using Capacity Planning Cloud computing makes use of virtualization - but cloud computing also focuses on allowing computing to be delivered as a service. http://www.accelacomm.com/jaw/sfnl/114/51521223/ _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
