Am 30.11.2012 14:05, schrieb Gael Varoquaux: > On Fri, Nov 30, 2012 at 01:58:48PM +0100, Philipp Singer wrote: >>> Don't. This is not a good model selection strategy, and it is very >>> costly. Use a stratified kfold with k between 5 or 10. >> Well, I only have a few samples and I am explicitely interested in >> seeing the performance of each sample alone trained on the rest. > Still. I keep having this discussion with users: leave one out is about > the worst strategy possible because it is the one with the largest > variance on the test accuracy estimation. You'll have to believe me on > this, because I don't have time to dig up the relevent papers (there is a > bit of literature on this, but it's really hard to find, and it's more > folk knowledge in machine learning. Gael did dig up the citations and posted them here:
https://github.com/scikit-learn/scikit-learn/issues/1427 ------------------------------------------------------------------------------ Keep yourself connected to Go Parallel: TUNE You got it built. Now make it sing. Tune shows you how. http://goparallel.sourceforge.net _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
