> But you're not breaking independence here: you're drawing iid from a > finite population. However, as pointed by Olivier, this may create some > artefact when used with some classifiers. > (the scores across folds are not independent but this is true for most > cv techniques, and this is another matter anyway).
if you look at my tests on scale_C you'll see that I test that if you duplicate each sample and C is fixed, you don't change the solution. It seemed at this time a very valid concern and a good argument for having the scale_C. I do believe now that it's not a valid argument. When you bootstrap with replacement then you're in the "middle“ as you have duplicated samples. does it make sense? Alex ------------------------------------------------------------------------------ Live Security Virtual Conference Exclusive live event will cover all the ways today's security and threat landscape has changed and how IT managers can respond. Discussions will include endpoint security, mobile security and the latest in malware threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
