Can you try running it under valgrind and sending back the memory errors you get? I have no idea how hoisy valgrind's output on the scikit is, but this should help narrow things down.
On Thu, Nov 3, 2011 at 12:17, Andreas Müller <[email protected]> wrote: > Hi folks. > Today I ran across a segfault doing sgd multi class classification. > I tracked the error down to sgd_fast but don't know how to proceed. > My data has shape (50000, 13824) and is dense, the parameters > of the classifier are: > SGDClassifier(loss="hinge", penalty="l2") > > I ran it through scaler and the features seem to be in a reasonable > range and there are no infs or nans. > If I train only on the first 5000 of the 13824 features, I get > reasonable results, for 6000 it segfaults. The features split > into three parts of equal lenght and training on any of > these parts works fine. > > My RAM is big and empty during the crash. > > Does anyone have any idea how to find the problem? > > Cheers, > Andy > > ------------------------------------------------------------------------------ > RSA(R) Conference 2012 > Save $700 by Nov 18 > Register now > http://p.sf.net/sfu/rsa-sfdev2dev1 > _______________________________________________ > Scikit-learn-general mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > -- - Alexandre ------------------------------------------------------------------------------ RSA(R) Conference 2012 Save $700 by Nov 18 Register now http://p.sf.net/sfu/rsa-sfdev2dev1 _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
