2012/8/22 Andreas Mueller <[email protected]>: > Hi everybody. > Recently I tried to hack at the SVM implementation, motivated by Lars' issue > #918. > While trying to figure out how to add support for callable kernels, I found > out > that precomputed kernels are not supported at all in the sparse svm, even > though they are documented and look like they are tested. > > Trying to implement sparse kernels, I ran in a dead end somewhere in the C > code > (see my PR). > > Do you think we should try to fix that? Or should we warn / disable this > "feature" > which gives arbitrary results and randomly segfaults? > > The use-case that Lars was interested in doesn't actually need the sparse > SVM, > and I don't really see many applications for having a sparse matrix as a > kernel. > I guess we should rather change the checks such that callable kernels can > take > sparse input but have to produce dense kernel matrices. > > What do you think?
+1 for raising an exception for when the precomputed kernel matrix is spares + documenting it in the docstring. +1 for opening a github issue / PR for adding support for sparse precomputed kernel matrix later in the unlikely event of a user encountering that use case "in real life". -- Olivier http://twitter.com/ogrisel - http://github.com/ogrisel ------------------------------------------------------------------------------ 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
