Joao: 1) The error message you get when setting nu=0 is due to the fact that no support vectors can be found with that extreme restriction, and this confuses the predict function (try svm(...., fitted = false): the model returned is empty). In fact, the C++ code interfaced by svm() clearly allows nu = 0 and nu = 1, although these aren't sensible values. I will add a check to the R code and drop Chih-Chen Lin, the author of the C code, a message -- thanks for pointing this out.
2) The libsvm code is not optimized for polynomial kernels and is known to perform quite badly in that case (in contrast to the RBF kernel for which it is very fast). Do you think you need the whole data set for tuning the parameters? Best, David ______________________________________________ R-help@stat.math.ethz.ch mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
