Additions / Bug fixes V 1.4

  • svm, svr : new svmtorch optimizer, libsvm should now work for regression properly, too.

  • gproc : new Gaussian process object

  • relvm_r: new Relevance Vector Machine object

  • mutinf: new mutual information feat. sel object

  • kmp: new kernel matching pursuit object

  • kpca object changed:
    o - In the old object, d.X had to be a kernelmatrix and the child field was emtpy. Now the child field (kernel) is kernel('linear') by default and d is assumed to be data. If you want to simulate the old behaviour you can use custom kernel.
    o - The eigenvecs of the kernelmatrix are sorted according to the absoulte value of the eigenvals in decreasing order.
    o - There is also some code added to treat degenerated subspaces correctly.