I just realized that svm.svm_model.get_sv_indices() returns an index of 1 for the first elements, 2 for the second element, etc. This makes sense for matlab, but not for python, where list indices start at 0. I noticed the issue trying to reproduce matlab code, comparing the training data for indices of model.get_sv_indices() to the corresponding output of model.get_sv_indices().
This isn't related to matlab. It's a C function in the core library. The reason why we decide to return [1... num_data] is related to the setting of precomputed kernels.
I just realized that svm.svm_model.get_sv_indices() returns an index of 1 for the first elements, 2 for the second element, etc. This makes sense for matlab, but not for python, where list indices start at 0. I noticed the issue trying to reproduce matlab code, comparing the training data for indices of model.get_sv_indices() to the corresponding output of model.get_sv_indices().