Use two kernel types at once

The computation of the Gram matrices requires the choice of a kernel. Two kernels are available, RBF and Delta. RBF is recommended for continuous features, Delta is recommended for categorical features. With real-world datasets, one has to select from a mix of continuous and categorical features. The current way to do it is to use RBF kernel and accept that it might not be optimal for categorical feature. Can we find a way to distinguish continuous and categorical features from a given dataset and use the post appropriate kernel for each of the two groups?

transferwise / hisel

Use two kernel types at once #34