Implementations as in SparseLib 1.0
** See : http://research.microsoft.com/mlp/RVM/relevance.htm
**
** Comments from -gb-:
**** Note : So far requires centered data !!
** Note that if you get an error due to "Bad conditioned Hessian"
its an inherent problem in the implementation. Mostly this implementation
works with rbf and rarely with a linear kernel.
Furthermore if you try to solve a multivariate regression problem the
code automatically recognises and you get a multi_reg(relvm_r) object
back!
** Example:
** d=gen(toyreg({'o=2','n=3'}))
[r,a]=train(relvm_r(kernel('rbf',1)),d)