Allow selecting proper training sample for endogenous_prior
Currently, we simply use
Y=data';
, but it is straightforward to include different data