Allow selecting proper training sample for endogenous_prior

Currently, we simply use Y=data';, but it is straightforward to include different data