running likelihood.eval() works, and it gives the (N x n_predictors) shape I’m after.
Strange when its within the model environment it doesn’t work.
I’ll have a look at the link for test for batch params sometime tomorrow and provide an update on how it goes.