Per-row logit of the classifier, as a torch tensor

This is the log ratio itself: the module's single output for each (theta, x) row.

Usage

nre_logit_tensor(net, theta, x)