Hey,
just wanted to let you know that the current NAMLSS implementation is not working. First off al, the dimensions do not add up, see
|
x = self.feature_dropout(torch.cat(all_outputs, dim=1)).sum(dim=1).unsqueeze(-1) |
Changing this to x = self.feature_dropout(torch.stack(all_outputs, dim=-1)).sum(dim=-1) at least trains the model. However, I get other errors related to saving the parameters and deprecated functions.
Hey,
just wanted to let you know that the current NAMLSS implementation is not working. First off al, the dimensions do not add up, see
NAMpy/nampy/basemodels/nam.py
Line 297 in 7fbb144
Changing this to
x = self.feature_dropout(torch.stack(all_outputs, dim=-1)).sum(dim=-1)at least trains the model. However, I get other errors related to saving the parameters and deprecated functions.