facebookresearch / modeling_long_term_future Goto Github PK

Could you breakdown the final loss function used in the code, comparing with eq 7 in the paper?
I can clearly see the log prob gaussian for forward net, the KL div term, the aux reconstruction term. The z-forcing paper also had the log prob gaussian for backward net in its loss formulation, but I dont understand it in eq7. Also, the action decoder loss is missing.

Hi,
Im trying to run your experiments but can't figure out where to find the code corresponding to the following (in scripts/rl_zforcing.py and scripts/rl_zforcing_dec.py)

from torch.nn._functions.thnn import rnnFusedPointwise as fusedBackend

used later on in the definition of class LSTMCell. However there is No module named 'torch.nn._functions' anymore in PyTorch it seems. If I understand correctly this code was available up until PyTorch version 0.4.1 but was discontinued after that ? So do you know where I should now point towards to use the same function ?

By searching I found a similar function in this old repo that this user probably forked from pytorch a while back. Is that the intended function to be used ?

Thanks