Trouble Training BiTrapGMM on JAAD and PIE

[werdnabae]

Hello authors,
I'm trying to train the BiTrapGMM on JAAD and PIE, but I'm running into some troubles while doing so.
This is the error message I'm getting.

Changing line 32 in bitrap_gmm.py to:
self.node_future_encoder_h = nn.Linear(self.cfg.GLOBAL_INPUT_DIM, 32)
solves this issue, but raises a new issue later in the forward method of bitrap_gmm.py

Any insight into these issues would be appreciated!

Thank you,
Andrew