Add pytorch demo script

[wanzysky]

500 iters:

1500 iters:

[oiao]

Thanks a lot for the implementation. I have noticed a strange behavior of Pytorch: When concatenating the return value of LinearNormalGamma at https://github.com/aamini/evidential-deep-learning/blob/71ebd59ab3a4b66c38d919e8aa9ad3711a416796/torch_hello_world.py#L22 such that:

ret = mu, self.evidence(logv), self.evidence(logalpha) + 1, self.evidence(logbeta)
return torch.cat(ret, dim=-1)

the training does not seem to progress and a constant value is predicted. Is there any reasonable explanation for this?