About calculating logdet

[ydzhang-stormstout]

Hi, thanks for your paper, and I really enjoy it.
I found the code about calculating the logdet is different from that in the paper.
Specifically, the torch.log(radius) might be redundant in https://github.com/oskopek/mvae/blob/master/mt/mvae/ops/hyperbolics.py#L63,
since log(R) is contained by torch.log(r).
Also, the logdet in sphere might have the same problem.
Could you please help me check the codes? Thanks a lot!

[oskopek]

Hm.. now that you mention it, it seems you're right. Did you try running training with the -log(r) removed from code?

Apologies for the late reply.

[ydzhang-stormstout]

Hm.. now that you mention it, it seems you're right. Did you try running training with the -log(r) removed from code?

Apologies for the late reply.

I just checked the implementation of the wrapped normal distribution and didn't conduct the experiments. Thanks for your help.

[HT-Everlasting]

I've encountered an issue while training models with the CIFAR and Omniglot datasets. When using the MNIST dataset, everything works fine. However, with CIFAR and Omniglot, I get the following error:AttributeError: Can't pickle local object 'ToDefaultTensor.init..' Thanks a lot!