Uniform distribution instead of Gaussian distribution

[jan1854]

Hi,

I think there might be a slight error in the LatentModel. For sampling the latent states, the function torch.rand_like() is used, which samples from a uniform distribution. According to the paper (see appendix B) and the original implementation the latent states should be sampled from a Gaussian distribution, however. This can be fixed by replacing torch.rand_like() with torch.randn_like() In lines 193, 196, 204, 207, 226, 229, 239, 242 of network/latent.py.

Best regards,
Jan

[toshikwa]

You're right, they are typos...
Could you send a pull request?

Thank you for pointing out.

[jan1854]

There you go: #6