kl-divergence formula in vae seems wrong about model-zoo HOT 6 CLOSED

@chengchingwen you're totally correct. The last term in my implementation is completely off. Not sure how this slipped through, but many thanks for finding it. @AbinavRavi I propose that you accept the proposed fix.

from model-zoo.

@MikeInnes

from model-zoo.

Can I take up this issue?

from model-zoo.

@AbinavRavi sure? I guess it should be easy to fix since it just looks like a typo. However, I'm not the guy who knows divergence very much. Most of the formula were found on the Internet, so maybe we need a double check on the equations.

from model-zoo.

Best to talk to @willtebbutt who wrote this.

from model-zoo.

@willtebbutt I was learning about vae and just stuck on that function for a few days. The interesting part is that even there are some weird term in the code, the model still works well. I guess that's the reason why nobody notice it before. @AbinavRavi you might also make a pr ?

from model-zoo.