autoencoders icon indicating copy to clipboard operation
autoencoders copied to clipboard

Published 20 hours ago • verified publisher icon caglar

KL Divergence

Open mdda opened this issue 9 years ago • 2 comments

When I look at your function at ae.py#L138, and compare it to https://stat.duke.edu/courses/Spring09/sta205/lec/hoef.pdf (in the middle of page 2) it seems like there should be different bracketing :

return p * T.log(p) - T.log(p_hat) + (1 - p) * T.log(1 - p) - (1 - p) * T.log(1 - p_hat)

should be

return p * (T.log(p) - T.log(p_hat)) + (1 - p) * (T.log(1 - p) - T.log(1 - p_hat))

Please be gentle if I've missed something...

mdda avatar Nov 11 '15 10:11 mdda

Yes, I agree with you. Be honest, I found that the files has some bugs, more than only one. Maybe these are caglar's challenge/test to us, haha!

veinpy avatar Jan 10 '16 16:01 veinpy

Hi, I'm also running this code. Can you show me the format of your test dataset mnist_all.pickle? Or where can I find it?

Great thanks!

liweijia avatar Jun 01 '16 09:06 liweijia