Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.Sign up
Cross entropy with logit #1104
We found that the layer works better than
SigmoidLayer + CrossEntropyError sometimes. But later we realized that we used incorrect parameters and these two approaches give equal results on correct parameters. So, I am not sure that the layer is needed except some special cases.
P.S. Sorry for that, that was my mistake.
I think valgrind could help.
Looks good to me, just made some minor issues abut the style.
Looks good to me, just a few comments about optimization and comments. I think maybe I found a bug in
Backward() too, please check on that. Otherwise I think it's ready.
I am just a bit busy right now. I will look at it on the weekend if that is okay.…
On Thu, Sep 14, 2017 at 8:28 PM, Ryan Curtin ***@***.***> wrote: @kris-singh <https://github.com/kris-singh>: did you want to handle the issues I pointed out? Once those are done I think that everyone agrees this is ready for merge ***@***.*** <https://github.com/zoq> or @lozhnikov <https://github.com/lozhnikov> correct me if I am wrong there please). — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1104 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AKvBS-PTsR_Gs8LrfJzpeqO0V-OWtF2Tks5siT74gaJpZM4PDp6L> .
@lozhnikov: I think that the issue wasn't pointed out by the tests because the Forward() function is only being tested with positive inputs: