New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Default to zero'ed initial state for all RNN #590

Merged

MikeInnes merged 1 commit into FluxML:master from oxinabox:patch-2

Feb 4, 2019

Member

oxinabox commented Feb 2, 2019 •

edited

Loading

Zero as the default initial state is AFAIK what every other neural network framework does.

It shouldn't actually have much (any?) effect, since the Recur remembers the initial state anyway,
but it makes things match expectations.

I also see no reason that the biases were zeroed.
I kinda suspect there may just have been a type on the order of arguments there.

Maybe I am wrong and just don't understand the code


          Default to zero'ed initial state

26550da

Member

MikeInnes commented Feb 4, 2019

Seems reasonable to me, thanks.

MikeInnes merged commit e774053 into FluxML:master

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet