X0 undefined in backward pass
Closed this issue · 3 comments
nestordemeure commented
X0 seem to be used but undefined in the backward pass of the current implementation.
sbarratt commented
Good point. fixed.
…On Wed, Mar 18, 2020 at 10:34 AM Nestor Demeure ***@***.***> wrote:
X0 seem to be used but undefined in the backward pass of the current
implementation
<https://github.com/sbarratt/torch_cg/blob/13f58a9ee5d53e2cd837d1cc9ea87929cac119ee/torch_cg/cg_batch.py#L135>
.
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<#4>, or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AB7LUGJX7U6XCIXERHHHG23RIEAZXANCNFSM4LOWZN7A>
.
nestordemeure commented
Note: it might be worth adding a test with an X0.
I am not sure the backward pass works when there is a X0 as it does not return a gradient for X0 (I encountered the bug but it was with a heavily modified version of your code so it might not be relevant).
sbarratt commented
It's not meant to return a derivative with respect to X0.
…On Thu, Mar 19, 2020 at 4:37 AM Nestor Demeure ***@***.***> wrote:
Note: it might be worth adding a test with an X0.
I am not sure the backward pass works when there is a X0 as it does not
return a gradient for X0 (I encountered the bug but it was with a heavily
modified version of your code so it might not be relevant).
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
<#4 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AB7LUGLFVX6Y4MWKEPQ46GTRIH7Y3ANCNFSM4LOWZN7A>
.