Skip to content
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.

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

libnd4j: biasadd_bp doesn't have isNCHW arg, wrong for NHWC case? #8280

Closed
AlexDBlack opened this issue Oct 9, 2019 · 1 comment · Fixed by KonduitAI/deeplearning4j#28
Closed
Labels
Bug
Milestone

Comments

@AlexDBlack
Copy link
Contributor

@AlexDBlack AlexDBlack commented Oct 9, 2019

https://github.com/eclipse/deeplearning4j/blob/master/libnd4j/include/ops/declarable/generic/parity_ops/bias_add.cpp#L83-L86

The bias gradient (via sum) is basically epsilonNext.sum(0, 2, 3) which is correct for the NCHW case but not the NHWC case (which should be .sum(0, 1, 2))

@AlexDBlack AlexDBlack added Bug C++ labels Oct 9, 2019
@AlexDBlack AlexDBlack added this to the 1.0.0-beta6 milestone Oct 31, 2019
@shyrma

This comment has been minimized.

Copy link
Contributor

@shyrma shyrma commented Nov 3, 2019

corrected!

AlexDBlack added a commit to KonduitAI/deeplearning4j that referenced this issue Nov 4, 2019
Signed-off-by: AlexDBlack <blacka101@gmail.com>
@AlexDBlack AlexDBlack mentioned this issue Nov 4, 2019
AlexDBlack added a commit to KonduitAI/deeplearning4j that referenced this issue Nov 4, 2019
* eclipse#8280 biasadd_bp nchw arg fixes (java side) + test

Signed-off-by: AlexDBlack <blacka101@gmail.com>

* eclipse#8285 Concat op Java side fixes

Signed-off-by: AlexDBlack <blacka101@gmail.com>

* Concat op cpp fix - allow dynamic axis to be negative, same as static axis

Signed-off-by: AlexDBlack <blacka101@gmail.com>

* ignores for deconv3d import tests until deconv3d_tf op is implemented

Signed-off-by: AlexDBlack <blacka101@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
2 participants
You can’t perform that action at this time.