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

DL4J Keras Import: scaled identity weight init #8395

Closed
AlexDBlack opened this issue Nov 15, 2019 · 0 comments · Fixed by KonduitAI/deeplearning4j#51
Closed

DL4J Keras Import: scaled identity weight init #8395

AlexDBlack opened this issue Nov 15, 2019 · 0 comments · Fixed by KonduitAI/deeplearning4j#51
Assignees

Comments

@AlexDBlack
Copy link
Contributor

@AlexDBlack AlexDBlack commented Nov 15, 2019

https://github.com/eclipse/deeplearning4j/blob/master/deeplearning4j/deeplearning4j-modelimport/src/main/java/org/deeplearning4j/nn/modelimport/keras/utils/KerasInitilizationUtils.java#L143

Scaled identity weight init not supported, setting gain=1
This is logged in a whole lot of the keras import tests.

This seems like something we should be able to add very easily, now that weight initialization is a function not just an enumeration.

@AlexDBlack AlexDBlack changed the title DL4J Keras Import: DL4J Keras Import: scaled identity weight init Nov 15, 2019
@AlexDBlack AlexDBlack self-assigned this Nov 15, 2019
AlexDBlack added a commit to KonduitAI/deeplearning4j that referenced this issue Nov 15, 2019
Signed-off-by: AlexDBlack <blacka101@gmail.com>
AlexDBlack added a commit to KonduitAI/deeplearning4j that referenced this issue Nov 16, 2019
Signed-off-by: AlexDBlack <blacka101@gmail.com>
AlexDBlack added a commit to KonduitAI/deeplearning4j that referenced this issue Nov 16, 2019
* eclipse#8395 Keras import - support scaled identity weight init

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

* More Keras scaled weight init fixes

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

* eclipse#8352 Deprecate duplicate SamplingDataSetIterator class

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

* Remove /O2 optimization for faster CUDA build

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

* Tweak regression test precision for CUDA

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

* Fix edge cases for buffer creation

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

* Update MKLDNN validation tests to new helper enable/disable settings

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

* Delete debugging class

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

* MKLDNN test - add proper skip for CUDA backend

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

* Align WeightInitUtil with weight init classes

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

* Fix for SameDiff test layers weight init when using IWeightInit classes

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
1 participant
You can’t perform that action at this time.