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 Fixes #5942

merged 5 commits into from Jul 23, 2018


Copy link

commented Jul 21, 2018

Fixes: #5907
Fixes: #5624

Note: This PR makes significant changes to mask arrays for CNN layers.

  • Previously: masks were always 2d, [minibatch,H] or [minibatch,W]. And global pooling layer could work only if exactly one of H or W of input was 1
  • Now: Masks for CNNs are 4d, in a broadcastable format: [mb,d/1, h/1, w/1]
    • Required to fix #5624
    • This also allows us to support more sophisticated masking situations: for example, different image sizes in same minibatch for fully convolutional networks (using either Pool+outputLayer, or CnnLossLayer)
  • CnnSentenceDataSetIterator now supports Cnn1d/RNN format also (plus has been updated to new 2d CNN mask format)

@AlexDBlack AlexDBlack changed the title [WIP] DL4J Fixes DL4J Fixes Jul 23, 2018

@AlexDBlack AlexDBlack merged commit 2f76896 into master Jul 23, 2018

0 of 2 checks passed

codeclimate Code Climate encountered an error attempting to analyze this pull request.
continuous-integration/jenkins/pr-merge This commit cannot be built

@AlexDBlack AlexDBlack deleted the ab_dl4j branch Jul 23, 2018

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
None yet
1 participant
You can’t perform that action at this time.