Jopyth opened a new pull request #17566: [WIP] Discussion on merging BMXNet 2 contributions URL: https://github.com/apache/incubator-mxnet/pull/17566 ## Description ## This PR can help us **start a discussion** on how to integrate the functions required for Binary Neural Networks (BNN) models into mxnet based on our code in BMXNet 2 ( https://github.com/hpi-xnor/BMXNet-v2 ). It should definitely not be merged in its current state. ## Checklist ## ### Essentials ### Please feel free to remove inapplicable items for your PR. - [ ] The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant [JIRA issue](https://issues.apache.org/jira/projects/MXNET/issues) created (except PRs with tiny changes) - [ ] Changes are complete (i.e. I finished coding on this PR) - [ ] All changes have test coverage: - Unit tests are added for small changes to verify correctness (e.g. adding a new operator) - Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore) - Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL) - [ ] Code is well-documented: - For user-facing API changes, API doc string has been updated. - For new C++ functions in header files, their functionalities and arguments are documented. - For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable - Check the API doc at https://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html - [ ] To the best of my knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change ### Changes ### We added three functions `det_sign` ([ada4ea1d](https://github.com/hpi-xnor/BMXNet-v2/commit/ada4ea1d4418cfdd6cbc6d0159e1a716cb01cd85)), `round_ste` ([044f81f0](https://github.com/hpi-xnor/BMXNet-v2/commit/044f81f028887b9842070df28b28de394bd07516)) and `contrib.gradcancel` to MXNet (see [src/operator/contrib/gradient_cancel[-inl.h|.cc|.cu]](src/operator/contrib)). - _The `gradcancel` operator is needed for training a BNN only, however it currently is also part of the binary layers._ The rest of the changes was made in the following folders/files: - Examples are in a submodule in [example/bmxnet-examples](https://github.com/hpi-xnor/BMXNet-v2-examples) - _might reside in a binary model zoo instead_ - Tests are in [tests/binary](tests/binary) - _we run our tests with `pytest`, so a conversion (and possibly integration into the existing test files) of these to `nosetests` should be made_ - Layers are in [python/mxnet/gluon/nn/binary_layers.py](python/mxnet/gluon/nn/binary_layers.py) - _needed for inference and training_ - Converter is in [tools/binary_converter](tools/binary_converter) - _only needed for optimized inference_ See our [Changelog](https://github.com/hpi-xnor/BMXNet-v2/blob/master/CHANGELOG.md) for a (hopefully) complete overview. ## Comments ## - Since we merged different mxnet releases one after another into our repository, a squash or interactive rebase could help clean the changes up.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
