How to Contribute

We'd love to accept your patches and contributions to this project. There are just a few small guidelines you need to follow.

Contributor License Agreement

Contributions to this project must be accompanied by a Contributor License Agreement. You (or your employer) retain the copyright to your contribution, this simply gives us permission to use and redistribute your contributions as part of the project. Head over to https://cla.developers.google.com/ to see your current agreements on file or to sign a new one.

You generally only need to submit a CLA once, so if you've already submitted one (even if it was for a different project), you probably don't need to do it again.

Code reviews

All submissions, including submissions by project members, require review. We use GitHub pull requests for this purpose. Consult GitHub Help for more information on using pull requests.

Continuous Integration

Continuous Integration tests run for every PR that is sent against TensorFlow.js repositories. Any time you push to the branch, they will re-run. Before asking for a review, make sure that the continuous integration tests are passing.

To see the logs from the Cloud Build CI, please join either our discussion or announcement mailing list.

Failing tests (click details for information):

Passing tests:

Filing GitHub issues

Please follow the guidelines specified in the ISSUE_TEMPLATE file.

Good places to start

Consider joining the tfjs mailing list to discuss the development of TensorFlow.js. If you're working on anything substantial it's better to discuss your design before submitting a PR. If you're looking for tasks to work on, we also mark issues as "good first issue" and "stat:contributions welcome"

Adding functionality

One way to ensure that your PR will be accepted is to add functionality that has been requested in Github issues. If there is something you think is important and we're missing it but does not show up in Github issues, it would be good to file an issue there first so we can have the discussion before sending us a PR.

In general, we're trying to add functionality when driven by use-cases instead of adding functionality for the sake of parity with TensorFlow python. This will help us keep the bundle size smaller and have less to maintain especially as we add new backends.

Adding an op

When adding ops to the library and deciding whether to write a kernel implementation in backend.ts, be sure to check out the TensorFlow ops list here. This list shows the kernels available for the TensorFlow C API. To ensure that we can bind to this with node.js, we should ensure that our backend.ts interface matches ops in the TensorFlow C API.

Code reviews

All submissions, including submissions by project members, require review. We use GitHub pull requests for this purpose. Consult GitHub Help for more information on using pull requests.

We require unit tests for most code, instructions for running our unit test suites are in the documentation.