[RFC] TensorBoard - Improved Plugin Extensibility

[Edd Wilder-James]

Hi folks,

I'm very excited to announce the first RFC from the TensorBoard project is available for comment - Improved Plugin Extensibility.

The feedback phase will be open until 2019-04-29. Comments are invited to the pull request linked. You can view the design doc there, and also leave comments inline on the document source.

Summary

This RFC proposes a friction-free approach to enabling TensorBoard contributors to create and share TensorBoard plugins as installable extensions.

TensorBoard is composed of various plugins that show up as separate dashboards to users (e.g., scalars, graphs, hparams, histograms, etc.). Today. anyone can contribute a new plugin by forking TensorBoard, adding their plugin, and getting the PR merged. However, this can require a lengthy review to make sure it is architected correctly and is aligned with TensorBoard’s goals.

This RFC covers two changes that will make this process easier: Plugins can be distributed as Python packages. Users can install them as extensions to TensorBoard. Plugins are no longer limited to use Polymer but can instead use any frontend library or framework (e.g., React)

The intention is to enable any prospective contributor to easily be able to create or reuse a visualization with tools familiar to them, and enable users to try it out quickly. Plugins that are useful to many users should still be added to TensorBoard, but this removes the friction needed to start getting feedback or trying out something new.

thanks in advance for your comments!

--

Edd Wilder-James, Open Source Strategy at Google

Submit a proposal for TensorFlow World!

Github @ewilderj • @tensorflow • @kubeflow