In this example, we render a run selector dropdown component. When the user selects a run, it shows a preview of all scalar data for tags within it. For a complete guide to plugin development, see ADDING_A_PLUGIN.
All files under static/* are served as static assets, with the frontend entry point being static/index.js. The plugin backend serves scalar summaries (e.g. values written by tf.summary.scalar) from runs within the --logdir passed to TensorBoard.
To generate some scalar summaries, you can run the demo.py. Alternatively, you can write scalars from your own Python program to a log directory, using Keras callbacks or tf.summary.scalar with a summary file writer.
Copy the directory tensorboard/examples/plugins/example_raw_scalars into a desired folder. In a virtualenv with TensorBoard installed, run:
python setup.py develop
This will link the plugin into your virtualenv. Then, just run
tensorboard --logdir /tmp/runs_containing_scalars
and open TensorBoard to see the raw scalars example tab.
After making changes to static/index.js or adding assets to static/, you can refresh the page in your browser to see your changes. Modifying the backend requires restarting the TensorBoard process.
To uninstall, you can run
python setup.py develop --uninstall
to unlink the plugin from your virtualenv, after which you can also delete the tensorboard_plugin_example_raw_scalars.egg-info/ directory that the original setup.py invocation created.