Ability to Collapse Sub-Runs with Hierarchical Naming Convention

Summary

As a user of TensorBoard, I find it challenging to navigate and analyze data when working with many runs or custom logs. I would like TensorBoard to have a built-in feature that allows me to collapse sub-runs with a hierarchical naming convention. This would enable me to better organize and visualize my runs and significantly improve my user experience.

Background

TensorBoard currently lacks the ability to collapse sub-runs in this way. Although the "Regex Filter" feature can group runs based on a regular expression, it is not as intuitive or convenient as the proposed feature. By collapsing sub-runs with hierarchical names, I would be able to easily see the relationship between different runs and better understand how they fit into my overall analysis.

Proposed Solution

I propose that TensorBoard add a feature to collapse sub-runs with the naming convention of "parent_run/child_run/...". Ideally, the collapsed sub-runs would appear as a single entry with a toggle button to expand or collapse the view. This feature would make it easier for users to visualize and organize their runs and improve the overall usability of TensorBoard.

Current solution:

Proposed solution:

Conclusion

As a researcher who frequently works with deep learning models, I rely on TensorBoard to monitor the performance of my experiments. By adding the ability to collapse sub-runs with a hierarchical naming convention, TensorBoard would become a more efficient tool for organizing and analyzing data. I hope this feature request can be considered for implementation in a future release. Thank you.

Start by reviewing TensorBoard's existing Regex Filter and the run visualization shown in the issue. Define how slash-separated parent_run/child_run names should appear, how expansion and collapse should work, and how the behavior should be validated against the proposed UI.