Step Log visualization optimization

[clemens-fricke]

While using the step log visualization tool more and more a few issue came to light.

1. When hiding a trace, by clicking on it in the legend, it would reappear after changing the episode
2. Colors in the top plot repeated for reward and the first objective
3. Only one legend existed and not separate legends for each subplot
4. File sizes get really big if the step log is big
5. Responsiveness of the visualization was sluggish for large files

To combat this the visualization was optimized. The biggest part is the fix for 1, 4, and 5 which were fixed by changing how the slider works. Before all episode traces were plotted separately at the same time but than hidden for all episode but the one that was shown. Now one one trace is plotted for each variable and the data is exchanged for each episode. This solves both the responsiveness, file size and trace hiding problems.

Quantifiable is the easiest for point 4: For a specifically large step log we get a visualization file that is 374Mb in size and for the optimized the file size shrinks to 78Mb.

The new visualization looks like this:

[Copilot]

Pull request overview

This PR optimizes the step log visualization tool to address performance and usability issues. The key improvement is replacing the approach of pre-plotting all episodes (and hiding unselected ones) with dynamically updating trace data via the slider control. This change significantly reduces file sizes (e.g., from 374MB to 78MB in large cases) and improves responsiveness while fixing trace hiding bugs and color repetition issues.

Key changes:

• Refactored slider implementation to update trace data dynamically instead of toggling visibility of pre-rendered traces
• Added separate legends for each subplot instead of a single shared legend
• Improved memory management with explicit cleanup of intermediate dataframes

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[danielwolff1]

Looks good to me so far, thanks for the very valuable improvements.

I would say let's have a brief discussion about this next Tuesday and merge it after that.