Dimensionally-stacked heatmaps

[EwoutH]

I recently implemented a custom extension to Seaborn's heatmap function to visualize hierarchical data with multi-level indices. This approach reshapes and formats hierarchical data into a dimensionally-stacked heatmap, enabling clear representation of multi-dimensional relationships (e.g., parameter sweeps or policy comparisons).

Here’s an example visualization:

Note the custom formatting of axis ticks to display hierarchical levels and dividing lines to visually separate higher-level categories (these could be thicker or a color though).

This type of plot is particularly useful for multi-dimensional data analysis, like parameter sweeps or experimental results. Practically anything you want to visualize with more than two dimensions. It was originally inspired from the EMA workbench.

@mwaskom Would you find such functionality to be useful for Seaborn, or should it remain an external utility built on top of it? If it’s a potential fit, I’d be happy to refine the design and open an PR.

Thanks for maintaining such an excellent library!

[simon-hirsch]

Super cool, would you like to share some of the if this does not make it into seaborn?

[EwoutH]

You can find some here:

• Code: https://github.com/EwoutH/urban-self-driving-effects/blob/main/model/analysis_ff.ipynb
• Images: https://github.com/EwoutH/urban-self-driving-effects/tree/main/img/exp4

[simon-hirsch]

Thanks! :)