Beeswarm plots are useful in visualizing dependencies between feature values and feature importance. A beeswarm plot could be adapted from the SHAP library by sub-dividing the y-axis for each interaction term:
- An interaction term (e.g., "Feature A x Feature B") will occupy a "block" on the y-axis.
- This block will be subdivided into multiple rows: one for each feature within the interaction.
- For the interaction (A, B):
- There will be a sub-row for Feature A. The x-position of every dot in this row will be the interaction value for that sample, but the color of the dot will be determined by the value of Feature A.
- There will be another sub-row for Feature B. The x-position of the dots is the same, but the color will be determined by the value of Feature B
Example of such plot could look like this:
Beeswarm plots are useful in visualizing dependencies between feature values and feature importance. A beeswarm plot could be adapted from the SHAP library by sub-dividing the y-axis for each interaction term:
Example of such plot could look like this: