Hi @slundberg

I'm trying to use KernelExplainer (on SHAP) and TabularExplainer (on SHAPIQ) to explain TabPFN predictions with beeswarm plot, following the documentation examples. However, the computation is taking extremely long for a very small test set (173 instances) on my personal computer, and I had to interrupt the process.

Setup:

Model: TabPFN

Test set size: 173 instances

Explainer: TabularExplainer

Questions:

Is TabPFN inherently slow for SHAP explanations compared to XGBoost/LightGBM/Random Forest?

Are there any optimization parameters or alternative approaches for explaining TabPFN more efficiently?

What would be a reasonable expected runtime for this dataset size?

System specs: Intel(R) Core(TM) Ultra 9 185H (2.30 GHz) processor and 32 GB of RAM.

Thank you!