We are facing issue with shap implementation using KernelExplainer.
Train data : 260mb
Test data : 50mb
GPU memory : 20GB
During implementation, we are not able to see any spike in memory utilization.
Let us know in case any further information is required.
We have gone through the documentation of shap kernel explainer here : API Reference — cuml 23.10.00 documentation
The implementation shown here also having the same issues once we increase the sample size to more than power of 5. The gpu utilization doesn’t show anything unusual but consistently throwing cuda memory error. It’s been more than a month, awaiting for your replies.