Hi, I am going through the ML Observability Fundamentals course and noticed something I think might be strange behavior when comparing feature importance between two copies of the exact same dataset for the same time period. When looking at global comparison they are equivalent but when I drill down to filter by cohort, the SHAP scores are different. It seems to me like they should be the same, am I misunderstanding something or could this be a bug?
checked on two of the demo models
arize-demo-fraud-detection-use-case
arize-demo-click-through-rate-use-case
