SageMaker Timeseries Explainability: Timewise bug

[Alex-Wenner-FHR]

Describe the bug When using the TSX feature, there appears to be a bug in the timewise granularity.

I see the following error not specifying static covariate baselines in the AsymmetricShapleyValueConfig instantiation: ValueError: Baseline for static_covariates must be set.

When I do specify static covariate baselines in the AsymmetricShapleyValueConfig instantiation, I see the following: ValueError: Baseline for static_covariates is only supported for fine_grained. Used granularity: timewise.

To reproduce Try to use the AsymmetricShapleyValueConfig instantiation with and without baselines for the static covariates.

Expected behavior I would expect either specifying or not specifying the static covariate baselines to function with timewise granularity.

System information A description of your system. Please provide:

• SageMaker Python SDK version: 2.217.0
• Framework name (eg. PyTorch) or algorithm (eg. KMeans): N/A (TSX)
• Framework version: N/A (TSX)
• Python version: 3.9 on the Processing Job, 3.10 on the notebook launching the job.
• CPU or GPU: GPU & CPU
• Custom Docker image (Y/N): N

[rvasahu-amazon]

Hi Alex,

Thanks for reaching out. This looks like an bug on our end. We're working on a fix and release for this. For now please use fine_grained explanations instead. We'll include an update here when a fix is released.

[rvasahu-amazon]

Hi, this issue has been resolved and a new release has since been made. It should now be possible to omit a static covariates baseline when selecting a timewise explanation granularity. Please reach out to the team if there are any other issues.