Known issues and limitations for Watson OpenScale

The following known issues and limitations apply to the IBM Watson OpenScale service.

Known issues

Limitations

Known issues for Watson OpenScale

Some explainability configurations that use SHAP cannot be saved

If you train a model on data from an S3 bucket and you use the watsonx.data Presto connector, you cannot save your explainability configuration if you use Shapley Additive explanations (SHAP).

To enable SHAP explanations, ensure that the model deployed in the deployment space is trained using the native Db2 connector, linked to a Db2 datasource within the Watson Studio project. During the online configuration in Watson OpenScale, select Db2 as the datasource, allowing Watson OpenScale to access the training data needed for SHAP-based explainability.

Note: Watson OpenScale currently does not support a native Presto connector, so the connector used to train the AutoAI model must be consistent with the connector specified during the Watson OpenScale model configuration step.

Attempt to import settings from pre-prod deployment space fails

If you are assigned the admin or editor role for a deployment space, your attempt to import settings from a pre-prod deployment space might fail.

You can work around this issue by being assigned the admin or editor role for the associated Watson OpenScale instance. You can also login as the cpadmin user to import settings.

Model details fail to display when viewing threshold violation alert notifications

If you are invited to a deployment space as an admin user and send metric threshold violation results to your email address, your attempt to view model details might fail. When you click View model details in your email notification, a WOS instance opens and fails to display model details.

To work around this issue, you must be assigned a collaborator role for the Watson OpenScale service instance that is associated with your deployment space to view model details.

Limitations for Watson OpenScale

Special characters are not supported

Special characters, such as é, are not supported. Errors can occur, for example, when you use special characters in the names of evaluations, models, or metrics. Replace special character with standard ASCII characters.