What should the ML specialist do to resolve the violations?
Manually trigger the monitoring job to re-evaluate the SageMaker endpoint traffic sample.
Run the Model Monitor baseline job again on the new training set. Configure Model Monitor to use the new baseline.
Delete the endpoint and recreate it with the original configuration.
Retrain the model again by using a combination of the original training set and the new training set.
Explanations:
Manually triggering the monitoring job does not address the root cause of the violations. The monitoring job needs to be based on an updated baseline that reflects the current model’s performance and the latest dataset. Simply re-evaluating the existing traffic sample may still result in violations if the baseline is outdated.
Running the Model Monitor baseline job again on the new training set is essential. This establishes a new baseline that reflects the updated model’s performance with the latest dataset, allowing the monitoring to accurately assess if the new model is meeting expectations based on current data. Configuring Model Monitor to use this new baseline is crucial to resolving the violations.
Deleting and recreating the endpoint with the original configuration does not solve the issue. The endpoint may still be using the old baseline data for monitoring, which does not align with the performance of the newly trained model. This approach is unnecessary and inefficient.
Retraining the model using a combination of the original and new training sets might not address the monitoring violations directly. While it could improve the model, if the baseline for monitoring is not updated to reflect this retraining, violations may persist. The monitoring violations need to be addressed through an updated baseline specifically tailored to the new data.
In my opinion, the answer is:
Run the Model Monitor baseline job again on the new training set. Configure Model Monitor to use the new baseline.