Best practices for monitoring ML models
This section discusses best practices for monitoring models using SageMaker Model Monitor and SageMaker Clarify, taking into consideration the under-the-hood operation of these features and a few limitations as they stand at the time of publication of this book:
- Choosing the correct data format: Model Monitor and Clarify can only monitor for drift in tabular data. Therefore, ensure that your training data is in tabular format. For other data formats, you will have to build custom monitoring containers.
- Choosing real-time endpoints as the mode of model deployment: Model Monitor and Clarify support monitoring for a single-model real-time endpoint. Monitoring a model used with batch transform or multi-model endpoints is not supported. So, ensure that the model you want to monitor is deployed as a single-model real-time endpoint. Additionally, if the model is part of an inference pipeline, the entire pipeline is monitored, not the individual...
