Use it when
- You want to monitor your deployed model to detect data drift, outlier observations far from your training data, and adversarial examples that maliciously cause your model to make prediction mistakes.
- You want a tool that has direct integration with Seldon Core and KServe.
Watch out
- Alibi Detect provides sophisticated detection algorithms out-of-the-box, but you may need to write custom code to integrate them into your monitoring system.
Example stacks
Airflow + MLflow stack
Installation
pip install alibi-detect