Deepchecks is an open-source Python library for validating and monitoring machine-learning data and models. It helps production teams detect data-quality problems, drift, leakage, performance degradation, imbalance, and unexpected bias before they undermine reliability. section 11.2.1
Deepchecks supports several production checks:
- Validate data integrity by checking consistency, missing values, data types, and other anomalies. section 11.2.1
- Compare training and incoming data distributions to identify drift that may signal future performance degradation. section 11.2.1
- Customize existing checks or create checks for application-specific monitoring requirements. section 11.2.1
Detecting image-property drift
Install the library with pip, load training and inference datasets as VisionData objects, and run ImagePropertyDrift(). The check compares properties such as brightness, contrast, aspect ratio, area, and color intensity, then saves an HTML report and exposes drift scores. In one test, brightness showed a drift score of 0.6189 while aspect ratio and area showed 0.0, isolating brightness as the major distribution change. section 11.2.1
