Give us a star ⭐️ on GitHub to support the project!
No model lasts forever. Detect shifts in model predictions and input data to get ahead of potential issues.
GET STARTED
Data drift is a shift in the statistical properties of the input data. When the distributions of the model predictions or features change, this might signal a shift in the model environment and lead to model performance decay.
Evidently simplifies detecting and exploring data drift. You can catch issues before they affect user experience and speed up debugging by quickly understanding where the change comes from.
Did your model start predicting one class more often than before? Is there a change in predicted probabilities? Quickly visualize and catch unexpected behavior.
Get an instant overview of the dataset drift. Find which features contribute to it and explore the distributions to interpret the change. Pick any of 15+ drift detection methods, or pass your own.
Make sense of the unstructured data changes. Monitor text descriptors, such as text length or sentiment. Find specific words that help explain the shifts.
Working directly with embeddings? Choose one of the available methods to detect distribution drift in the input vectors.
Supplement distribution drift detection with rule-based checks. Detect data loss, values outside the min-max range, or new categories.
Capture how feature distributions change with continuous monitoring. Log, track, and explore data drift over time to know if your models operate in a familiar environment.
Easily add Evidently to existing workflows, no matter where you deploy.
