DriftSource
Enums · AiDotNet.DriftDetection
Which distribution shift a DriftReport attributes drift to.
For Beginners
A model can go stale for two different reasons, and the fix differs:
- Concept drift — the relationship between inputs and the answer changed
(P(y|x)); the model’s predictions are getting wrong. Usually means retrain.
- Covariate drift — the inputs themselves shifted (P(x)); the model may not
be wrong yet, but it is now operating outside the data it learned from. Usually means investigate the data pipeline before accuracy degrades.
Reporting which is happening — not just that something is — is what makes this more actionable than a single error-stream detector.
Fields
| Field | Summary |
|---|---|
DriftSource Both |
Both concept and covariate drift detected. |
DriftSource Concept |
Concept drift only: the error stream drifted (P(y|x) changed). |
DriftSource Covariate |
Covariate drift only: the prediction distribution shifted (P(x) changed). |
DriftSource None |
No drift detected on either lens. |
int value__ |