NCA-AIIO · Question #64
Which assumption is violated when deployment data follows a different distribution than training data?
The correct answer is B. Stationarity. Stationarity is the assumption that the statistical properties of the data-generating process remain constant over time (or across contexts). When a model trained on one distribution is deployed against a different distribution - a phenomenon called dataset shift or…
Question
Which assumption is violated when deployment data follows a different distribution than training data?
Options
- AIndependence
- BStationarity
- CLinearity
- DCompleteness
How the community answered
(43 responses)- A2% (1)
- B88% (38)
- C2% (1)
- D7% (3)
Explanation
Stationarity is the assumption that the statistical properties of the data-generating process remain constant over time (or across contexts). When a model trained on one distribution is deployed against a different distribution - a phenomenon called dataset shift or distribution shift - the stationarity assumption is violated, and model performance degrades. Independence (A) is violated when data points are correlated (e.g., time series without proper handling). Linearity (C) is a modeling assumption about the functional form of relationships. Completeness (D) refers to whether all necessary features or data points are present. Distribution mismatch between training and deployment is precisely what stationarity violation means in a machine learning context.
Topics
Community Discussion
No community discussion yet for this question.