Model performance
Model performance monitors significant changes in the model’s output and provides feedback as compared to the expected trends.
Superwise offers you a number of Model performance policy templates:
Prediction shift
Using the prediction shift policy, you can observe the shift in the model’s predictions at the segment level.
Metric | Role | Segments |
---|---|---|
distribution_shift | Numeric or categorical Predictions | All Segments |
Decision proportion
The decision proportion policy tracks whether the proportion of a specific decision is above a certain threshold (e.g., approval is below 10%) for Boolean types.
Metric | Role | Segments |
---|---|---|
proportion | Boolean Predictions | All Segments |
Label shift
The label shift policy tracks the shift in the model prediction at the segment level.
Metric | Role | Segments |
---|---|---|
distribution_shift | Numeric or categorical Labels | All Segments |
Label proportion
The label proportion policy monitors what proportion of a specific decision is above a certain threshold (e.g., approval is below 10%) for Boolean types.
Metric | Role | Segments |
---|---|---|
proportion | Boolean Labels | All Segments |
Updated over 2 years ago