DATA DRIFT & MODEL EVALUATION
Model monitoring.
Trace a distribution shift. Understand its impact on your model.
Four features shifted by 2 training standard deviations.
Features flagged
4/ 13
Distribution or missingness alertsModel accuracy
53.7%
-44.4 pp vs untouched holdoutMacro F1
0.479
Equal weight across classesCurrent observations
54
Perturbed test rows · original labelsFEATURE DISTRIBUTION
flavanoids
ReferenceCurrent
Feature diagnostics 13
Select a feature to inspect its distribution.
| Feature | PSI ↓ | KS q-value | Missing (current) | Status |
|---|---|---|---|---|
| 2.794 | < 0.001 | 0.0% | Alert | |
| 2.424 | < 0.001 | 0.0% | Alert | |
| 2.313 | < 0.001 | 0.0% | Alert | |
| 2.173 | < 0.001 | 0.0% | Alert | |
| 0.470 | 0.682 | 0.0% | No alert | |
| 0.288 | 0.937 | 0.0% | No alert | |
| 0.277 | 0.720 | 0.0% | No alert | |
| 0.199 | 0.850 | 0.0% | No alert | |
| 0.183 | 0.850 | 0.0% | No alert | |
| 0.181 | 0.720 | 0.0% | No alert | |
| 0.172 | 0.345 | 0.0% | No alert | |
| 0.147 | 0.937 | 0.0% | No alert | |
| 0.137 | 0.929 | 0.0% | No alert |
13 features shown · BH-adjusted KS p-values · threshold changes are exploratory
