Drift
The drift monitor measures how different the selected data distribution is from the baseline.
Superwise offers you a number of templates for policies that monitor drift.
![Screen Shot 2022-03-29 at 17.17.30.png 1974](https://files.readme.io/f212b46-Screen_Shot_2022-03-29_at_17.17.30.png)
Feature stability
The feature stability metric monitors the five most important features that contribute to long-lasting drift across the entire data set.
Metric | Role | Segments |
---|---|---|
distribution_shift | Top 5 important features | Entire set |
Dataset Shift
You can scan input drift at the segment level.
Metric | Role | Segments |
---|---|---|
input_drift | Cross Features | All Segments |
Training-Serving skew
alert when input is above a certain threshold from the baseline
Metric | Role | Segments |
---|---|---|
input_drift | Cross Features | Entire set |
Updated over 2 years ago