abacusai.feature_drift_summary
Module Contents
Classes
Summary of important model monitoring statistics for features available in a model monitoring instance |
- class abacusai.feature_drift_summary.FeatureDriftSummary(client, featureIndex=None, name=None, distance=None, jsDistance=None, wsDistance=None, ksStatistic=None, predictionDrift=None, targetColumn=None, dataIntegrityTimeseries=None, nestedSummary=None, nullViolations={}, typeViolations={}, rangeViolations={}, catViolations={})
Bases:
abacusai.return_class.AbstractApiClass
Summary of important model monitoring statistics for features available in a model monitoring instance
- Parameters:
client (ApiClient) – An authenticated API Client instance
featureIndex (list[dict]) – A list of dicts of eligible feature names and corresponding overall feature drift measures.
name (str) – Name of feature.
distance (float) – Symmetric sum of KL divergences between the training distribution and the range of values in the specified window.
jsDistance (float) – JS divergence between the training distribution and the range of values in the specified window.
wsDistance (float) – Wasserstein distance between the training distribution and the range of values in the specified window.
ksStatistic (float) – Kolmogorov-Smirnov statistic computed between the training distribution and the range of values in the specified window.
predictionDrift (float) – Drift for the target column.
targetColumn (str) – Target column name.
dataIntegrityTimeseries (dict) – Frequency vs Data Integrity Violation Charts.
nestedSummary (list[dict]) – Summary of model monitoring statistics for nested features.
nullViolations (NullViolation) – A list of dicts of feature names and a description of corresponding null violations.
typeViolations (TypeViolation) – A list of dicts of feature names and corresponding type mismatches.
rangeViolations (RangeViolation) – A list of dicts of numerical feature names and corresponding prediction range discrepancies.
catViolations (CategoricalRangeViolation) – A list of dicts of categorical feature names and corresponding prediction range discrepancies.
- __repr__()
Return repr(self).