hsfs.core.statistics_comparison_config #
StatisticsComparisonConfig #
Defines a single metric comparison for feature monitoring.
Each instance specifies the metric to evaluate, a threshold for the difference, and whether the comparison is relative, strict, or against a fixed reference value.
Exactly one of metric (scalar) or distribution_metric (distribution) must be set.
binning_strategy property #
binning_strategy: str | None
Strategy used to discretize continuous features into histogram bins.
custom_bin_edges property #
Custom bin edges (required when binning_strategy is CUSTOM_EDGES).
distribution_metric property #
distribution_metric: str | None
Distribution distance metric for PDF-based comparison.
relative property #
relative: bool | None
Whether to compute the relative difference instead of the absolute one.
smoothing_epsilon property #
smoothing_epsilon: float | None
Small additive constant applied to bin probabilities to avoid log(0).
specific_value property #
specific_value: float | None
Fixed reference value to compare against instead of reference statistics.