Struct ShiftDetector 

pub struct ShiftDetector { /* private fields */ }

Implementations

impl ShiftDetector

pub fn from_dataset(dataset: &BinnedDataset) -> Self

Create a detector from a training BinnedDataset

Pre-computes histograms from the training data for efficient comparison with inference data.

Uses dynamic histogram sizing based on each feature’s actual num_bins to minimize memory usage (vs fixed 256-bin arrays).

pub fn from_raw( features: &[f32], num_features: usize, feature_names: Option<&[String]>, ) -> Self

Create from raw feature data

Arguments

• features - Row-major feature matrix
• num_features - Number of features
• feature_names - Optional feature names

Note: Raw data uses MAX_HISTOGRAM_BINS (256) since actual bin count is unknown. For memory efficiency with binned data, use from_dataset instead.

pub fn with_thresholds(self, warning: f32, critical: f32) -> Self

Set custom thresholds

pub fn with_metric<M: DistributionMetric + 'static>(self, metric: M) -> Self

Set a custom distribution metric

pub fn check(&self, inference_data: &BinnedDataset) -> ShiftResult

Check inference data for distribution shift

Compares each feature’s distribution in the inference data against the reference (training) distribution. Automatically adjusts thresholds based on sample size differences.

Uses dynamic histogram sizing matching reference histograms for memory efficiency.

pub fn feature_info(&self) -> &[FeatureInfo]

Get reference feature info

pub fn warning_threshold(&self) -> f32

Get the warning threshold

pub fn critical_threshold(&self) -> f32

Get the critical threshold

pub fn metric_name(&self) -> &'static str

Get the metric name