Module discovery 

Shapelet discovery & ranking: candidate generation, discriminative quality scoring (information gain / F-statistic), and top-K selection with self-similarity pruning.

Builds on the Phase 57 distance core (shapelet_distance, Shapelet): given a labeled training curve set, enumerate candidate subsequences (exhaustively or via deterministic seeded random sampling bounded by max_candidates), score each by how well its distance orderline separates the class labels, and greedily select a non-redundant ShapeletSet.

Determinism

The candidate SET is fixed by config.seed before scoring, scoring is pure, and the final ranking uses f64::total_cmp on quality with a (series_idx, start, length) tie-break. Two fits with the same config are therefore byte-identical, and the sequential (parallel off) result matches the parallel one exactly.

Structs

ShapeletDiscoveryConfig

Configuration for discover_shapelets.

ShapeletSet

A discovered, ranked, non-redundant set of shapelets.

Enums

QualityMeasure

Discriminative quality measure used to score a candidate shapelet’s distance orderline against the class labels.

Functions

discover_shapelets

Discover a non-redundant ShapeletSet from a labeled training curve set.