Module cohort

Cohort-level temporal drift analytics.

Measures how a group of entities evolves collectively in embedding space, complementing the single-entity analytics in crate::calculus.

Key metrics

Metric	What it captures
centroid_drift	How the group center moved
dispersion_change	Did the group spread or compress?
convergence_score	Are entities moving in the same direction?
outliers	Entities drifting abnormally vs the group

Structs

CohortDriftReport

Full cohort drift analysis between two time points.

CohortOutlier

An entity whose drift deviates significantly from the cohort.

Functions

centroid 🔒

Compute the centroid (element-wise mean) of a set of vectors.

cohort_drift

Compute cohort-level drift analysis.

compute_convergence_score 🔒

Mean cosine similarity of all drift vectors against their mean direction.

nearest_vector_at

Find the vector closest in time to target within a trajectory.