Spectral Closure

Module: spectralbridge.spectral_closure.

Focused scientific closure diagnostics; experimental, no production changes.

Paired partitions are descriptive balanced-panel sums of squares, not causal variance components or random regional effects. GlobalID is a polygon identity.

API glossary · Workflow / scientific context

Signatures and documentation below are collected from the current source. Advanced objects are not newly declared stable by inclusion here.

Audience: advanced.

spectral_shape

spectral_shape(values, weights)

Per-observation weighted unit norm and amplitude; never fit on test rows.

Audience: advanced.

closure_labels

closure_labels(table)

Explicit partial biological hierarchy; mixed cover is a proxy, not purity.

Audience: advanced.

paired_partition

paired_partition(first, second, labels, weights)

Exact class/polygon/year/interaction/residual SS on a complete two-year panel.

Labels must be stable per identity. Columns are spectral coordinates with explicit positive quadrature weights. Class and polygon are nested; the balanced time structure makes this decomposition orthogonal even with unequal class sizes. Components describe this panel, not a causal population.

Audience: advanced.

paired_partition_bootstrap

paired_partition_bootstrap(first, second, labels, weights, *, replicates=499, seed=20261015)

Class-stratified identity resampling preserves both years of each pair.

Audience: advanced.

one_vs_rest_metrics

one_vs_rest_metrics(truth, predicted, classes)

Multiclass one-versus-rest rates; undefined denominators stay NaN.

Audience: advanced.

cluster_rate_intervals

cluster_rate_intervals(truth, predicted, identities, classes, *, replicates=499, seed=20261015)

Conditional class-stratified whole-identity intervals for detection rates.