Researcher Information¶
Module: spectralbridge.researcher_information.
Experimental researcher-cover spectral information and synthetic fractions.
Only recorded cover categories define biological groups. Utilities do not infer physical purity, recover missing labels or validate field abundance.
API glossary ยท 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.
Audience: advanced.
identity_weights ¶
identity_weights(identities, labels=None)
Each identity has total weight one; optionally balance class totals.
Audience: advanced. Also exported as spectralbridge.library_analysis.project_simplex.
project_simplex ¶
project_simplex(values)
Euclidean projection onto nonnegative sum-one fractions (sorting rule).
Audience: advanced.
weighted_auc ¶
weighted_auc(truth, scores, weights=None)
Weighted binary ROC AUC per column, with half credit for tied values.
Audience: advanced.
univariate_discrimination ¶
univariate_discrimination(train, train_labels, train_ids, test, test_labels, test_ids)
Balanced-prior single-coordinate LDA; threshold learned on training only.
Audience: advanced.
orthogonal_feature_path ¶
orthogonal_feature_path(X, labels, count=100, weights=None)
Greedy complementary coordinates via balanced one-hot residual gains.
Every input must belong to caller's training fold. Does not examine targets or claim globally optimal sensor design. Orthogonalizes duplicate channels.
Audience: advanced.
disconnected_derivative ¶
disconnected_derivative(X, wavelengths, order=1, max_gap=8)
Derivatives within supported segments, with no atmospheric-gap bridges.
Audience: advanced.
disjoint_parent_bundles ¶
disjoint_parent_bundles(table, count, seed)
Sample one observation per class/ID, never reuse an ID across bundles.
Audience: advanced.
fraction_metrics ¶
fraction_metrics(truth, pred)
Per-class fraction errors and presence metrics at a fixed .10 cutoff.
Audience: advanced.