API reference

The public API is re-exported from the top-level xpectra namespace. The tables below are generated from the installed package; click any name for the full signature and docstring.

Configuration & data

xpectra.TrustConfig
xpectra.load_labeled	Labeled spectra (4 studies, unknowns dropped), grouped 6-class labels.
xpectra.load_unknown	The kedzierski_2019_u unlabeled environmental pool, on the labeled grid.
xpectra.load_external	One external augmentation pool (openspecy / simple / awi2024) from the norm route.
xpectra.loso_splits	Leave-one-study-out folds over the labeled studies.
xpectra.leave_one_class_out	Masks for synthetic OOD: train without one class, hold it out as OOD.

Metrics

xpectra.expected_calibration_error	Top-label ECE plus the per-bin table used for reliability diagrams.
xpectra.multiclass_brier	Mean squared error between one-hot labels and predicted probabilities.
xpectra.risk_coverage	Selective-risk curve: abstain on the least confident samples first.
xpectra.aurc	Area under the risk-coverage curve (lower is better).
xpectra.selective_metrics_at_coverage	Accuracy/macro-F1/MCC on the retained fraction at a target coverage.

Calibration

xpectra.CalibratedModel	One estimator + one calibration method behind a uniform proba API.
xpectra.TemperatureScaler	Single-parameter temperature scaling fitted by NLL minimization.
xpectra.reliability_curve	Per-bin confidence/accuracy table for reliability diagrams.

Out-of-distribution detection & abstention

xpectra.AbstainingClassifier	Predict a class when the OOD score is below tau, otherwise abstain.
xpectra.MaxSoftmax	1 - max predicted probability.
xpectra.Entropy	Normalized Shannon entropy of the predictive distribution.
xpectra.Energy	Negative log-sum-exp of logits (log-proba fallback for CV-calibrated models).
xpectra.Mahalanobis	Min per-class Mahalanobis distance in PCA space (shared LW covariance).
xpectra.KNNDistance	Distance to the k-th nearest training sample in PCA space.
xpectra.SpectralAngleScorer	Min spectral angle to the class-mean spectra (norm route only).

Semi-supervised learning

xpectra.self_train
xpectra.label_spread	LabelSpreading with a sparse kNN kernel on PCA-reduced features.
xpectra.ssl_loso_experiment	Per LOSO fold: supervised-only vs self-training vs label-spreading, all evaluated on the held-out study.

Weathering analysis

xpectra.band_area	Integrated area (or peak height) of a wavenumber window.
xpectra.carbonyl_index	C=O band relative to the CH2-scissoring reference band.
xpectra.weathering_report	Spearman correlations of carbonyl index vs confidence and OOD scores.
xpectra.unknown_composition	Class fractions among the unknowns, with and without abstentions.

Module index

xpectra.config	Configuration for the xpectra weathered-unknowns study.
xpectra.data	Data loading for the trust study, built on xpectra.pipeline.
xpectra.metrics	Calibration and selective-prediction metrics not provided by sklearn.
xpectra.calibration	Leakage-safe probability calibration wrappers.
xpectra.ood	Out-of-distribution scoring and abstention.
xpectra.ssl	Semi-supervised learning: do the 4,058 unlabeled weathered spectra help?
xpectra.weathering	Band-ratio weathering indicators (carbonyl index) and composition tables.
xpectra.protocols	Experimental orchestrators shared by the driver scripts.
xpectra.plotting	Publication figures for the trust study.
xpectra.pipeline	Reusable training and prediction utilities for the FTIR notebooks.