Spectral Library Analytics¶
Module: spectralbridge.spectral_library_analytics.
Experimental integrated spectral-library analytics, with immutable inputs.
A source-specific scientific experiment, not a stable top-level package API. Read the predeclared design before interpreting predictive or mixture outputs.
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.
stable_seed ¶
stable_seed(name, seed=20261014)
Derive a deterministic 32-bit-compatible random seed from a name and base seed.
Audience: advanced.
load_analytics_inputs ¶
load_analytics_inputs(source, compact, output)
Verify upstream compact ownership, write only metadata/reference index.
Audience: advanced.
scheme_data ¶
scheme_data(table, values, field)
Subset paired arrays to assigned, nonmissing labels with nonzero spectral support.
Audience: advanced.
fold_support ¶
fold_support(table, train, test, config)
Report per-label training/test identity counts and explicit closed/open/insufficient-support statuses.
Audience: advanced.
prepare_fold ¶
prepare_fold(table, arrays, actual, fold, kind, train, test, config)
Class support/calibration gates, using actual flight targets when required.
Audience: advanced.
representation_values ¶
representation_values(arrays, representation)
Decode a representation name into its NEON/OLI sensor and feature budget; the arrays argument is retained but unused.
Audience: advanced.
fit_predict_case ¶
fit_predict_case(table, arrays, waves, prepared, context, representation, model_name, root, identity, config)
One owned restartable fit; no final-test labels influence fitting choices.
Audience: advanced.
profile_learning ¶
profile_learning(source, compact, output, config=None)
Profile declared models without computing or inspecting outer-test scores.
Audience: advanced.
run_learning ¶
run_learning(source, compact, output, *, config=None, first_pass=False, workers=2)
Run declared classical models/representations with per-fit restart checks.
All-flight diagnostics use logistic/full/OLI. Plant/site/year comparisons use all model baselines at full/OLI plus logistic selected-coordinate curves. No ordinary pixel-level observations are materialized.