Skip to content

index.md

Latest commit

 

History

History
35 lines (33 loc) · 2.41 KB

index.md

File metadata and controls

35 lines (33 loc) · 2.41 KB
Raw

API reference

  • tierra_plus.diagnostics.spectral_design(template, sigma, overlap, ...): noise-whitened synthetic target/nuisance geometry with explicit shape overlap.
  • tierra_plus.diagnostics.gaussian_posterior(design, observations, rho=...): exact Gaussian conditioning with a correlated unit-variance prior.
  • tierra_plus.diagnostics.nuisance_projected_gls(design, observations): prior-free target estimate and variance; refuses non-identifiable designs.
  • tierra_plus.diagnostics.constrained_linear_estimator(design, rho, max_nuisance_response): optimal linear readout and expected MSE under a nuisance-response cap.
  • tierra_plus.diagnostics.network.fit(x, y, seed=..., ...): seeded NumPy tanh regressor with training-only normalization, explicit hidden activations, paired direction discovery and reloadable numerical state.
  • tierra_plus.models.Spectrum(wavelength_microns, flux, uncertainty=None): immutable spectrum values with finite-array and grid validation.
  • tierra_plus.inference.spectral.UniformPrior(name, lower, upper): finite ordered bounds for one parameter.
  • tierra_plus.inference.spectral.SpectralLikelihood(observed, model, priors): prior transform and normalized diagonal Gaussian likelihood for a callable model.
  • tierra_plus.inference.NestedSampler(config=..., model_bridge=...): optional dynesty sampler; run() returns weighted samples, log likelihoods, log evidence/error and metadata. A missing backend raises an explicit error.
  • tierra_plus.contamination.contaminate_transit(planet, photosphere, spots, faculae, spot_fraction=0, facula_fraction=0): quiet-chord stellar intensity mixing.
  • tierra_plus.forward.opacity.ensemble.posterior_weights_for_case(case, priors=None): normalized weights for fixed spectra, not fitted-model marginal likelihoods. Extreme positive uncertainties use log-space normalization; an all-zero representable likelihood fails explicitly instead of selecting a NaN winner.
  • tierra_plus.provenance: content hashing, append-only run-log and replay-manifest helpers. Structured hashes require string mapping keys and finite JSON numbers. File-byte SHA256 and structured-payload digests retain separate identity namespaces.

See the source docstrings for parameter definitions, the causal protocol for experimental assumptions, and the tutorial for the supporting nested-sampling run.