Expand description
Re-exported for the Python bindings (gam-pyffi), which build their
analytic-penalty registry through the single shared descriptor parser that
also serves the in-process workflow pipeline. Exposed here so PyFFI can name
it without the (crate-private) workflow module being publicly reachable.
§Shared model-estimation contract types
Lower-layer types that both the families layer (which constructs penalty
and dispersion specifications and propagates estimation errors) and the
solver layer (which consumes them) need to name. Hosting them here breaks
the families → solver::estimate back-edge that #1135 tracks: families now
import these from crate::model_types instead of reaching up into
crate::solver::estimate.
§Layering
These types depend only on lower or sibling layers (linalg, terms,
families’ error types) — never on solver. EstimationError carries
#[from] conversions for the family error types it wraps, which is the
allowed downward direction.
Structs§
- Active
Linear Constraint Block - Active row block of the joint linear inequality constraint matrix at the converged inner iterate.
- Adaptive
Regularization Options - Certified
Rung - The rung a CERTIFICATE was decided against, in owned form (#2530).
- Curvature
Floor Clearance - Analytic optimality certificate at the returned optimum (#934).
- Dispersion
- Dispersion contract used by inferential covariance and reference distributions.
- Display
Coefficient Uncertainty - A definition-consistent coefficient-uncertainty view (#2296): standard
errors and (optional) covariance from the SAME definition, tagged with
that definition. Produced by
UnifiedFitResult::display_coefficient_uncertainty. - FitArtifacts
- Post-fit artifacts needed by downstream diagnostics/inference without re-running PIRLS.
- FitGeometry
- Coefficient geometry retained at convergence for inference and post-fit diagnostics.
- FitInference
- FitOptions
- Fitted
Block - Inference quantities for one coefficient block.
- Outer
Criterion Certificate - Projected
KktResidual - KKT residual
r = grad_beta L_pen(beta_hat)at the converged inner iterate, tagged with the exact represented subspace. - Rail
Coordinate - One outer smoothing coordinate certified stationary-at-asymptote: it has no
interior optimum (its gradient never clears the fixed bound), but the fitted
model has provably reached its rail limit to within tolerance along it
(#2348 Inc 1 / #2299 layer 3, #2337 Thm 2.1). The certified facts are the
per-coordinate outputs of
crate::rho_optimizer::asymptote_certificate::assess_coordinate. - Railed
Coordinate Fact - One railed coordinate together with the interval and margin it was judged against (#2530).
- Unified
FitResult - Unified fit result for all model types (standard GAM, GAMLSS, survival).
- Unified
FitResult Parts - Working
Geometry - Owned diagonal working-set evidence at convergence.
Enums§
- Block
Role - Role of a coefficient block within a multi-parameter model.
- Coefficient
Covariance Definition - Exact coefficient-covariance definition (#2296).
- Coefficient
Prior Mean - Programmatic prior mean for a coefficient penalty block.
- Covariance
Declined - Why a fit that could have published a coefficient covariance deliberately did not (gam#2718, gam#2484).
- Curvature
Evidence - What curvature evidence a certificate actually has (#2561).
- Estimation
Error - A comprehensive error type for the model estimation process.
- Execution
Path - Truthful execution-path classifier for a fit’s hot inner solve (issue #1017).
- Fitted
Link State - Inference
Covariance Mode - Which posterior covariance definition an uncertainty surface was built from.
- KktResidual
Subspace - Subspace represented by a stored KKT residual.
- Outer
Stationarity Certificate - The stationarity equation that certified an outer optimum.
- Penalty
Spec - A penalty specification for the public estimate API.
- Rail
Tail Evidence - What established a rail coordinate’s tail law, and the standard it cleared (#2348 Inc 5 build-out).
- Smoothing
Correction Method - Serialized provenance of a retained smoothing-uncertainty correction.
Constants§
- NO_
CRITERION_ AT_ EXACT_ FIT - The one explanation every ranking surface gives for an absent criterion.
Functions§
- criterion_
display - Render a possibly-absent criterion for human output.
- ensure_
finite_ scalar - Public wrapper returning
Stringerrors for use outside the estimation module. - is_
zero_ dispersion_ boundary - The exact-fit Gaussian boundary, from the three quantities that define it.
- saved_
latent_ cloglog_ state_ from_ fit - saved_
mixture_ state_ from_ fit - saved_
sas_ state_ from_ fit - validate_
all_ finite - Public wrapper returning
Stringerrors for use outside the estimation module. - validate_
dense_ hessian_ export - Validate the structural integrity of an exported penalized Hessian.
- validate_
explicit_ dense_ hessian_ for_ whitening - Validate that a saved penalized Hessian is an explicit dense precision matrix suitable for HMC/NUTS whitening.