pymixef.backends.glmm module

Dense Laplace-approximation generalized linear mixed-model backend.

Implemented likelihoods are Bernoulli, binomial-logit, Poisson-log, and negative-binomial-2-log. The engine jointly optimizes fixed effects and random covariance parameters while solving the conditional random-effect mode with an exact Newton Hessian at every outer evaluation. Constants in the conditional probability masses are retained.

Extended families, zero inflation, censoring, AGHQ, and noncanonical links are rejected explicitly; the reference engine never substitutes PQL or another scientifically different estimator.

pymixef.backends.glmm.GLMMBackend

alias of LaplaceGLMMBackend

class pymixef.backends.glmm.LaplaceGLMMBackend

Bases: object

Dense grouped-random-effect Laplace GLMM reference backend.

name = 'laplace'

fit(data, *, family=None, maxiter=500, tolerance=1e-7, inner_maxiter=100, inner_tolerance=1e-9, compute_hessian=True, **options)

Parameters:

• data (Any)

• family (Any)

• maxiter (int)

• tolerance (float)

• inner_maxiter (int)

• inner_tolerance (float)

• compute_hessian (bool)

• options (Any)

Return type:

dict[str, Any]

pymixef.backends.glmm.fit_glmm(data, **options)

Fit a supported GLMM by dense first-order Laplace approximation.

Parameters:

• data (Any)

• options (Any)

Return type:

dict[str, Any]