pymixef.backends.mmrm module

Dense repeated-measures Gaussian REML (MMRM) reference backend.

Expected fields in addition to the common backend contract are subject (or subjects/group), visit (or visits), and a covariance structure in covariance/covariance_structure/residual. Supported structures are homogeneous, diagonal, unstructured, compound symmetry, AR(1), heterogeneous AR(1), Toeplitz, heterogeneous Toeplitz, first-order ante-dependence, and spatial power. Missing visits are handled by selecting the observed submatrix of the shared visit covariance.

Satterthwaite denominator degrees of freedom use a finite-difference delta method for contrast variance and the REML covariance-parameter Hessian. kenward-roger is intentionally not claimed. An explicitly requested kenward-roger-approximate option uses a second-order delta adjustment and is labelled KR-inspired, not exact KR.

class pymixef.backends.mmrm.MMRMBackend

Bases: object

Dense subject-block MMRM REML engine.

name = 'mmrm'

fit(data, *, covariance=None, method='REML', df_method='satterthwaite', confidence_level=0.95, maxiter=1_000, tolerance=1e-8, compute_hessian=True, **options)

Parameters:

• data (Any)

• covariance (Any)

• method (str)

• df_method (str)

• confidence_level (float)

• maxiter (int)

• tolerance (float)

• compute_hessian (bool)

• options (Any)

Return type:

dict[str, Any]

pymixef.backends.mmrm.contrasts(beta, covariance, matrix, *, names=None, degrees_of_freedom=np.inf, confidence_level=0.95, method='Wald')

Build a tidy table for linear EMM/contrast estimates.

Parameters:

• beta (ArrayLike)

• covariance (ArrayLike)

• matrix (ArrayLike)

• names (Sequence[str] | None)

• degrees_of_freedom (ArrayLike | float)

• confidence_level (float)

• method (str)

Return type:

dict[str, list[Any]]

pymixef.backends.mmrm.estimated_marginal_means(beta, covariance, matrix, *, names=None, degrees_of_freedom=np.inf, confidence_level=0.95, method='Wald')

Build a tidy table for linear EMM/contrast estimates.

Parameters:

• beta (ArrayLike)

• covariance (ArrayLike)

• matrix (ArrayLike)

• names (Sequence[str] | None)

• degrees_of_freedom (ArrayLike | float)

• confidence_level (float)

• method (str)

Return type:

dict[str, list[Any]]

pymixef.backends.mmrm.fit_mmrm(data, **options)

Fit a Gaussian MMRM with dense subject covariance blocks.

Parameters:

• data (Any)

• options (Any)

Return type:

dict[str, Any]

pymixef.backends.mmrm.linear_inference(beta, covariance, matrix, *, names=None, degrees_of_freedom=np.inf, confidence_level=0.95, method='Wald')

Build a tidy table for linear EMM/contrast estimates.

Parameters:

• beta (ArrayLike)

• covariance (ArrayLike)

• matrix (ArrayLike)

• names (Sequence[str] | None)

• degrees_of_freedom (ArrayLike | float)

• confidence_level (float)

• method (str)

Return type:

dict[str, list[Any]]