MixedEffectResults#

class mmer.MixedEffectResults(model)[source]#

Bases: object

MixedEffectRegressor fitted model results and inference interface.

Provides convenient access to fitted model components, variance estimates, and methods for prediction and random effect estimation.

Variables:

  • model (MixedEffectRegressor) – The source model instance.

  • fe_model (RegressorMixin) – Fitted fixed effects model.

  • m (int) – Number of output responses.

  • k (int) – Number of grouping factors.

  • random_effect_terms (list of RandomEffectTerm) – Learned random effect states.

  • residual_term (ResidualTerm) – Learned residual state.

  • log_likelihood (list) – History of log-likelihood values during training.

Parameters:

model (MixedEffectRegressor)

predict(X)[source]#

Predict using fixed effects component only.

Makes predictions using only the learned fixed effects model, ignoring random effects. For predictions that include random effects, use compute_random_effects() and add the total effect to predictions.

Parameters:

X (np.ndarray) – Covariates for prediction, shape (n, p).

Returns:

predictions – Predicted values from fixed effects only, shape (n, m).

Return type:

np.ndarray

Notes

Current implementation does not support random effects in prediction. To obtain predictions including random effects:

  1. Call compute_random_effects() to get random effect estimates

  2. Add total_effect to fixed effect predictions

Examples

>>> # Predict with fixed effects only
>>> y_pred = results.predict(X_new)

>>> # Predict with both fixed and random effects
>>> y_fixed = results.predict(X_new)
>>> _, total_re, _ = results.compute_random_effects(X_new, y_new, groups_new)
>>> y_pred_full = y_fixed + total_re
compute_random_effects(X, y, groups)[source]#

Estimate posterior random effects for new or existing data.

Parameters:

  • X (np.ndarray) – Features, shape (n, p).

  • y (np.ndarray) – Targets, shape (n, m).

  • groups (np.ndarray) – Grouping factors, shape (n, k).

Returns:

  • residuals (np.ndarray) – Estimated residuals after accounting for fixed and random effects.

  • total_effect (np.ndarray) – Total estimated random effects.

  • mu (tuple of np.ndarray) – Estimated random effects for each grouping factor.

property residual_covariance: ndarray#

Get the estimated residual covariance matrix.

property random_effects_covariances: tuple[ndarray]#

Get the estimated covariance matrices for each random effect grouping factor.

static cov_to_corr(cov)[source]#

Convert a covariance matrix to a correlation matrix. Protects against zero variances for numerical stability.

Parameters:

cov (ndarray)

Return type:

ndarray

property residual_correlation: ndarray#

Get the estimated residual correlation matrix.

property random_effects_correlations: tuple[ndarray]#

Get the estimated correlation matrices for each random effect grouping factor.

get_marginal_correlation(slope_covariates=None)[source]#

Compute the total marginal correlation matrix (m x m) for a single observation profile.

Parameters:

slope_covariates (list of np.ndarray or None, optional) – List of 1D arrays containing random slope covariates for each grouping factor. Omit the intercept (1.0) as it is automatically included.

Returns:

corr – Total marginal correlation matrix.

Return type:

np.ndarray

get_marginal_covariance(slope_covariates=None)[source]#

Compute the total marginal covariance matrix (m x m) for a single observation profile.

Parameters:

slope_covariates (list of np.ndarray or None, optional) – List of 1D arrays containing random slope covariates for each grouping factor. Omit the intercept (1.0) as it is automatically included. If a group has only a random intercept, pass None or an empty array for that group. If entirely None, assumes random intercepts only for all groups.

Returns:

cov – Total marginal covariance matrix combining residuals and random effects.

Return type:

np.ndarray

summary()[source]#

Generate a comprehensive text summary of the fitted multivariate mixed effects model.

Returns:

A formatted string containing model metadata and variance estimates.

Return type:

str