MaxDiffMixedLogit.sample_posterior_predictive

MaxDiffMixedLogit.sample_posterior_predictive(task_df=None, extend_idata=True, **kwargs)

Sample from the posterior predictive distribution.

Appropriate for in-sample posterior predictive checks on training data. When task_df is provided the model data containers are updated via pm.set_data before sampling.

Warning

Due to the sequential best-worst likelihood, worst_pick is conditioned on the observed best position, not on the sampled best_pick. The joint draw is therefore incoherent for generative or counterfactual use. See the class-level Posterior predictive limitations note for details, and use predict_choices() / apply_intervention() instead for any out-of-sample or counterfactual simulation.