Compute Posterior Expectation Predictions

Compute posterior draws of the expected value of the response distribution (mu, without observation noise). This is the epred quantity used by brms' loo_R2() and bayesim's r2_loo_metric() (F3). It must NOT include observation-level noise — only the model's conditional mean.

Fitters that cannot provide expectation predictions should return NULL; the r2_loo metric then degrades to NA. The default Fitter method returns NULL.

Usage

predict_epred(fitter, fit_result, newdata = NULL)

Arguments

fitter

An S7 Fitter object

fit_result

A bayesim_fit_result object from fit_model()

newdata

Data frame with observations. If NULL, uses training data.

Value

A matrix with dimensions S x N (draws x observations), or NULL if not supported by the fitter.