Skip to contents

Compute Pointwise Log-Likelihood

Source: R/fitter.R
log_lik_matrix.Rd

Compute pointwise log-likelihood values. Named log_lik_matrix (rather than log_lik) so that bayesim does not mask brms::log_lik or the rstantools log_lik generic for users who load bayesim alongside brms.

Usage

log_lik_matrix(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 where:

  • S = number of posterior draws (rows)

  • N = number of observations (columns)

  • Entry (s, i) is log p(y_i | parameters_s)

This is the brms/loo convention (draws as rows), matching brms::log_lik and what loo::psis/loo::E_loo/loo::relative_eff expect.