Convenience function for extracting the pointwise log-likelihood
matrix or array from a
stanfit object from the rstan package.
Note: recent versions of rstan now include a
loo() method for
stanfit objects that handles this internally.
extract_log_lik(stanfit, parameter_name = "log_lik", merge_chains = TRUE)
A character string naming the parameter (or generated quantity) in the Stan model corresponding to the log-likelihood.
merge_chains=TRUE, an \(S\) by \(N\) matrix of
(post-warmup) extracted draws, where \(S\) is the size of the posterior
sample and \(N\) is the number of data points. If
merge_chains=FALSE, an \(I\) by \(C\) by \(N\) array, where
\(I \times C = S\).
Stan does not automatically compute and store the log-likelihood. It is up to the user to incorporate it into the Stan program if it is to be extracted after fitting the model. In a Stan model, the pointwise log likelihood can be coded as a vector in the transformed parameters block (and then summed up in the model block) or it can be coded entirely in the generated quantities block. We recommend using the generated quantities block so that the computations are carried out only once per iteration rather than once per HMC leapfrog step.
For example, the following is the
generated quantities block for
computing and saving the log-likelihood for a linear regression model with
N data points, outcome
y, predictor matrix
beta, and standard deviation
for (n in 1:N) log_lik[n] = normal_lpdf(y[n] | X[n, ] * beta, sigma);
Stan Development Team (2017). The Stan C++ Library, Version 2.16.0. https://mc-stan.org/
Stan Development Team (2017). RStan: the R interface to Stan, Version 2.16.1. https://mc-stan.org/