NEWS.md
run_stage
argument: pdist_update_n
. By default this is set to 500, so after 500 iterations the proposal distribution will be regenerated and used for subsequent iterations of the sampler.accept_rate
now operates over the entire sampler data store, and by default will return the rate of accepted newly generated random effects over a window covering the last 200 iterations. This will effect how the acceptance rate progress bar display progress, being more responsive to local changes in acceptance rate. This also means secondary runs of the adaptation stage can take into account unique random effect samples from previous runs for it’s internal testing.relabel_samples
, added to allow a user to change the label indicating the samples stage of origin. This function takes a pmwg samplers object, a set of indices corresponding to the samples you wish to relabel and two arguments detailing to expected value for the current stage (default “burn”) and the stage to would like to relabel the samples to (default “adapt”). The goal for the function is to take samples generated in a burn in stage that after visualising appear to have successfully converged around the posterior and make them available for internal function that check for adaptation and generate the proposal distribution used in the final sampling stage.sampled_forstmann
, a pmwgs object with a simple LBA model (requires rtdists package) applied to the previously included forstmann
dataset. Includes 3 runs of the sampler with low numbers of iterations and particles. It has the data element stripped so see the Details in the documentation for the object (?sampled_forstmann
) for advice on using the object.gibbs_step
internal functioncheckmate
used for testing arguments to run_stage
as_mcmc
function to export mcmc objects (from coda package) or a list of mcmc objects for the covariance matrix and subject random effects.init
function to reduce time for sampler initialisation.