R2admb (version 0.7.13)

mcmc.control: Control options for MCMC after ADMB fitting

Description

Determines the options (number of steps, save interval, etc.) for running MCMC based on the estimated mode (maximum likelihood estimate) and parameter variance-covariance matrix

Usage

mcmc.control(mcmc = 1000, mcmc2 = 0, mcsave, mcnoscale = FALSE, mcgrope = FALSE, mcmult = 1, mcmcpars = NULL)

Arguments

mcmc

Total number of MCMC steps

mcmc2

MCMC2 steps (see ADMB-RE manual)

mcsave

Thinning interval for values saved in the PSV file. Default is pmax(1,floor(mcmc/1000)), i.e. aim to save 1000 steps

mcnoscale

don't rescale step size for mcmc depending on acceptance rate

mcgrope

(double) Use a candidate distribution that is a mixture of a multivariate normal and a fatter-tailed distribution with a proportion mcmcgrope of the fatter-tailed distribution; the ADMB manual suggests values of mcgrope between 0.05 and 0.1

mcmult

Multiplier for the MCMC candidate distribution

mcmcpars

(character) vector of parameters to track in MCMC run. At least one must be specified. ADMB produces two kinds of output for MCMC. For any sdreport parameters it will produce a hst file that contains a summary histogram; mcmcpars constructs appropriate sdreport parameters in the auto-generated TPL file. Step-by-step output for all parameters (regulated by mcsave) is saved in the PSV file.

Value

Returns a list of options suitable for passing as the mcmc.opts argument to do_admb

Details

See the AD Model Builder reference manual. The mcrb option (reduce correlation of the Hessian when constructing the candidate distribution) and the mcseed options (seed for random number generator) are not yet implemented; mcnoscale above may not work properly

Examples

mcmc.control(mcmc=2000)