init_tuning: Initialize tuning for M-H kernels.

In fintzij/BDAepimodel: Fit Stochastic Epidemic Models in R via Bayesian Data Augmentation

Description

Initialize tuning for M-H kernels.

Usage

init_tuning(epimodel, tune_params, cov_mtx = NULL,
  cov_scale = (2.38^2/length(tune_params)), mintune = 2,
  maxtune = 24, ntu = 500, tunewt = 0, target_accept = c(0.15,
  0.5))

Arguments

tune_params	character vector of parameters for which proposal is to be tuned.
cov_mtx	covariance matrix for parameters to be used in M-H updates - supercedes the covariance matrix that already exists in the epimodel object if it was specified in both.
cov_scale	scale tuning parameter for the covariance matrix - i.e. c * Sigma, defaults to (2.38^2 / length(tune_params)).
mintune	minimum number of tuning loops, defaults to 2.
maxtune	maximum number of tuning loops, defaults to 24.
ntu	number of parameter updates per tuning loop, defaults to 500.
tunewt	initial weight to put on covariance matrix from most recent tuning loop if it is desired to tune the covariance matrix according to a weighted average of the previous covariance matrix and the new one. Defaults to 0 - i.e. scale tuning only. The tuning weight will decrease in proportion to the total number of iterations run - e.g. if the initial weight is 1, the covariance matrix will receive weight 1. The next 500 will receive weight 0.5. the 500 after that will receive weight 500/1500.
target_accept	vector of length 2 giving the acceptable range for the acceptance rate for M-H parameter updates, defaults to c(0.15, 0.5). Tuning ends once the acceptance rate either falls within this range or the maximum number of tuning loops is exhausted.

Value

epimodel object with initialized tuning settings

Examples

## Not run: init_tuning(epimodel, 
                 tune_params = c("beta", "mu", "rho"),
                 cov_mtx = matrix(c(1, -0.865, -0.747, 
                                     -0.865, 1, 0.831,
                                     -0.747, 0.831, 1), nrow = 3),
                 cov_scale = 0.02, 
                 mintune = 2,
                 maxtune = 10, 
                 ntu = 500,
                 tunewt = 0,
                 target_accept = 0.15, 0.5)
## End(Not run)

fintzij/BDAepimodel documentation built on Sept. 20, 2020, 1:44 p.m.