bayesControls {icenReg} | R Documentation |
Control parameters for ic_bayes
Description
Control parameters for ic_bayes
Usage
bayesControls(
samples = 4000,
chains = 4,
useMLE_start = TRUE,
burnIn = 2000,
samplesPerUpdate = 1000,
initSD = 0.1,
updateChol = TRUE,
acceptRate = 0.25,
thin = 5
)
Arguments
samples |
Number of samples. |
chains |
Number of MCMC chains to run |
useMLE_start |
Should MLE used for starting point? |
burnIn |
Number of samples discarded for burn in |
samplesPerUpdate |
Number of iterations between updates of proposal covariance matrix |
initSD |
If |
updateChol |
Should cholesky decomposition be updated? |
acceptRate |
Target acceptance rate |
thin |
Amount of thinning |
Details
Control parameters for the MH block updater used by ic_bayes
.
The samples
argument dictates how many MCMC samples are taken. One
sample will be saved every thin
iterations, so there will a total of
thin * samples + burnIn
iterations. The burn in samples are not saved at all.
Default behavior is to first calculate the MLE (not the MAP) estimate and use
Hessian at the MLE to seed the proposal covariance matrix. After this, an updative
covariance matrix is used. In cases with weakly informative likelihoods,
using the MLE startpoint may lead to overly diffuse proposal or even undefined
starting values.
In this case, it suggested to use a cold start by setting useMLE_start = F
for the controls
argument. In this case, the initial starting proposal
covariance matrix will be a diagonal matrix with initSD
standard deviations.