mlParams: Parameters for evaluating marginal likelihood
In CNPBayes: Bayesian mixture models for copy number polymorphisms

Description Usage Arguments Details Value See Also Examples

Parameters for evaluating marginal likelihood

mlParams(root = 1/10, reject.threshold = exp(-10),
  prop.threshold = 0.5, prop.effective.size = 0.05,
  ignore.effective.size = FALSE, ignore.small.pstar = FALSE,
  warnings = TRUE)

`root`	length-one numeric vector. We exponentiate `p(theta* \| ...)` by the value of `root`. Values less than one reduce the influence of extreme observations.
`reject.threshold`	length-one numeric vector between 0 and 1. Probabilities in the reduced Gibbs model for the thetas that are less than this threshold are flagged.
`prop.threshold`	length-one numeric vector between 0 and 1. If more than `prop.threshold` are flagged, the marginal likelihood is not evaluated.
`prop.effective.size`	Logical. If the effective size / total iterations is less than `prop.effective.size`, the marginal likelihood is not evaluated (unless `ignore.effective.size` is `TRUE`).
`ignore.effective.size`	Logical. By default, if the effective size of any theta chain is less than 0.02, the marginal likelihood is not calculated. If this parameter is set to TRUE, the effective size is ignored. Occasionally, the effective size is misleading. See details.
`ignore.small.pstar`	Logical. Flags from the `reject.threshold` parameter are ignored and the marginal likelihood is calculated.
`warnings`	Logical. If FALSE, warnings are not issued. This is FALSE by default for the marginalLikelihood-list method, and TRUE otherwise.

For mixture models, a low effective size of one or more theta chains can occur for the following reasons:

A. the model has not yet converged

B. the model is overfit and there is lots of mixing (label swapping )between some of the chains

C. the model is not overfit but there is a lot of mixing of the thetas

For both (A) and (B) it is desirable to return NAs. While (C) can also occur, it can be easily diagnosed by visual inspection of the chains. To the extent that (C) occurs, the correction factor may not be needed.

a list of parameters to be passed to marginalLikelihood.

effectiveSize marginalLikelihood