fitSEMMS: Run the Generalized Alternating Maximization algorithm to fit...

In haimbar/SEMMS: Variable selection in Generalized Linear Models - an empirical Bayes approach

Description

Coefficients of predictors in a GLM are modeled as a three-component normal mixture, where the majority of predictors are assumed to belong to the null component (i.e., they have no effect on the mean of the response) and the others can have a positive, or negative effect. fitSEMMS runs a Generalized Alternating Maximization algorithm to fit the model.

Usage

fitSEMMS(
  dat,
  mincor = 0.7,
  nn = 5,
  nnset = NULL,
  distribution,
  rnd = F,
  BHthr = 0.01,
  initWithEdgeFinder = T,
  minchange = 1,
  maxst = 20,
  ptf = F,
  verbose = FALSE
)

Arguments

dat	The dataset, as generated by readInputFile().
mincor	The minimum correlation coefficient between pairs of putative variable, over which they are considered highly correlated. Default is 0.75.
nn	The initial value for the number of non-null variables. Default is 5.
nnset	Optional: instead of an initial number of candidates, can specify the column numbers in the Z matrix for the first iteration. Default is null.
distribution	The distribution of the response (N, P, or B).
rnd	Whether to run the greedy (F, default) algorithm, or the randomized.
BHthr	The Benjamini-Hochberg threshold to be used when determining the initial set of non-null variables. Default=0.01.
initWithEdgeFinder	Determines whether to use the edgefinder package to find highly correlated pairs of predictors (default=TRUE).
minchange	The minimum change in the log-likelihood that is to be considered meaningful. Default=1.
maxst	The maximum number of iterations of the algorithm. Default=20.
ptf	Whether to print output at each iteration to a file called SEMMS.log. Default=FALSE.
verbose	Whether to show progress message to the user. Default=FALSE.

Value

inits	The initial values obtained from initVals().
initNN	The column numbers of the putative variables to be included in the model in the first iteration.
gam.out	The output from the GAMupdate() function, which is a list containing nn=the selected variables, mu,beta,s2r,s2e=the parameter estimates of the mixture model, gam_nn=the sign (+/-1) of the effect of the non-null variables, pp=a table of posterior probabilities, and lockedOut=any variables found to be highly correlated with selected variables.
distribution	The distribution selected by the user.
mincor	The user's input for mincor above. Ignored if initWithEdgeFinder is set to TRUE

Examples

## Not run: 
fn <- system.file("extdata", "AR1SIM.RData", package = "SEMMS", mustWork = TRUE)
dataYXZ <- readInputFile(fn, ycol=1, Zcols=2:100)
fittedSEMMS <- fitSEMMS(dataYXZ, mincor=0.8, nn=15, minchange= 1,
                        distribution="N",verbose=T,rnd=F)
## End(Not run)

haimbar/SEMMS documentation built on Dec. 20, 2021, 2:44 p.m.