pargee: Parallel GEE solver
In Bioconductor/geeni: Parallel implmentation of bounded memory GEE

Description Usage Arguments Details Value Author(s) Examples

A parallelized, out-of-memory based GEE solver

1
2
3

  ## S4 method for signature 'gdManager'
pargee(gd, family, binit, maxit=20, 
            tol=1e-06, sandwich=TRUE, ...)

`gd`	A `gdManager-class`.
`family`	A GLM stats `family` instance, e.g., `gaussian`. See ?`family` for details.
`binit`	Initial values for the regression parameter vector.
`maxit`	Maximum number of iterations allowed.
`tol`	Convergence criterion. The absolute value of the incremental change in parameter estimates is measured in each iteration step. Iterations will stop when the largest of these changes is smaller than `tol`.
`sandwich`	Logical indicating if the robust variance should be computed.
`...`	Arguments passed to other methods.

Preliminary demonstration, mildly validated on orthodontic data for gaussian case and respiratory data for binomal case.

A list with two named elements: 'coefficients' and 'robust.variance'.

Vince Carey <stvjc@channing.harvard.edu>

## Create a 'groupedData' object from 'Orthondont' data set:
library(nlme)
  orthGD <-
       groupedData(distance ~ age | Subject,
                   data = as.data.frame(Orthodont),
                   FUN = mean,
                   outer = ~ Sex,
                   labels = list(x = "Age",
                     y = "Distance from pituitary to pterygomaxillary fissure"),
                   units = list( x = "(yr)", y = "(mm)"))
dim(orthGD)

## Create a 'gdManager' that points to the ff files in tempdir():
orthMgr <- groupedData2ff(gd=orthGD, gcol=3, prefix=tempfile())

## In the ff file 'Sex' is coded as `1` for male and `2` for female. 
## We adjust the coding to `0` and `1`.
numdat(orthMgr)[,"Sex"] <- numdat(orthMgr)[,"Sex"] - 1
 
## Call the GEE solver:
res <- pargee(orthMgr, gaussian, c(0, 0, 0))

## Coefficients and standard errors:
round(res$coefficients, 4)
round(sqrt(diag(res$robust.variance)), 4)

## Compare results with those from geeglm() from the geepack package:
library(geepack)
geeglm <- geeglm(distance ~ age + Sex, data=orthGD, id=Subject)
round(summary(geeglm)$coefficients, 4)