EMLasso.1l: Fit EMLasso for 1 lambda
In EMLasso: Fit (logistic) LASSO when there is missing data in the predictors

Description Usage Arguments Value Author(s) References Examples

Fit EMLasso for 1 lambda

  EMLasso.1l(ds, out, lambda, nrOfSamplesPerMDRow = 10,
    rowsToUseForFit = seq(nrow(ds)),
    firstTimeCompleter = marginalCompleter(),
    imputeDs2FitDsProperties = normalImputationConversion(),
    fitPredictor = GLoMo, family = "binomial",
    convergenceChecker = convergenceCheckCreator(),
    postProcess = postProcessEMLasso1l, verbosity = 0,
    extraLog = function(...) { })

`ds`	dataset (`numdfr` or `data.frame`) to fit it to
`out`	outcome vector
`lambda`	the single lambda
`nrOfSamplesPerMDRow`	For rows with missing data, how many rows to sample. Defaults to 10.
`rowsToUseForFit`	Which of the rows of dfr/out can be used for fitting the LASSO (int vector). Defaults to all rows.
`firstTimeCompleter`	object supported by `completeMarginal` or of similar form that will complete the dataset the first time. Defaults to `marginalCompleter` with default parameters.
`imputeDs2FitDsProperties`	see `imputeDs2FitDs` object that will provide the conversion from imputed dataset to one that is ready for fitting the predictor model
`fitPredictor`	function that will perform the predictor model fit. Defaults to `GLoMo` and should be of this signature.
`family`	see `glmnet`. Defaults to "binomial" (i.e. lasso penalized logistic regression).
`convergenceChecker`	function that will check based on the coefficients over the repeats whether convergence has been achieved. Defaults to `convergenceCheckCreator()`, so a wrapper around `checkConvergence.glmnet`. The function should have parameters `coefs` and `verbosity=0`
`postProcess`	function, like `postProcessEMLasso1l` (its default) and of that signature, to do some extra work before returning the result of `EMLasso.1l`
`verbosity`	The higher this value, the more levels of progress and debug information is displayed (note: in R for Windows, turn off buffered output)
`extraLog`	A function (by default one that does nothing) that will be used in each iteration to perform extra logging

An object of class EMLasso1l. This is a list with the following items:

lasso.fit: glmnet object
predictorModel: final predictor fit (typically of class GLoMo)
coefs: coefs for all iterations (rows) and dummycoded columnames (columns), incl. (intercept).
ds: original dataset passed along
out: outcome variable (1 for each row in ds)
lambda: 1 lambda value
nrOfSamplesPerMDRow: how many imputations per row with missing data
convergence: return value of last call to convergenceChecker. Will typically contain information like minIt and maxIt.
rowsToUseForFit: which of the rows in ds was used to fit the lasso
iterCount: how many iterations occurred before convergence / maxIt
imputeDs2FitDsProperties: as passed in, but first fed to imputeDs2FitDsProps
any other items added by postProcess

Nick Sabbe nick.sabbe@ugent.be

[PENDING]

aDfr<-generateTypicalIndependentDfr(numCat=10, numCnt=10, numObs=100, catProbs=rep(1/3,3),
rcnt=typicalRandomNorm, doShuffle=TRUE, verbosity=1)

outlins<- -mean(aDfr$cnt1)+aDfr$cnt1+2*(aDfr$cat1=="b")
outprobs<-expit(outlins)
y<-factor(sapply(outprobs, function(prob){sample(c("no", "yes"), 1, prob=c(1-prob,prob))}))

aDfr.MD<-randomNA(aDfr, 0.01, verbosity=1)
rlh<-findReasonableLambdaHelper(aDfr, y, verbosity=10)
aLam<-rlh$regionDfr[2,"lambda"]
emlfit1<-EMLasso.1l(aDfr.MD, y, lambda=aLam, nrOfSamplesPerMDRow=7,
	convergenceChecker=convergenceCheckCreator(minIt=5, maxIt=10), verbosity=10)