crossValidate: Crossvalidate a model
In EMLasso: Fit (logistic) LASSO when there is missing data in the predictors

Description Usage Arguments Value Note Author(s) See Also Examples

Crossvalidate a model

  crossValidate(model, ..., verbosity = 0)

  ## S3 method for class 'EMLassoGLoMo'
crossValidate(model,
    ds=model$result[[1]]$ds, out=model$result[[1]]$out,
    wts=rep(1, nrow(ds)),
    imputeDs2FitDsProperties=model$imputeDs2FitDsProperties,
    imputations=10, ..., type.measure="auc",
    keepResultPerLambda=FALSE, nobs=1, unpenalized=FALSE,
    verbosity=0)

  repeatedlyPredictOut(glomo, ds, out, varsets, reps = 10,
    nfolds = 10,
    imputeDs2FitDsProperties = normalImputationConversion(),
    returnGroups = FALSE, returnCoefs = FALSE, ...,
    reusabledata, noScaling = TRUE, verbosity = 0)

  repeatedPredictedProbAUC(reppredprob, out, verbosity = 0,
    groups, onlyresult = TRUE, glmnetlike = TRUE)

  cv.MI.logreg(glomo, ds, out, useVarNames, reps,
    imputeDs2FitDsProperties = normalImputationConversion(),
    lambda, useAsGlmnetFit, ..., verbosity = 0)

`model`	model fit
`...`	for flexibility in 'derived' implementation (passed on to `collectImputationModels` in `crossValidate.EMLassoGLoMo`)
`verbosity`	The higher this value, the more levels of progress and debug information is displayed (note: in R for Windows, turn off buffered output)
`ds`	dataset with predictors
`out`	vector (binary factor) of outcomes
`wts`	vector of weights (defaults to equal weights for all rows)
`imputeDs2FitDsProperties`	see `imputeDs2FitDs` object that will provide the conversion from imputed dataset to one that is ready for fitting the predictor model
`imputations`	Number of multiple imputations on the complete dataset (defaults to 10)
`type.measure`	see `cv.glmnet`
`keepResultPerLambda`	if `TRUE` (not the default), the individual results from the `crossValidate.EMLasso1l` are also returned in an extra item `resultPerLambda`
`nobs`	how many observations are simulated for each row with missing data
`unpenalized`	if `TRUE` (not the default) a simple regression model is fit with the selected variables
`glomo`	GLoMo model to predict from
`varsets`	list of character vectors holding the variables (names) to be checked
`reps`	number of predictions
`nfolds`	number of folds for crossvalidation
`returnGroups`	if `TRUE`, a list is returned with the normal result as its `result` item and a matrix holding the group assignment per repetition as the `groups` item.
`returnCoefs`	if `TRUE`, a list is returned with the normal result as its `result` item and a list of matrices holding the coefficient values per repetition and per fold as the `coefs` item.
`reusabledata`	optional premade result of `reusableDataForGLoMoSampling`
`noScaling`	if `TRUE` (the default) the data is not scaled upon conversion
`reppredprob`	one of the matrices as return by `repeatedlyPredictOut`
`groups`	vector/matrix of fold membership assignment. If nor present, 10 random groups are created
`onlyresult`	if `FALSE`, the return value holds (as list members) the the overall AUC and sd, but also the results per repetition
`glmnetlike`	if `TRUE`, `calcAUC.glmnet` is used to calculate AUC and sd, otherwise `calcAUC.Binary` is used
`useVarNames`	names of columns to include in the model (character vector)
`lambda`	value to use as lambda in the return value (note: ignored for the rest)
`useAsGlmnetFit`	object that can be used for the `glmnet.fit` item in the return value

object that has as class: "cv." pasted before the class of model. Normally, model will will be the return value of EMLasso, so this result is mainly the same as a cv.glmnet. The added/altered items are:

`glmnet.fit`	is now the `model` passed in, so has more classes besides "glmnet" (e.g. "EMLasso")
`resultPerLambda`	matrix with one column per imputation. The top rows are the estimates for the criterion per lambda, below that are their SD estimates. Not present if `keepResultPerLambda=FALSE`

List of the same length as varsets (unless it was length 1, then the first object is simply returned). Each item is a matrix with one row for each row in ds and one column per reps, and holds the predicted probability in a crossvalidation.

named vector of length 2, holding the "AUC" and the "AUCSD"

object of classes "cv.MI.logreg", "cv.glmnet" and "cv.lognet". Has exactly the items of a cv.glmnet object

aids to generalize crossvalidation

Nick Sabbe nick.sabbe@ugent.be

EMLasso, cv.glmnet

y<-rbinom(nrow(iris), 1, 0.5)
iris.cpy<-randomNA(iris, n=0.1)
iris.emlognet<-EMLasso(ds=numdfr(iris.cpy), out=y,
	lambdas=c(0.03,0.002,0.0003), nrOfSamplesPerMDRow=7, verbosity=2,
	convergenceChecker=convergenceCheckCreator(minIt=5, maxIt=10))
sfStop()
iris.cv.emlognet<-crossValidate(iris.emlognet, verbosity=2)