findReasonableLambdaHelper: Function to run on a dataset with not too much missing data...

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

Description

Expects a singly imputed dataset and fits a logistic LASSO so the user can pick a set that wil probably be interesting.

Usage

  findReasonableLambdaHelper(ds, out, family = "binomial",
    showFirst = 20, showPlot = TRUE, type.measure = "auc",
    repsNeededForFirstOccurrence = 3,
    weights = rep(1, nrow(ds)), ..., verbosity = 0,
    minNumHigher = 20, minNumLower = 20, maxNumLower = 30,
    imputeDs2FitDsProperties = normalImputationConversion(),
    standardize = FALSE, nfolds = 10)

  ## S3 method for class 'LambdaHelper'
object[i, j, drop = TRUE]

  getLambdas(x, ...)

  ## S3 method for class 'lambdaregion'
getLambdas(x, ...)

  ## S3 method for class 'LambdaHelper'
getLambdas(x, ...)

Arguments

ds	dataset to investigate
out	outcome vector
family	see glmnet. Defaults to "binomial" (i.e. lasso penalized logistic regression).
showFirst	show the top coefficients (first showFirst occurring)
showPlot	if TRUE (the default), visually supports the decision
type.measure	see cv.glmnet
repsNeededForFirstOccurrence	How many times (i.e. for how many lambda values) must a coefficient be consecutively nonzero before we count it as "occurring"
weights	vector with weight to be assigned to each row of ds
...	passed on to plotex (if relevant)
verbosity	The higher this value, the more levels of progress and debug information is displayed (note: in R for Windows, turn off buffered output)
minNumHigher	How many lambdas higher than the optimum do you minimally want (if available)
minNumLower	How many lambdas lower than the optimum do you minimally want (if available)
maxNumLower	How many lambdas lower than the optimum do you maximally want
imputeDs2FitDsProperties	see imputeDs2FitDs and EMLasso
standardize	see glmnet. Defaults to FALSE.
nfolds	see glmnet. Defaults to 10.
object	LambdaHelper
i	row index
j	column index. If this is missing, the ith lambda is returned
drop	if TRUE the result is coerced to the simplest structure possible
x	object to find 'interesting' set of lambdas for

Value

list of class "LambdaHelper":

topres	data.frame with showFirst rows, and columns: variable (name), lambda,critl (lower bound of criterion), crit (estimate of criterion), critu (upper bound of criterion), critsd (sd of criterion), index (at which lambda index does this variable first occur)
allLambda	vector of lambda values
regionDfr	data.frame w 3 rows 3 columns: name (values: "lower lambda", "optimum", and "higher lambda"), idx and lambda
regionOfInterestData	see getMinMaxPosLikeGlmnet

depends on the parameters

vector of lambda values, normally high to low

Note

EMLasso is pretty heavy and has to be run per lambda. This functions helps preselect some lambda values, and can typically avoid useless calculations for non-interesting lambda values.

Author(s)

Nick Sabbe [e-mail]

Examples

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))}))

rlh<-findReasonableLambdaHelper(aDfr, y, verbosity=10)
data(emlcvfit, package="EMLasso")
rlh<-findReasonableLambdaHelper(aDfr, y, verbosity=10)
rlh[1]
rlh[1:5, NULL]
data(emlcvfit, package="EMLasso")
rlh<-findReasonableLambdaHelper(aDfr, y, verbosity=10)
getLambdas(rlh$regionOfInterestData)
data(emlcvfit, package="EMLasso")
rlh<-findReasonableLambdaHelper(aDfr, y, verbosity=10)
getLambdas(rlh)

EMLasso documentation built on May 2, 2019, 5:49 p.m.