cv.pengls: Peform cross-validation pengls

In sthawinke/pengls: Fit Penalised Generalised Least Squares models

Peform cross-validation pengls

Description

Peform cross-validation pengls

Usage

cv.pengls(
  data,
  glsSt,
  xNames,
  outVar,
  corMat,
  nfolds,
  foldid,
  scale = FALSE,
  center = FALSE,
  cvType = "blocked",
  lambdas,
  transFun = "identity",
  exclude = NULL,
  transFunArgs = list(),
  loss = c("R2", "MSE"),
  verbose = FALSE,
  ...
)

Arguments

data	A data matrix or data frame
glsSt	a covariance structure, as supplied to nlme::gls as "correlation"
xNames	names of the regressors in data
outVar	name of the outcome variable in data
corMat	a starting value for the correlation matrix. Taken to be a diagonal matrix if missing
nfolds	an integer, the number of folds used in cv.glmnet to find lambda
foldid	An optional vector defining the fold
scale, center	booleans, should regressors be scaled to zero mean and variance 1? Defaults to TRUE
cvType	A character vector defining the type of cross-validation. Either "random" or "blocked", ignored if foldid is provided
lambdas	an optional lambda sequence
transFun	a transformation function to apply to predictions and outcome in the cross-validation
exclude	indices of predictors to be excluded from intercept + xNames
transFunArgs	Additional arguments passed onto transFun
loss	a character vector, currently either 'R2' or 'MSE' indicating the loss function (although R2 is not a proper loss...)
verbose	a boolean, should output be printed?
...	passed onto glmnet::glmnet

Value

A list with components

lambda	The series of lambdas
cvm	The vector of mean R2's
cvsd	The standard error of R2 at the maximum
cvOpt	The R2 according to the 1 standard error rule
coefs	The matrix of coefficients for every lambda value
bestFit	The best fitting pengls model according to the 1 standard error rule
lambda.min	Lambda value with maximal R2
lambda.1se	Smallest lambda value within 1 standard error from the maximum
foldid	The folds
glsSt	The nlme correlation object
loss	The loss function used

Examples

library(nlme)
library(BiocParallel)
n <- 20 #Sample size
p <- 50 #Number of features
g <- 10 #Size of the grid
#Generate grid
Grid <- expand.grid("x" = seq_len(g), "y" = seq_len(g))
# Sample points from grid without replacement
GridSample <- Grid[sample(nrow(Grid), n, replace = FALSE),]
#Generate outcome and regressors
b <- matrix(rnorm(p*n), n , p)
a <- rnorm(n, mean = b %*% rbinom(p, size = 1, p = 0.2)) #20% signal
#Compile to a matrix
df <- data.frame("a" = a, "b" = b, GridSample)
# Define the correlation structure (see ?nlme::gls), with initial nugget 0.5 and range 5
corStruct = corGaus(form = ~ x + y, nugget = TRUE,
value = c("range" = 5, "nugget" = 0.5))
#Fit the pengls model, for simplicity for a simple lambda
register(MulticoreParam(3)) #Prepare multithereading
penglsFitCV = cv.pengls(data = df, outVar = "a", xNames = grep(names(df),
pattern = "b", value = TRUE),
glsSt = corStruct, nfolds = 5)
penglsFitCV$lambda.1se #Lambda for 1 standard error rule
penglsFitCV$cvOpt #Corresponding R2
coef(penglsFitCV)
penglsFitCV$foldid #The folds used
#With MSE as loss function
penglsFitCVmse = cv.pengls(data = df, outVar = "a",
xNames = grep(names(df), pattern = "b", value =TRUE),
glsSt = corStruct, nfolds = 5, loss = "MSE")
penglsFitCVmse$lambda.1se #Lambda for 1 standard error rule
penglsFitCVmse$cvOpt #Corresponding MSE
coef(penglsFitCVmse)
predict(penglsFitCVmse)

sthawinke/pengls documentation built on July 2, 2023, 7:27 a.m.