spadd.presmth.Legr.cv: Choose tuning parameters for fitting the desparsified lasso...

In gregorkb/spaddinf: Fit two-step estimator for inference in sparse high-dimensional additive models

Description

Choose tuning parameters for fitting the desparsified lasso presmoothing estimator with Legendre polynomials

Usage

spadd.presmth.Legr.cv(
  X,
  Y,
  d.pre,
  K = 1,
  n.lambda,
  n.eta,
  n.folds,
  plot = FALSE
)

Arguments

X	the design matrix
Y	the response vector
d.pre	the number of intervals in which to divide the support of each covariate
K	the order of the Legendre polynomials. E.g. K=0 fits piecwise constant, K=1 fits piecewise linear functions.
n.lambda	the number of candidate lambda values
n.eta	the number of candidate eta values
n.folds	the number of crossvalidation folds
plota	logical indicating whether crossvalidation output should be plotted

Value

a list with the chosen values of the tuning parameters

Examples

data <- data_gen(n = 200,q = 50,r = .9)

spadd.presmth.Legr.cv.out <- spadd.presmth.Legr.cv(X = data$X,
                                                   Y = data$Y,
                                                   d.pre = 10,
                                                   n.lambda = 25,
                                                   n.eta = 25,
                                                   n.folds = 5,
                                                   plot = TRUE)

spadd.presmth.Legr.out <- spadd.presmth.Legr(X = data$X,
                                             Y = data$Y,
                                             d.pre = 10,
                                             lambda = spadd.presmth.Legr.cv.out$cv.lambda,
                                             eta = spadd.presmth.Legr.cv.out$cv.eta,
                                             n.foi = 6)

plot_presmth_Legr(x = spadd.presmth.Legr.out,
                  true.functions = list( f = data$f,
                                         X = data$X))

gregorkb/spaddinf documentation built on July 23, 2021, 4:02 a.m.