NonpModelCheck (version 1.0)

group.npvarselec: Group variable selection for nonparametric regression

Description

Performs group variable selection in a completely nonparametric regression model using hypothesis testing for high-dimensional one-way ANOVA and False Discovery Rate (FDR) corrections.

Usage

group.npvarselec(X, Y, groups, method = "backward", p = 7, fitSPC = TRUE,
    degree.pol = 0, kernel.type = "epanech", bandwidth = 0, gridsize = 10, 
    dim.red = c(1, 10))

Arguments

X

matrix with observations, rows corresponding to data points and columns correspond to covariates.

Y

vector of observed responses.

groups

a variable of type "list" containing, in each item, a vector of indices of the covariates in each group.

method

type of algorithm to run variable selection, options are "backward", "forward" and "forward2".

p

size of the window Wi. See npmodelcheck for details.

fitSPC

a logical indicating whether to use the first supervised principal component (SPC) of each group in the local polynomial fitting. See Details.

degree.pol

degree of the polynomial to be used in the local fit.

kernel.type

kernel type, options are "box", "trun.normal", "gaussian", "epanech", "biweight", "triweight" and "triangular". "trun.normal" is a gaussian kernel truncated between -3 and 3.

bandwidth

bandwidth for the local polynomial fit at each step of the elimination (or selection). Options are: 0 for leave-one-out cross validation with criterion of minimum MSE to select a unique bandwidth that will be used for all dimensions; -1 for Generalized Cr

gridsize

number of possible bandwidths to be searched in cross-validation. Default is set to 10. If cross-validation is not performed, it is ignored.

dim.red

vector with first element indicating 1 for Sliced Inverse Regression (SIR) and 2 for Supervised Principal Components (SPC); the second element of the vector should be number of slices (if SIR), or number of principal components (if SPC). If 0, no dimensio

Value

• selected
  groups selected
• p_values
  p-values of the tests of the selected groups

Details

The selection procedure is based on the nonparametric test npmodelcheck, which for testing the significance of group i, uses the residuals of the local polynomial regression of all the other covariates that are not in that group. When "fitSPC" is TRUE, the residuals for each test are computed based on the estimated regression curve m(S_{(-i)}), where S has d columns, each containing the first SPC of the corresponding group, and S_{(-i)} is the matrix S without column i.

Backward elimination is done by removing, at each step, the least significant group in the model if its p-value, obtained from the test npmodelcheck, is not significant according to False Discovery Rate (FDR) corrections (Benjamini and Yekutieli, 2001). The final model contains only groups that have significant p-values (based on FDR).

Forward selection is done by adding to the model, at each step, the group with the smallest p-value (when tested with all covariates that are already in the model), if when added, every group in the model is significant according to FDR corrections.

Forward2 selection is as follows: at each step, denote by Z = (Z_1, ..., Z_q) the groups in the model and by W = (W_1, ..., W_r) the groups not in the model (note that (Z,W) = X). Let p_j, j = 1,...r, be the maximum of the set of q+1 p-values obtained from testing each group (Z1,...,Z_q,W_j). Add to the model the group corresponding to the smallest p_j as long as, when added, all the p-values of the groups in the model are significant according to FDR corrections.

See also details of npmodelcheck and localpoly.reg.

References

Zambom, A. Z. and Akritas, M. G. (2012). a) Nonparametric Model Checking and Variable Selection. arXiv 1205.6761.

Zambom, A. Z. and Akritas, M. G. (2012). b) Signicance Testing and Group Variable Selection. arXiv 1205.6843.

Benjamini, Y. and Yekutieli, D. (2001) The control of false discovery rate in multiple testing under dependency. Annals of Statistics, 29, 1165-1188.

See Also

npmodelcheck, localpoly.reg, npvarselec

Examples

groups = vector("list",7)

groups[[1]] = c(3,8,5,12,14)
groups[[2]] = c(6,7,9,10)
groups[[3]] = 13
groups[[4]] = c(1,2,4,11)
groups[[5]] = c(15,16, 20)
groups[[6]] = 17
groups[[7]] = c(18,19)

X = matrix(1,100,20)
for (i in 1:20)
X[,i] = rnorm(100)

Y = X[,13]^3 + X[,7] + X[,15]^2 + X[,16] + rnorm(100)

group.npvarselec(X,Y,groups)