select_smoothing: Optimal Smoothing Parameter for ssMRCD based on Residuals

In ssMRCD: Spatially Smoothed MRCD Estimator

Optimal Smoothing Parameter for ssMRCD based on Residuals

Description

The optimal smoothing value for the ssMRCD estimator is based on the residuals and the trimmed mean of the norm.

Usage

select_smoothing(
  X,
  groups,
  weights,
  lambda = seq(0, 1, 0.1),
  TM = NULL,
  alpha = 0.75,
  seed = 123436,
  return_all = TRUE,
  cores = 1
)

Arguments

X	data matrix containing observations.
groups	grouping vector corresponding to X.
weights	weight matrix for groups, see rescale_weights, and geo_weights.
lambda	vector of parameter values for smoothing, between 0 and 1.
TM	target matrix, if not given MCD (or MRCD if non regular) is used with default values and alpha.
alpha	percentage of outliers to be expected.
seed	seed for ssMRCD calculations.
return_all	logical, if FALSE the function returns only the optimal lambda.
cores	integer, number of cores used for parallel computing.

Value

lambda_opt	optimal lambda for smoothing.
COVS	ssMRCD object with optimal parameter setting.
plot	plot for optimal parameter setting.
residuals	mean of norm of residuals for varying lambda.

Examples

# create data set
x1 = matrix(runif(200), ncol = 2)
x2 = matrix(rnorm(200), ncol = 2)

# create weighting matrix
W = matrix(c(0, 1, 1, 0), ncol = 2)

select_smoothing (X = rbind(x1, x2),
                 groups = rep(c(1,2), each = 100),
                 weights = W,
                 lambda = seq(0, 1, 0.1),
                 return_all = TRUE,
                 cores = 1)

ssMRCD documentation built on Sept. 11, 2024, 5:14 p.m.