grid_ridge: Computes grid of predictions for ridge regression

In bgs25/SubsetGridRegression: Subset Grid Regression for Lasso, Ridge and Logistic Lasso Regression Models with Potential Prior Information

Description

Computes prediction errors for a grid (over lambda and variable subsets) of ridge regression models

Usage

grid_ridge(
  x,
  z,
  y,
  yz,
  var_order = NULL,
  lambda = NULL,
  nlambda = 100,
  lambda.min.ratio = 1e-05,
  grid.size = p,
  lambda.mult = 1e+05,
  errors_mean = TRUE
)

Arguments

x	design matrix for training
z	design matrix for testing
y	response for training
yz	response for testing
var_order	For user-specified ordering of variables. Indices start at 0, start with least important variable and end with most. By default order will be induced from scaling of columns in design matrix
lambda	For user-specified sequence of tuning parameter lambda
nlambda	Length of automatically generated sequence of tuning parameters lambda
lambda.min.ratio	Ratio of max/min lambda for automatically generated sequence of tuning parameters lambda
grid.size	Number of subsets of variables for which a solution path will be computed for
lambda.mult	Scales the sequence of lambda by a constant
errors_mean	Controls whether MSPE or SPE (without dividing by sample size) is returned. Used only for within cv_grid_ridge

Value

A list of objects:

• errors – a matrix of errors for the models (grid.size times nlambda)

• lambda – sequence of lambda values used

Examples

set.seed(1)
X = matrix(0, 50, 500)
Z = matrix(0, 10, 500)
betavec = c(rep(1,5),rep(0,495))
X[ , 1:5 ] = matrix(rnorm(250), 50, 5)
Z[ , 1:5 ] = matrix(rnorm(50), 10, 5)
YZ = as.vector(Z %*% betavec)
Y = as.vector(X %*% betavec)
Y = Y + rnorm(50)
X = X + matrix(rnorm(50*500), 50, 500)
mod1 = grid_ridge(X, Z, Y, YZ, grid.size = 50)

bgs25/SubsetGridRegression documentation built on Dec. 19, 2021, 8:50 a.m.