Skip to contents

Fast Ridge Regression

Source: R/lmRidge.R
lmRidge.Rd

Fit a linear model by ridge regression over many values of the regularization parameter

Usage

lmRidge(y, design, X, lambda, weights, scaleX = TRUE)

Arguments

y

response vector

design

unpenalized design matrix, including intercept term

X

matrix of features whose coefficients are penalized

lambda

vector or scalar of regularization parameter values

weights

vector of weights

scaleX

scale the X matrix so all features have the same variance. defaults to TRUE

Value

none

list of regression results

Details

Adapt ridge fitting algorithing using SVD from MASS::lm.ridge(), adding unpenalized design matrix and weights

See also

MASS::lm.ridge()

Examples

y = longley$GNP.deflator
design = model.matrix(~ 1, longley)
X = longley[,-1]

lambda = seq(0, .05, length.out=40)

fit = lmRidge(y, design, X, lambda = lambda)

par(pty="s")
plot(fit)


# get coefs with best GCV
i = which.min(fit$GCV)
fit$coef[,i,drop=FALSE]
#>              0.005128205
#> GNP           17.6910338
#> Unemployed     1.8533830
#> Armed.Forces   0.4525287
#> Population    -9.2444737
#> Year           0.6904862
#> Employed      -0.1858366

fit$coef_design[,i,drop=FALSE]
#>             0.005128205
#> (Intercept)    101.6812

# predict
predict(fit, 
  newdata = list(design = design, X = X), 
  lambda = fit$lambda[i])
#>      0.005128205
#> 1947    83.70544
#> 1948    86.87336
#> 1949    88.14397
#> 1950    90.87599
#> 1951    95.90758
#> 1952    97.69792
#> 1953    98.49161
#> 1954   100.11293
#> 1955   103.26701
#> 1956   105.16366
#> 1957   107.47808
#> 1958   109.44567
#> 1959   112.71412
#> 1960   113.94012
#> 1961   115.38703
#> 1962   117.69551