Here in https://github.com/cran/ridge/blob/master/R/linearRidge.R#L96,

nPCs <- which.max(propVar[propVar < 90]) + 1

what if all the elements in propVar are greater than 90? It will throw an error, which is the title of this issue.

Hello,
I am using logisticRidge function to ask how my binary variable (0/1) is related to a set of linear predictors. I am writing to ask for clarification on the ‘predict’ portion of the model.

After fitting a logisticRidge model, I am trying to use the predict function to get fitted probabilities of outcome. However, I appear to be getting different values when I use ‘predict’ and ‘predict’ but supplying the original data that I fitted the model to as new data:

data(mtcars)
library(ridge)
model<-logisticRidge(vs~.,data=mtcars)
predict(model,type="response")
predict(model,type="response",newdata=mtcars[,-8])

Which usage of predict provides accurate values?

I have tested this out on a several datasets and with different combinations of binary predictors and continuous predictors. They seem to produce the same mismatches in predict values. I am happy to provide more replicable examples if you would like me to.

Thank you for your time!

Dear Steffen,

I believe there are some errors in the expression employed to compute the standard errors of the ridge linear coefficients (ridge:::vcov.ridgeLinear). Specifically, it always penalizes the intercept (even when argument scaling = "none" on the ridge::linearRidge call) and it uses an incomplete formula. The proper formula for the Ridge standard errors is:

Where Gamma is the unitary matrix of dimensions (m+1)×(m+1) but where element i=1, j=1, is set to zero to avoid penalizing the intercept (only in the case of scaling = "none"). And it is not as programmed in ridge:::vcov.ridgeLinear:

You may find references for this formula below:
https://www.statlect.com/fundamentals-of-statistics/ridge-regression
https://ncss-wpengine.netdna-ssl.com/wp-content/themes/ncss/pdf/Procedures/NCSS/Ridge_Regression.pdf
https://stats.stackexchange.com/questions/2121/how-can-i-estimate-coefficient-standard-errors-when-using-ridge-regression

See also below a script reproducing what I believe the proper Ridge standard errors are:

# Load libraries
library(MASS)
library(ridge)

# Generate X, y
set.seed(1)
n = 100
m = 10
mu = rnorm(m) # Generate mean vector
Sigma = rWishart(n = 1, df = m, Sigma = diag(m))[,,1] # Generate Covariance matrix
X = mvrnorm(n = n, mu = mu, Sigma = Sigma)
X_ = cbind(1, X) # X with an intercept column
THETA = rnorm(m+1)
y = X_ %*% THETA + rnorm(n, sd = 3)

# Using unscaled ridge Regression
lambda = 100
ridge_model = linearRidge(y ~ X, lambda = lambda, scaling = "none")
I = diag(m+1); I[1,1] = 0

# Linear coefficients
solve(t(X_) %*% X_ + lambda*I) %*% t(X_) %*% y
(theta_ridge = coef(ridge_model)) # Ridge coefficients coincides

# Variance of residuals
(var_epsilon = sum((y - X_ %*% theta_ridge)^2)/(n-(m+1)))

# Standard errors
var_epsilon * solve(t(X_) %*% X_ + lambda*I) %*% t(X_) %*% X_ %*% solve(t(X_) %*% X_ + lambda*I)

# However, ridge:::vcov.ridgeLinear computes:
ridge:::vcov.ridgeLinear(ridge_model)
var_epsilon * solve(t(X_) %*% X_ + lambda*diag(m+1)) # WRONG FORMULA

I am aware that due to the lack of scale invariance of the ridge estimator, that center-scaling is recommended (and hence dropping the intercept column). Also, the degrees of freedom in ridge regression are not really n-(m+1) (some authors speak of effective degrees of freedom as n minus the trace of the hat matrix).

FYI, I am using ridge_3.0 on R version 4.1.0 (2021-05-18), platform: x86_64-w64-mingw32/x64 (64-bit), running under: Windows >= 8 x64 (build 9200).

Please let me know if I have overseen anything. Have a nice day,

Best,
Ben.

1. I try to use ‘conda install -c r r-ridge’， but it keep Solving environment without stopped
2. I try to install ridge with https://cran.r-project.org/src/contrib/ridge_3.2.tar.gz，i apt install gcc and GSL， but it still can not install.
   information is as follows:

install.packages("/home/shufan/software/oncoPredict/ridge_3.2.tar.gz",repos = NULL,type = "source")

• installing source package ‘ridge’ ...
  ** package ‘ridge’ successfully unpacked and MD5 sums checked
  ** using staged installation
  /home/xxx/anaconda3/envs/drug/lib/R/bin/config: 1: eval: make: not found
  /home/xxx/anaconda3/envs/drug/lib/R/bin/config: 1: eval: make: not found
  /home/xxx/anaconda3/envs/drug/lib/R/bin/config: 1: eval: make: not found
  /home/xxx/anaconda3/envs/drug/lib/R/bin/config: 1: eval: make: not found
  /home/xxx/anaconda3/envs/drug/lib/R/bin/config: 1: eval: make: not found

checking for gcc... gcc
checking whether the C compiler works... yes
checking for C compiler default output file name... a.out
checking for suffix of executables...
checking whether we are cross compiling... no
checking for suffix of object files... o
checking whether the compiler supports GNU C... yes
checking whether gcc accepts -g... yes
checking for gcc option to enable C11 features... none needed
checking for gsl-config... /usr/bin/gsl-config
checking for stdio.h... yes
checking for stdlib.h... yes
checking for string.h... yes
checking for inttypes.h... yes
checking for stdint.h... yes
checking for strings.h... yes
checking for sys/stat.h... yes
checking for sys/types.h... yes
checking for unistd.h... yes
checking for gsl/gsl_version.h... yes
configure: creating ./config.status
config.status: creating R/linearRidgeGenotypes.R
config.status: creating R/linearRidgeGenotypesPredict.R
config.status: creating R/logisticRidgeGenotypes.R
config.status: creating R/logisticRidgeGenotypesPredict.R
config.status: creating src/Makevars
config.status: creating src/config.h
config.status: src/config.h is unchanged
** libs
sh: 1: make: not found
Warning in system(cmd) : 调用命令时发生了错误
ERROR: compilation failed for package ‘ridge’

Test case:

library(testthat)

tol <- 0.0001

test_that("test linearRidge with formula variable that goes out of scope", {
  the_formula <- formula('mpg ~ wt + cyl')
  model1 <- lm(the_formula, data = mtcars)
  model2 <- linearRidge(the_formula, data = mtcars, lambda = 0)
  # suppose the_formula goes away:
  rm(the_formula)
  
  expect_equal(coef(model1), coef(model2), tolerance = tol, label = "coefficients")
  expect_equal(predict(model1), predict(model2), tolerance = tol, label = "predictions")
})

This is separate from my changes, and also fails on 2.4.

Here is an example:

library(ridge)

foo <- mtcars
foo$cyl_factor <- as.factor(paste0("cyl", foo$cyl))
model1 <- lm(mpg ~ wt + cyl_factor, data = foo)
model2 <- linearRidge(mpg ~ wt + cyl_factor, data = foo, lambda = 0.0)

predict(model1)
# has output..
predict(model2)
Error in as.matrix(mm) %*% beta : 
  requires numeric/complex matrix/vector arguments

It's because cyl_factor breaks out into multiple columns, while mm has only one column for that factor. In predict.linearRidge, right before res <- drop(as.matrix(mm) %*% beta:

Browse[2]> names(mm)
[1] "1"          "wt"         "cyl_factor"
Browse[2]> names(beta)
[1] "(Intercept)"    "wt"             "cyl_factorcyl6" "cyl_factorcyl8"

Since the last commit (548f48d), on line 47 in linearRidge.R there is a misplaced print(m). This is probably a leftover from a debug session that causes unwanted clutter in the output when using ridge.

Hello Steffen, How are you doing?

Thanks for keeping this package.

I'm thinking of using your package for a set of panel data, 30 years, 38 c_id.

Your package can be used for such setting?
Thank you very much

Hi, I kept getting error when installing ridge in the latest version of R4.0.2 in MacOS (fine in Windows system).
Below is the error information:

Any solutions?
Many thanks advanced!

Hello,

I am having significant issues installing the ridge toolbox (see below error). I have used it before, so not sure what is going on. I have installed GSL. For reference, I am working on a MacOS (Catalina, 10.15.4).

Thank you,
Cristina

install.packages("ridge")
Installing package into ‘/Users/cristinaroman/Library/R/4.0/library’
(as ‘lib’ is unspecified)
Package which is only available in source form, and may need compilation of C/C++/Fortran: ‘ridge’
Do you want to attempt to install these from sources? (Yes/no/cancel) yes
installing the source package ‘ridge’

trying URL 'https://cloud.r-project.org/src/contrib/ridge_2.5.tar.gz'
Content type 'application/x-gzip' length 124719 bytes (121 KB)

downloaded 121 KB

• installing source package ‘ridge’ ...
  ** package ‘ridge’ successfully unpacked and MD5 sums checked
  ** using staged installation
  checking for gsl-config... no
  configure: WARNING: gsl-config not found, is GSL installed?
  configure: WARNING: ridge will be installed but some functions will be unavailable
  configure: creating ./config.status
  config.status: creating R/linearRidgeGenotypes.R
  config.status: creating R/linearRidgeGenotypesPredict.R
  config.status: creating R/logisticRidgeGenotypes.R
  config.status: creating R/logisticRidgeGenotypesPredict.R
  config.status: creating src/config.h
  ** libs
  xcrun: error: invalid active developer path (/Library/Developer/CommandLineTools), missing xcrun at: /Library/Developer/CommandLineTools/usr/bin/xcrun
  ERROR: compilation failed for package ‘ridge’
• removing ‘/Users/cristinaroman/Library/R/4.0/library/ridge’

The downloaded source packages are in
‘/private/var/folders/p4/1n_2sb096sg87thjdrz9whwm0000gn/T/RtmpTeLO5O/downloaded_packages’
Warning message:
In install.packages("ridge") :
installation of package ‘ridge’ had non-zero exit status