CVforRidge_Starter (1).R

# Create a toy data example
p = 200
n = 50

# Generate matrix of covariates
library(mnormt)
set.seed(23945)
Sigma = matrix(0.7, p, p) + diag(rep(0.3, p)) # covariance matrix for covariates
X = rmnorm(n, varcov = Sigma)

# Generate response
beta0 = 2 # intercept
sigma = 0.7 # noise standard deviation
beta = runif(p, min = -2, max = 2)
Y = beta0 + X %*% beta + rnorm(n, sd = sigma)


# Split the data into K = 2 folds
# [ToDo] Create idfold vector of length n indicating the fold id, should have 1 or 2 in each position
idfold = sample(1:n) %% 4 + 1

# Correct approaches that are simple
idfold = sample(rep(1:4, length.out = n), size = n)

# For loop over folds and lambdas
##########################################
lambda_seq = seq(0.01, 10, length.out = 20)
nlambda = length(lambda_seq)
cvm = rep(NA, nlambda) # want to have CV(lambda)
cvse = rep(NA, nlambda) # want to have SE_CV(lambda)
# Feel free to store additional things
cv_folds = matrix(NA, K, nlambda) # fold-specific errors

K = 7
idfold= sample(rep(1:K, length.out = n), size = n)
for (fold in 1:K){
  #[ToDo] Create training data xtrain and ytrain, everything except fold
  Xtrain = X[idfold != fold, ]
  Xtest = X[idfold == fold, ]
  Ytrain = Y[idfold != fold]
  Ytest = Y[idfold == fold]
  
  #[ToDo] Create testing data xtest and ytest, everything in fold

  
  #[ToDo] Center training data 


  #[ToDo] For loop over lambdas
  for (i in 1:length(lambda_seq)){
    # [ToDo] Calculate ridge solution
    
    # [ToDo] Complete with anything else you need for cvm and cvse
    
    # [ToDo] Errors calculation
    
  }
}

# [ToDo] get cvm and cvse from cv_folds


# Plot the output from above
plot(lambda_seq, cvm, col = "red", ylim = c(40, 140))
lines(lambda_seq, cvm + cvse)
lines(lambda_seq, cvm - cvse)