cv_grid_lasso: Computes cross-validated solutions for continuous response...

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

Description

Cross-validation wrapper for grid_lasso that computes solutions, selects and fits the optimal model.

Usage

cv_grid_lasso(
  x,
  y,
  K = 5,
  var_order = NULL,
  lambda = NULL,
  nlambda = 100L,
  lambda.min.ratio = ifelse(n < p, 0.01, 1e-04),
  grid.size = p,
  thresh = 1e-10,
  maxit = 1e+05,
  sparse = TRUE,
  mc.cores = 1,
  return.full.beta = FALSE,
  silent = TRUE,
  early.stopping = TRUE,
  early.stopping.factor = 0.5,
  fold_assign = NULL,
  missing.data = F,
  psd.method = "enet"
)

Arguments

x	Design matrix, n x p.
y	Vector of responses, length n.
K	Number of folds for cross-validation. Must be at least 2
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
thresh	Convergence threshold for coordinate descent for difference in objective values between successive iterations
maxit	Maximum number of iterations for coordinate descent routine
sparse	Whether to use sparse matrices in computation (setting FALSE recommended for advanced users only)
mc.cores	Number of cores to be made available for computing the cross-validation estimates in parallel
return.full.beta	Return the entire solution path for the chosen variable subset, as opposed to only the estimate for estimated optimal lambda
silent	Suppress some text to console
early.stopping	Whether square-root lasso condition for early stopping along lambda path should be used
early.stopping.factor	Factor of correction in square-root lasso early stopping criterion
fold_assign	For user-specified vector of assignment of folds for cross-validation. Must be of the form of integer vector with entries in 1 , ... , K.
missing.data	If TRUE then will use (slower) procedure that corrects for missing data
psd.method	The way that the gram matrix is made positive semidefinite. By default an elastic net term, alternatives are "coco" for CoCoLasso

Value

A list of objects:

• mu – estimated intercept

• beta – coefficient estimate

• beta.full – full solution path (only returned if return.full.beta = TRUE)

• lambda – Vector of values for lambda used

• cv – matrix of cross-validation error for each value of lambda and gridpoint

• col.means – vector of column means in unstandardized design matrix; important for making predictions on new data.

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)
Y = X %*% betavec
Y = Y + rnorm(50)
X = X + matrix(rnorm(50*500), 50, 500)
mod1 = cv_grid_lasso(X, Y, grid.size = 50)
predict(mod1, Z)

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