grid_lasso: Computes solution paths for continuous response model

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

Description

Computes solutions for grid lasso method

Usage

grid_lasso(
  x = NULL,
  y,
  XtX = NULL,
  Xty = NULL,
  standardize = TRUE,
  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,
  return.list = TRUE,
  sparse = TRUE,
  grid.size.truncate = grid.size,
  early.stopping = TRUE,
  early.stopping.factor = 0.5,
  missing.data = F,
  psd.method = "enet",
  enet.scale = F
)

Arguments

x	Design matrix, n x p
y	Vector of responses, length n
XtX	User-specified (scaled and centred) gram matrix if this is known to avoid its recomputation each time
Xty	User-specified (scaled and centred) t(X) times y / n, if this is know to avoid its recomputation each time
standardize	Scales design matrix before computation. Setting FALSE recommended for advanced use only
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
return.list	Returns all solution paths as a list. If set to false this is returned as one large concatenated vector. FALSE should only be used when one is interested in running speed tests
sparse	Whether to use sparse matrices in computation (setting FALSE recommended for advanced users only)
grid.size.truncate	Not for user modification and is only altered when called from cv_grid_lasso
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
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
enet.scale	Experimental and to be removed

Value

A list of objects:

• mu – estimated intercept

• beta – a list of matrices, one for each subset of variables in the grid. Each matrix contains a full solution path

• lambda – Vector of values of lambda used

• 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 = grid_lasso(X, Y, grid.size = 50)
predict(mod1, Z, 45, 80)

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