Estimate target rows of a high-dimensional precision matrix — estimate_precision

Nodewise lasso is the general option. The diagonal option is intentionally restricted to designs whose independence/diagonal-covariance structure is assumed in advance.

Usage

estimate_precision_rows(
  X,
  targets,
  method = c("nodewise", "diagonal", "user"),
  lambda_node = NULL,
  theta_rows = NULL,
  assume_independent = FALSE,
  glmnet_thresh = 1e-10
)

Arguments

X: Centered and standardized numeric design matrix.
targets: One-based target-column indices.
method: One of "nodewise", "diagonal", or "user".
lambda_node: Nodewise lasso penalty.
theta_rows: User-supplied rows when method = "user".
assume_independent: Must be TRUE for diagonal precision.
glmnet_thresh: Convergence threshold passed to glmnet::glmnet().

Value

A precision-direction specification.