EqVarDAG_HD_CLIME: Estimate topological ordering and DAG using high dimensional...

In WY-Chen/EqVarDAG: Causal Discovery with Equal Variance Assumption

Estimate topological ordering and DAG using high dimensional bottom-up CLIME approach Estimate DAG using topological ordering

Description

Estimate topological ordering and DAG using high dimensional bottom-up CLIME approach Estimate DAG using topological ordering

Usage

EqVarDAG_HD_CLIME(
  X,
  mtd = "dlasso",
  alpha = 0.05,
  threshold = 0.1,
  FCD = TRUE,
  precmtd = "sqrtlasso"
)

Arguments

X, Y:	n x p and 1 x p matrix
alpha:	desired selection significance level
mtd:	methods for learning DAG from topological orderings. "ztest": (p<n) [Multiple Testing and Error Control in Gaussian Graphical Model Selection. Drton and Perlman.2007] "rls": (p<n) fit recursive least squares using ggm package and threshold the regression coefs "chol": (p<n) perform cholesky decomposition and threshold the regression coefs "dlasso": debiased lasso (default with FCD=True and precmtd="sqrtlasso"); "lasso": lasso with fixed lambda from [Penalized likelihood methods for estimation of sparse high-dimensional directed acyclic graphs. Shojaie and Michailidis. 2010]; "adalasso": adaptive lasso with fixed lambda from [Shojaie and Michailidis. 2010]; "cvlasso": cross-validated lasso from glmnet; "scallasso": scaled lasso.
threshold:	for rls and chol, the threshold level.
FCD:	for debiased lasso, use the FCD procedure [False Discovery Rate Control via Debiased Lasso. Javanmard and Montanari. 2018] or use individual tests to select support.
precmtd:	for debiased lasso, how to compute debiasing matrix "cv": node-wise lasso w/ joint 10 fold cv "sqrtlasso": square-root lasso (no tune, default)

Value

Adjacency matrix with ADJ[i,j]!=0 iff i->j, and topological ordering

Examples

X1<-rnorm(100)
X2<-X1+rnorm(100)
EqVarDAG_HD_TD(cbind(X1,X2),2)

#$adj
#[,1] [,2]
#[1,]    0    1
#[2,]    0    0
#
#$TO
#[1] 1 2

WY-Chen/EqVarDAG documentation built on May 10, 2022, 7:08 a.m.