intDAG: A DAG function for interventional data

In intdag: Reconstruction of a Directed Acyclic Graph with Interventions

Description

This function allows you to learn the DAG structure from interventional data

Usage

intDAG(X, p, lambda, lambda2, tau = 0.05, rho = 1, A_NZ0 = NULL,
  A0 = NULL, sigma = NULL, Sig = NULL, variance.constraint = TRUE,
  opts.tol = 0.001, maxIter = 1000)

Arguments

X	The n by M data matrix
p	The dimension of the adjacency matrix
lambda	tuning parameter for the first penalty of the adjacency matrix
lambda2	tuning parameter for the sparsity penalty of the intervention matrix
tau	tuning parameter of the TLP function, default is 0.05
rho	the ADMM penalty parameter, default is 1
A_NZ0	An p by M matrix indicating nonzero elements as initial values
A0	An p by M matrix as initial values for (A, B)
sigma	the parameter in the variance constraint, not needed when variance.constraint is set to FALSE
Sig	vector of length p, the error variances of each node, not needed when variance.constraint is set to FALSE
variance.constraint	a flag indicating if the variance constraint is included, default is TRUE
opts.tol	Tolerance for convergence
maxIter	maximum number of iterations in ADMM loop

Value

A list with components

A	Estimated adjacency matrix
B	Estimated intervention matrix
Sig	Estimated vector of error variances of each node
sigma	Estimated paramter in the variance constraint

Examples

p <- w <- 10
s0 <- p # number of edges
lower <- rep(0, (p*(p-1)/2)) # num of possible edges
nz_set <- sample(1:(p*(p-1)/2), s0) # sample a non-zero edge set
lower[nz_set] <- 0.5
amat <- matrix(0, p, p)
amat[lower.tri(amat)] <- lower
bmat <- diag(sqrt(seq(1, 1.5, length=p)))
Sig <- seq(1.5, 1, length = p)
X <- rmvDAG_int(100, amat, bmat, Sig)
Sig0 <- rep(1, p)
sigma0 <- 3
out <- intDAG(X, p, 2, 2, 0.05, rho=10, sigma=sigma0, Sig=Sig0)

intdag documentation built on Oct. 9, 2019, 9:05 a.m.