bang: Causal discovery with latent confounding and non-Gaussian...

In ysamwang/ngBap: Estimates BAP structures assuming non-Gaussian errors

Causal discovery with latent confounding and non-Gaussian data

Description

Estimates a graph from data generated by a linear SEM with non-Gaussian errors corresponding to a BAP

Usage

bang(Y, K, level = 0.01, verbose = T, restrict = 1)

Arguments

Y	n x p matrix of observations with sample in row, variable in column
K	The degree of the moment which is non-Gaussianity (i.e., this is K in the paper)
level	the nominal level for the empirical likelihood tests
verbose	Print progress updates
restrict	Which moment restrictions to test in empirical likelihood restrict = 1 restricts all moments up to degree K; i.e. gamma_v^k gamma_u and gamma_v gamma_u^k for k < K restrict = 2, tests moments of degree K-1 and 1; i.e. gamma_v^(K-1) gamma_u and gamma_v gamma_u^(K-1)

Value

• totalEffect estimate of (I-B)^(-1)

• directEffect estimate of B

• omega estimate of Omega

• dEdge estimate of adjacncey matrix corresponding to directed edges

• bEdge estimate of adjacncey matrix corresponding to bidirected edges

• errs estimated epsilons

Examples

## Not run: 
dat <- ngBap::rBAP(n = 50000, p = 7, dist = "gamma", d = 3, b = 5, ancestral = F, shuffle = T, signs = T)
Y <- dat$Y
out <- ngBap::bang(Y, K = 3, level = .01, verbose = F, restrict = 1)

## End(Not run)

ysamwang/ngBap documentation built on Nov. 1, 2022, 3:20 a.m.