boot.notears: Linear NO-TEARS Algorithm (Reimplemented) With Bootstrapping

In rlebron-bioinfo/gnlearn: Genetic Network Learning

Description

This function allows you to learn a directed graph from a dataset using the Linear NO-TEARS algorithm.

Usage

boot.notears(
  df,
  lambda1 = 0.1,
  loss.type = c("l2", "logistic", "poisson"),
  max.iter = 100,
  h.tol = 1e-06,
  rho.max = 1e+06,
  w.threshold = 0.1,
  R = 200,
  m = NULL,
  threshold = 0.5,
  to = c("igraph", "adjacency", "edges", "graph", "bnlearn"),
  cluster = parallel::detectCores(),
  seed = sample(1:10^6, 1)
)

Arguments

df	Dataset.
lambda1	L1 regularization parameter. Default: 0.1
loss.type	Type of loss function to be used: 'l2', 'logistic', or 'poisson'. Default: 'l2'
max.iter	Maximum number of dual ascent steps. Default: 100
h.tol	Minimum absolute value of h. Default: 1e-8
rho.max	Maximum value of rho. Default: 1e+16
w.threshold	Threshold of absolute value of weight. Default: 0.1
R	Number of bootstrap replicates (optional). Default: 200
m	Size of training set (optional). Default: nrow(df)/2
threshold	Minimum strength required for a coefficient to be included in the average adjacency matrix (optional). Default: 0.5
to	Output format ('adjacency', 'edges', 'graph', 'igraph', or 'bnlearn') (optional).
cluster	A cluster object from package parallel or the number of cores to be used (optional). Default: parallel::detectCores()
seed	Seed used for random selection. Default: NULL

Examples

obj <- boot.notears(df)
avg.g <- obj$average
g.rep <- obj$replicates

rlebron-bioinfo/gnlearn documentation built on July 25, 2020, 12:38 p.m.