opt.target: Get an optimal intervention target

Description

Given a (observational or interventional) essential graph (or "CPDAG"), find the optimal intervention target that maximizes the number of edges that can be oriented after the intervention.

Usage

opt.target(essgraph, max.size, use.node.names = TRUE)

Arguments

essgraph

An '>EssGraph or '>graphNEL object representing a (observational or interventional) essential graph (or "CPDAG").

max.size

Maximum size of the intervention target. Only 1 and the number of vertices of essgraph are allowed; the latter means no size limit is applied (the default if the parameter is missing).

use.node.names

Indicates if the intervention target should be returned as a list of node names (if TRUE) or indices (if FALSE).

Value

A character vector of node names (if use.node.names = TRUE), or an integer vector of node indices (if use.node.names = FALSE) indicating the optimal intervention target.

Details

This function implements active learning strategies for structure learning from interventional data, one that calculates an optimal single-vertex intervention target, and one that calculates an optimal intervention target of arbitrary size. "Optimal" means the proposed intervention target guarantees the highest number of edges that can be oriented after performing the intervention, assuming the essential graph provided as input is the true essential graph under the currently available interventional data (i.e., neglecting possible estimation errors).

Implementation corresponds to algorithms "OptSingle" and "OptUnb" published in Hauser and B<U+00FC>hlmann (2012).

References

A. Hauser and P. B<U+00FC>hlmann (2012). Two optimal strategies for active learning of causal models from interventions. Proceedings of the 6th European Workshop on Probabilistic Graphical Models (PGM-2012), 123--130

Examples

Run this code

# NOT RUN {
## Load predefined data
data(gmG)

## Define the score (BIC)
score <- new("GaussL0penObsScore", gmG8$x)

## Estimate the essential graph using GES
ges.fit <- ges(score)
essgraph <- ges.fit$essgraph

## Plot the estimated essential graph
if (require(Rgraphviz)) {
  plot(essgraph, main = "Estimated CPDAG")
}
## The CPDAG has 1 unoriented component with 3 edges (Author <-> Bar, Bar <->
## Ctrl, Bar <-> V5)

## Get optimal single-vertex and unbounded intervention target
opt.target(essgraph, max.size = 1)
opt.target(essgraph, max.size = essgraph$node.count())
# }

Run the code above in your browser using DataCamp Workspace