MarkovBlanketAlgorithm<T>

Models & Types · AiDotNet.CausalDiscovery.ConstraintBased

Markov Blanket (Grow-Shrink) Algorithm — discovers the Markov blanket of each variable.

Inheritance: CausalDiscoveryBase<T> → ConstraintBasedBase<T> → MarkovBlanketAlgorithm<T>

For Beginners

Think of a variable’s Markov blanket as a “shield” — if you know all variables in the blanket, no other variable can provide additional information about the target. This algorithm finds that shield by adding helpful variables (growing) and then removing redundant ones (shrinking).

How It Works

The Grow-Shrink algorithm identifies the Markov blanket of each variable through two phases:

  • Growing: Add variables that increase conditional mutual information with the target.
  • Shrinking: Remove variables that become conditionally independent given the rest.

Reference: Margaritis and Thrun (1999), “Bayesian Network Induction via Local Neighborhoods”.

Constructors

Constructor Summary
MarkovBlanketAlgorithm(CausalDiscoveryOptions options, int discretizationBins, double miThreshold, int minGroupSizeForMI) Creates a new Markov Blanket algorithm instance.

Properties

Property Summary
bool SupportsLatentConfounders { get; }
bool SupportsNonlinear { get; }
string Name { get; }