IAMBAlgorithm<T>

Models & Types · AiDotNet.CausalDiscovery.ConstraintBased

IAMB (Incremental Association Markov Blanket) — efficient Markov blanket discovery.

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

For Beginners

IAMB finds the “neighborhood” (Markov blanket) of each variable — the set of variables that directly influence it or are directly influenced by it. It does this by adding the most relevant variables one at a time, then pruning any that turn out to be redundant. The blankets are then combined into a causal graph.

How It Works

IAMB discovers the causal structure by first finding the Markov blanket of each variable using a two-phase approach:

  • Forward phase: Greedily add the variable most associated with the target

(given the current blanket) until no more significantly associated variables exist

  • Backward phase: Remove any variable from the blanket that becomes

conditionally independent of the target given the remaining blanket members

  • Build the skeleton from pairwise Markov blanket membership
  • Orient edges using v-structure detection and Meek rules

Reference: Tsamardinos et al. (2003), “Algorithms for Large Scale Markov Blanket Discovery”.

Constructors

Constructor Summary
IAMBAlgorithm(CausalDiscoveryOptions options) Initializes IAMB with optional configuration.

Properties

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