ANMAlgorithm<T>

Models & Types · AiDotNet.CausalDiscovery.Functional

ANM (Additive Noise Model) — pairwise causal discovery via independence of residuals.

Inheritance: CausalDiscoveryBase<T> → FunctionalBase<T> → ANMAlgorithm<T>

For Beginners

If X truly causes Y, then the “noise” left over after predicting Y from X should have nothing to do with X. But if you try predicting X from Y, the leftover noise will still be related to Y. ANM uses this asymmetry to figure out which variable causes which.

How It Works

ANM determines causal direction between variable pairs by fitting Y = f(X) + N in both directions and checking which direction yields residuals N that are independent of the cause. If the residuals from X → Y are more independent of X than the residuals from Y → X are of Y, then X → Y is the inferred causal direction.

Reference: Hoyer et al. (2008), “Nonlinear Causal Discovery with Additive Noise Models”, NIPS.

Constructors

Constructor Summary
ANMAlgorithm(CausalDiscoveryOptions options)

Properties

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