CDNODAlgorithm<T>

Models & Types · AiDotNet.CausalDiscovery.ConstraintBased

CD-NOD — Constraint-based Discovery from Non-stationary / heterogeneous Data.

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

For Beginners

CD-NOD exploits the idea that when conditions change, causes tend to stay the same while effects change. By adding a “time” or “context” variable, it can detect which relationships shift, giving extra information for causal direction.

How It Works

CD-NOD extends constraint-based causal discovery to handle data from changing environments or multiple domains. It augments the data with a context variable C (representing time index or domain indicator), then uses conditional independence tests to detect which causal mechanisms change across contexts and leverages these changes to orient more edges than standard PC.

Algorithm:

  • Augment data with a context variable C (row index normalized to [0,1])
  • Run PC skeleton phase on augmented data (d+1 variables)
  • Identify variables connected to C (these have changing distributions)
  • For edges where one endpoint changes and the other doesn’t:

orient non-changing → changing (cause is stable, effect changes)

  • Apply standard v-structure and Meek rules on remaining edges
  • Return the d x d adjacency matrix (dropping the context variable)

Reference: Huang et al. (2020), “Causal Discovery from Heterogeneous/Nonstationary Data”, JMLR.

Constructors

Constructor Summary
CDNODAlgorithm(CausalDiscoveryOptions options) Initializes CD-NOD with optional configuration.

Properties

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