OCSEAlgorithm<T>

Models & Types · AiDotNet.CausalDiscovery.InformationTheoretic

oCSE — optimal Causation Entropy for causal network inference.

Inheritance: CausalDiscoveryBase<T> → InfoTheoreticBase<T> → OCSEAlgorithm<T>

For Beginners

oCSE measures how much a variable helps predict another variable’s CHANGES over time (not just its values). This is closer to true causation — a cause should affect how the effect changes.

How It Works

oCSE uses causation entropy — a measure of the information a variable provides about another variable’s transition — to identify causal links. It greedily selects the optimal conditioning set that maximizes the causation entropy criterion.

Algorithm:

  • For each target Y, compute transition: delta_Y[t] = Y[t+1] - Y[t]
  • For each candidate cause X, compute causation entropy:

CE(X→Y|S) = MI(delta_Y ; X | S) where S is the current conditioning set

  • Greedily add variables to S that maximize CE until no variable exceeds threshold
  • Variables in S are the discovered causal parents of Y
  • Edge weight = causation entropy value (Gaussian MI approximation)

Reference: Sun et al. (2015), “Causal Network Inference by Optimal Causation Entropy”, SIAM Journal on Applied Dynamical Systems.

Constructors

Constructor Summary
OCSEAlgorithm(CausalDiscoveryOptions options)

Properties

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