RSMAX2Algorithm<T>
Models & Types · AiDotNet.CausalDiscovery.Hybrid
RSMAX2 — Restricted Maximization, a hybrid constraint-based + score-based algorithm.
Inheritance: CausalDiscoveryBase<T> → HybridBase<T> → RSMAX2Algorithm<T>
For Beginners
RSMAX2 is a general framework for combining any “candidate finder” with any “best structure finder.” It first quickly identifies which variables MIGHT be connected, then carefully selects the best connections from those candidates.
How It Works
RSMAX2 (Restricted Structural Maximization, 2-phase) is a general framework for hybrid causal discovery. Phase 1 uses conditional independence tests to learn each variable’s candidate parent/child set (restricting the search space). Phase 2 uses greedy hill climbing with BIC scoring restricted to the candidate sets.
Algorithm:
- Restrict phase: For each variable, find candidate parents/children
using conditional independence tests (MMPC-like forward-backward selection)
- Maximize phase: Starting from an empty graph, greedily add edges
between variables and their candidates that improve BIC score
- Enforce acyclicity by rejecting edge additions that create cycles
- Apply backward deletion to remove edges that no longer improve score
Reference: Scutari (2010), “Learning Bayesian Networks with the bnlearn R Package”, JSS.
Constructors
| Constructor | Summary |
|---|---|
RSMAX2Algorithm(CausalDiscoveryOptions options) |
Initializes RSMAX2 with optional configuration. |
Properties
| Property | Summary |
|---|---|
bool SupportsNonlinear { get; } |
|
string Name { get; } |