Package edu.cmu.tetrad.search

Class SingletonAttacher

java.lang.Object

edu.cmu.tetrad.search.SingletonAttacher

public final class SingletonAttacher
extends Object

Utility to attach singleton (measured) variables as children of latent blocks using a Wilks rank-drop criterion, analogous to the latent->latent hierarchy test.

For each singleton s and each latent L with indicator set C_L, we form D = V \ (C_L ∪ {s}) and compute:

drop_L(s) = rank({s}, D) - rank({s}, D | C_L)

If max_L drop_L(s) >= minDrop, we add the edge L* -> s, where L* maximizes the drop. Ties are broken by latent name for determinism.

Method Summary

Modifier and Type	Method	Description
static List<Edge>	attachSingletonChildren(org.ejml.simple.SimpleMatrix S, int sampleSize, double alpha, int minDrop, int p, List<List<Integer>> latentBlocks, List<Node> latentNodes, List<Integer> singletonCols, List<Node> singletonNodes)	Returns latent->singleton edges to add.

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Method Details

attachSingletonChildren

public static List<Edge> attachSingletonChildren(org.ejml.simple.SimpleMatrix S,
 int sampleSize,
 double alpha,
 int minDrop,
 int p,
 List<List<Integer>> latentBlocks,
 List<Node> latentNodes,
 List<Integer> singletonCols,
 List<Node> singletonNodes)

Returns latent->singleton edges to add.

Parameters:

S - correlation/covariance matrix (SimpleMatrix)

sampleSize - number of rows

alpha - alpha for Wilks rank tests

minDrop - minimum required drop (e.g., 1)

p - number of observed variables (columns)

latentBlocks - list of indicator column indices for each latent (size>1)

latentNodes - parallel list of latent meta-nodes (same order as latentBlocks)

singletonCols - list of singleton column indices

singletonNodes - parallel list of singleton meta-nodes (same order as singletonCols)

Returns:

list of directed edges (latent -> singleton); does not modify any graph.