Package edu.cmu.tetrad.search.score

Class ZsbScore

java.lang.Object
edu.cmu.tetrad.search.score.ZsbScore
All Implemented Interfaces:
Score
public class ZsbScore extends Object implements Score
Implements an unpublished score based on a risk bound due to Zhang and Shen. It adapts Theorem 1 in the following reference:

Zhang, Y., & Shen, X. (2010). Model selection procedure for high‐dimensional data. Statistical Analysis and Data Mining: The ASA Data Science Journal, 3(5), 350-358

The score uses Theorem 1 in the above to numerically search for a lambda value that is bounded by a given probability risk, between 0 and 1, if outputting a local false positive parent for a variable. There is a parameter m0, which is a maximum number of parents for a particular variable, which is free. The solution of this score is to increase m0 from 0 upward, re-evaluating with each scoring that is done using that variable as a target node. Thus, over time, a lower bound on m0 is estimated with more and more precision. So as the score is used in the context of FGES or GRaSP, for instance, so long as the score for a given node is visited more than once, the scores output by the procedure can be expected to improve, though setting m0 to 0 for all variables does not give bad results even by itself.

This score is conservative for large, dense models and faster than other available scores in this package. The risk bound is easily interpreted.

As for all scores in Tetrad, higher scores mean more dependence, and negative scores indicate independence.

Version:
$Id: $Id
Author:
josephramsey