Weakening faithfulness : some heuristic causal discovery algorithms

Zhalama, Jiji ZHANG, Wolfgang MAYER

Research output: Journal PublicationsJournal Article (refereed)Researchpeer-review

Abstract

We examine the performance of some standard causal discovery algorithms, both constraint-based and score-based, from the perspective of how robust they are against (almost) failures of the Causal Faithfulness Assumption. For this purpose, we make only the so-called Triangle-Faithfulness assumption, which is a fairly weak consequence of the Faithfulness assumption, and otherwise allows unfaithful distributions. In particular, we allow violations of Adjacency-Faithfulness and Orientation-Faithfulness. We show that the (conservative) PC algorithm, a representative constraint-based method, can be made more robust against unfaithfulness by incorporating elements of the GES algorithm, a representative score-based method; similarly, the GES algorithm can be made less error-prone by incorporating elements of the conservative PC algorithm. As our simulations demonstrate, the increased robustness seems to matter even when faithfulness is not exactly violated, for with only finite sample, distributions that are not exactly unfaithful may be sufficiently close to being unfaithful to make trouble.
Original languageEnglish
Pages (from-to)93-104
Number of pages12
JournalInternational Journal of Data Science and Analytics
Volume3
Issue number2
Early online date25 Nov 2016
DOIs
Publication statusPublished - 1 Mar 2017

Fingerprint

heuristics
simulation
distribution
method

Keywords

  • Causal discovery
  • Faithfulness
  • GES
  • PC

Cite this

@article{c03fd205e2084ce9a1b18b3ce1dc38e0,
title = "Weakening faithfulness : some heuristic causal discovery algorithms",
abstract = "We examine the performance of some standard causal discovery algorithms, both constraint-based and score-based, from the perspective of how robust they are against (almost) failures of the Causal Faithfulness Assumption. For this purpose, we make only the so-called Triangle-Faithfulness assumption, which is a fairly weak consequence of the Faithfulness assumption, and otherwise allows unfaithful distributions. In particular, we allow violations of Adjacency-Faithfulness and Orientation-Faithfulness. We show that the (conservative) PC algorithm, a representative constraint-based method, can be made more robust against unfaithfulness by incorporating elements of the GES algorithm, a representative score-based method; similarly, the GES algorithm can be made less error-prone by incorporating elements of the conservative PC algorithm. As our simulations demonstrate, the increased robustness seems to matter even when faithfulness is not exactly violated, for with only finite sample, distributions that are not exactly unfaithful may be sufficiently close to being unfaithful to make trouble.",
keywords = "Causal discovery, Faithfulness, GES, PC",
author = "Zhalama and Jiji ZHANG and Wolfgang MAYER",
year = "2017",
month = "3",
day = "1",
doi = "10.1007/s41060-016-0033-y",
language = "English",
volume = "3",
pages = "93--104",
journal = "International Journal of Data Science and Analytics",
issn = "2364-415X",
number = "2",

}

Weakening faithfulness : some heuristic causal discovery algorithms. / Zhalama, ; ZHANG, Jiji; MAYER, Wolfgang.

In: International Journal of Data Science and Analytics, Vol. 3, No. 2, 01.03.2017, p. 93-104.

Research output: Journal PublicationsJournal Article (refereed)Researchpeer-review

TY - JOUR

T1 - Weakening faithfulness : some heuristic causal discovery algorithms

AU - Zhalama,

AU - ZHANG, Jiji

AU - MAYER, Wolfgang

PY - 2017/3/1

Y1 - 2017/3/1

N2 - We examine the performance of some standard causal discovery algorithms, both constraint-based and score-based, from the perspective of how robust they are against (almost) failures of the Causal Faithfulness Assumption. For this purpose, we make only the so-called Triangle-Faithfulness assumption, which is a fairly weak consequence of the Faithfulness assumption, and otherwise allows unfaithful distributions. In particular, we allow violations of Adjacency-Faithfulness and Orientation-Faithfulness. We show that the (conservative) PC algorithm, a representative constraint-based method, can be made more robust against unfaithfulness by incorporating elements of the GES algorithm, a representative score-based method; similarly, the GES algorithm can be made less error-prone by incorporating elements of the conservative PC algorithm. As our simulations demonstrate, the increased robustness seems to matter even when faithfulness is not exactly violated, for with only finite sample, distributions that are not exactly unfaithful may be sufficiently close to being unfaithful to make trouble.

AB - We examine the performance of some standard causal discovery algorithms, both constraint-based and score-based, from the perspective of how robust they are against (almost) failures of the Causal Faithfulness Assumption. For this purpose, we make only the so-called Triangle-Faithfulness assumption, which is a fairly weak consequence of the Faithfulness assumption, and otherwise allows unfaithful distributions. In particular, we allow violations of Adjacency-Faithfulness and Orientation-Faithfulness. We show that the (conservative) PC algorithm, a representative constraint-based method, can be made more robust against unfaithfulness by incorporating elements of the GES algorithm, a representative score-based method; similarly, the GES algorithm can be made less error-prone by incorporating elements of the conservative PC algorithm. As our simulations demonstrate, the increased robustness seems to matter even when faithfulness is not exactly violated, for with only finite sample, distributions that are not exactly unfaithful may be sufficiently close to being unfaithful to make trouble.

KW - Causal discovery

KW - Faithfulness

KW - GES

KW - PC

UR - http://commons.ln.edu.hk/sw_master/5492

U2 - 10.1007/s41060-016-0033-y

DO - 10.1007/s41060-016-0033-y

M3 - Journal Article (refereed)

VL - 3

SP - 93

EP - 104

JO - International Journal of Data Science and Analytics

JF - International Journal of Data Science and Analytics

SN - 2364-415X

IS - 2

ER -