Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Approach to Analyze Niche Evolution in XCS Models

Pier Luca Lanzi

TL;DR

This work addresses how to analyze niche evolution in XCS classifier systems by identifying currently active evolutionary niches (CAN) using occurrence-based activation and augmenting classifiers with time-stamp histories. The authors introduce a lightweight, representation-independent approach that leverages existing XCS data, including a timestamp list L per classifier, to estimate niche counts via CAN and MAN metrics and to track niche composition over time. Through experiments on binary single-step and multi-step problems with non-overlapping and overlapping solutions, the method yields robust niche-count estimates, with CAN and MAN converging toward the optimal niche count after condensation, and reveals overlapping niches that standard snapshots may miss. The approach enables deeper insight into XCS dynamics and offers potential for improved condensation strategies and explainable classifier systems, with extensions to more advanced representations as a promising direction.

Abstract

We present an approach to identify and track the evolution of niches in XCS that can be applied to any XCS model and any problem. It exploits the underlying principles of the evolutionary component of XCS, and therefore, it is independent of the representation used. It also employs information already available in XCS and thus requires minimal modifications to an existing XCS implementation. We present experiments on binary single-step and multi-step problems involving non-overlapping and highly overlapping solutions. We show that our approach can identify and evaluate the number of niches in the population; it also show that it can be used to identify the composition of active niches to as to track their evolution over time, allowing for a more in-depth analysis of XCS behavior.

An Approach to Analyze Niche Evolution in XCS Models

TL;DR

This work addresses how to analyze niche evolution in XCS classifier systems by identifying currently active evolutionary niches (CAN) using occurrence-based activation and augmenting classifiers with time-stamp histories. The authors introduce a lightweight, representation-independent approach that leverages existing XCS data, including a timestamp list L per classifier, to estimate niche counts via CAN and MAN metrics and to track niche composition over time. Through experiments on binary single-step and multi-step problems with non-overlapping and overlapping solutions, the method yields robust niche-count estimates, with CAN and MAN converging toward the optimal niche count after condensation, and reveals overlapping niches that standard snapshots may miss. The approach enables deeper insight into XCS dynamics and offers potential for improved condensation strategies and explainable classifier systems, with extensions to more advanced representations as a promising direction.

Abstract

We present an approach to identify and track the evolution of niches in XCS that can be applied to any XCS model and any problem. It exploits the underlying principles of the evolutionary component of XCS, and therefore, it is independent of the representation used. It also employs information already available in XCS and thus requires minimal modifications to an existing XCS implementation. We present experiments on binary single-step and multi-step problems involving non-overlapping and highly overlapping solutions. We show that our approach can identify and evaluate the number of niches in the population; it also show that it can be used to identify the composition of active niches to as to track their evolution over time, allowing for a more in-depth analysis of XCS behavior.

Paper Structure

This paper contains 19 sections, 6 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. The XCS Classifier System
  3. The Performance Component
  4. The Evolutionary (Discovery) Component
  5. Active Evolutionary Niches
  6. A Simple Binary Problem
  7. Modifications to XCS Models
  8. Number of Currently Active Niches
  9. Composition of Active Niches
  10. The Case of Overlapping Solutions
  11. Discussion
  12. Experiments on Binary Problems
  13. Design of Experiments
  14. Boolean Functions
  15. Multi-Step Problems
  16. ...and 4 more sections

Figures (3)

  • Figure 1: Example of overlapping classifiers from an optimal solution for MAJ3; the black boxes represent input-action pairs; the white boxes show classifiers using their condition:action pairs, prediction $p$, fitness $f$, and the content of the list $L$ of $ats$ values starting from the last one added.
  • Figure 2: XCS applied to the 20-multiplexer: (a) performance, number of classifiers, and currently active niches ( CAN), curves are averages over 20 runs); (b) evolution of niche size using the data collected of all the 20 runs.
  • Figure 3: Grid environments that have been used for evaluating XCS performance in multistep problems: (a) Woods1, (b) Woods14, (c) Maze4, (d) Woods2, (e) Maze5, and (f) Maze6wilson:1995lanzi:1997:icgawilson:1998lanzi:1999:analysis.