LogFileAnalyzer for Learning Classifier Systems
The LogFileAnalyzer is a Java program that supports the analysis of Learning Classifier System (LCS) experiments. LCSs are rule-based evolutionary learning systems that code their knowledge in a population of rules (called classifiers). In an LCS experiment, the classifier population evolves and changes constantly. New classifiers are created, bad classifiers are deleted, and existing classifiers are modified and updated.
The LogFileAnalyzer can help to better understand this dynamic process by visualizing classifier sets in tables and diagrams:
For each iteration in an LCS experiment, the classifier population, the match set, and the action set can be displayed in separate tables. The tables can be sorted using any classifier property as criterion. Therefore, classifiers of special interest (e.g. those having a high prediction) can easily be determined for further investigation. The development of classifier sets over time can be tracked iterationwise or by directly addressing the sets of an iteration by its number.
Besides displaying classifier sets in tables, sets can be visualized using histograms. Histograms are created automatically for each classifier property and visualize the distribution of classifiers with respect to the displayed property. Filters can be applied to each histogram chart, additionally charts can be saved and printed.
Single classifiers can be compared to all other classifiers in a set. Diagrams display the classifier's properties relative to the minimum and maximum values of the containing set.