Accuracy-based Learning Classifier Systems with Rule Combining mechanism (XCS-RC)
Based on Martin Butz's XCS code (2001) in Java. Read my PhD thesis here for a complete algorithmic description.
A new function called combining that employs inductive reasoning replacing all Darwinian genetic operation like mutation and crossover. This significantly enables the algorithm to reach maximal correctness rate and minimal population size quicker than other methods. My earlier papers comparing them can be obtained here and here.
New parameters in XCS-RC:
- Tcomb: combining period, after how many learning cycles the new technique will be applied
- predTol: prediction tolerance, the maximum difference between two classifiers to be combined
- predErrTol: prediction error tolerance, threshold for rule deletion, indicated inappropriate combining
Removed/unused parameters: all related to mutation and crossover.
This identical algorithm in different languages:
Should there be questions, feel free to ask. 🙂