mk5135795 / pyalcs

Implementation of Anticipatory Learning Classifiers System (ALCS) in Python

Anticipatory Learning Classifier Systems in Python

Repository containing code implementation for various Anticipatory Learning Classifier Systems (ALCS).

Agents

ACS

Introduced by Stolzmann in 1997 originally intended to simulate and evaluate Hoffmann's learning theory of anticipations.

• LCS framework with explicit representation of anticipations
• directed anticipatory learning process

ACS2

Added modifications:

• start with initially empty population of classifiers that are created by covering mechanism,
• genetic generalization mechanism
• population includes C-A-E triples that anticipate no change in the environment (ACS by default assumes no changes),
• after executing an action modification are applied to all action set [A],
• classifier has an extra property of "immediate reward".

YACS

todo

MACS

todo

Documentation

Documentation is available here.

Citation

If you want to use the library in your project please cite the following:

@inbook{
    title = "Integrating Anticipatory Classifier Systems with OpenAI Gym",
    keywords = "Aniticipatory Learning Classifier Systems, OpenAI Gym",
    author = "Norbert Kozlowski, Olgierd Unold",
    year = "2018",
    doi = "10.1145/3205651.3208241",
    isbn = "978-1-4503-5764-7/18/07",
    booktitle = "Proceedings of the Genetic and Evolutionary Computation Conference (GECCO '18)",
    publisher = "Association for Computing Machinery",
}

Prior to PR please execute to check if standards are holding:

make test