Abstract
Board game research has pursued two distinct but linked objectives: solving games, and strong play using heuristics. In our case study in the game of chess, we analyze how current AlphaZero type architectures learn and play late chess endgames, for which perfect play tablebases are available. We study the open source program Leela Chess Zero in three and four piece chess endgames. We quantify the program’s move decision errors for both an intermediate and a strong version, and for both the raw policy network and the full MCTS-based player. We discuss a number of interesting types of errors by using examples, explain how they come about, and present evidence-based conjectures on the types of positions that still cause problems for these impressive engines.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Ballicora, M.: Gaviota. https://sites.google.com/site/gaviotachessengine/Home/endgame-tablebases-1. Accessed 6 June 2021
Fiekas, N.: python-chess, August 2014. https://python-chess.readthedocs.io
Forstén, H.: Purpose of the moves left head. https://github.com/LeelaChessZero/lc0/pull/961#issuecomment-587112109. Accessed 24 Oct 2021
Huntington, G., Haworth, G.: Depth to mate and the 50-move rule. ICGA Journal 38(2), 93–98 (2015)
Kryukov, K.: Number of unique legal positions in chess endgames (2014). http://kirill-kryukov.com/chess/nulp/. Accessed 30 Oct 2021
Lassabe, N., Sanchez, S., Luga, H., Duthen, Y.: Genetically programmed strategies for chess endgame. In: Genetic and Evolutionary Computation, pp. 831–838 (2006)
Lc0 Authors: What is Lc0?. http://lczero.org/dev/wiki/what-is-lc0/. Accessed 24 Oct 2021
Lc0 Authors: Win-draw-loss evaluation. http://lczero.org/blog/2020/04/wdl-head/. Accessed 24 Oct 2021
de Man, R.: Syzygy. http://github.com/syzygy1/tb. Accessed 6 June 2021
de Man, R., Guo, B.: Syzygy endgame tablebases. http://syzygy-tables.info/. Accessed 6 June 2021
Romein, J.W., Bal, H.E.: Awari is solved. ICGA J. 25(3), 162–165 (2002)
Silver, D., et al.: A general reinforcement learning algorithm that masters chess, shogi, and go through self-play. Science 362(6419), 1140–1144 (2018)
Silver, D., et al.: Mastering the game of Go without human knowledge. Nature 550(7676), 354–359 (2017)
Acknowledgements
The authors gratefully acknowledge support from NSERC, the Natural Sciences and Engineering Research Council of Canada, and from Müllers DeepMind Chair in Artificial Intelligence and Canada CIFAR AI Chair.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2022 Springer Nature Switzerland AG
About this paper
Cite this paper
Haque, R., Wei, T.H., Müller, M. (2022). On the Road to Perfection? Evaluating Leela Chess Zero Against Endgame Tablebases. In: Browne, C., Kishimoto, A., Schaeffer, J. (eds) Advances in Computer Games. ACG 2021. Lecture Notes in Computer Science, vol 13262. Springer, Cham. https://doi.org/10.1007/978-3-031-11488-5_13
Download citation
DOI: https://doi.org/10.1007/978-3-031-11488-5_13
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-11487-8
Online ISBN: 978-3-031-11488-5
eBook Packages: Computer ScienceComputer Science (R0)