Deep Learning Approaches to Predicting the Optimal Chess Moves from Board Positions

Deep Learning Approaches to Predicting the Optimal Chess Moves from Board Positions
  IJETT-book-cover           
  
© 2024 by IJETT Journal
Volume-72 Issue-4
Year of Publication : 2024
Author : Muhammad Faiz Arsalan, Haryono Soeparno
DOI : 10.14445/22315381/IJETT-V72I4P105

How to Cite?

Muhammad Faiz Arsalan, Haryono Soeparno, "Deep Learning Approaches to Predicting the Optimal Chess Moves from Board Positions," International Journal of Engineering Trends and Technology, vol. 72, no. 4, pp. 43-50, 2024. Crossref, https://doi.org/10.14445/22315381/IJETT-V72I4P105

Abstract
In this study, the authors explore advanced methodology, which consists of two methods to predict chess strategies utilizing a neural network approach: tensor construction and a novel method, which is the move-to-delta framework. The approach commences with dataset curation from the esteemed Lichess database, transitions through tensor construction using a refined piece-centric representation, and an innovative model training underpinned by the "move-to-delta" conceptual framework. Pivotal components of the methodology are the strategic utilization of seed variability ranging from number 1 to 100 and exploring the impact of four different batch sizes (64, 128, 256, and 512), illuminating the nuanced interplay of weight initialization in neural training. The model’s performance is evaluated using two evaluation methods: the number of puzzles solved and performance metrics (MSE, MAE, and R-squared). Notably, the model initialized with seed number 33 and batch size 128 achieved exceptional capability, solving four positions out of the 25 Kaufman Test puzzles. This signifies an achievement that significantly surpasses extant chess engines, which, at best, resolve two Kaufman puzzles. This finding underscores the essential role of weight initialization, the usage of the move-to-delta framework, and the value of rigorous experimentation in the realm of chess move prediction through deep learning.

Keywords
Chess, Convolutional Neural Network, Deep learning, Kaufman Puzzle, Supervised learning.

References
[1] David Silver et al., “A General Reinforcement Learning Algorithm that Masters Chess, Shogi and Go through Self-Play,” Science, vol. 362, no. 6419, pp. 1140-1144, 2018.
[CrossRef] [Google Scholar] [Publisher Link]
[2] Omid E. David, Nathan S. Netanyahu, and Lior Wolf, “DeepChess: End-to-End Deep Neural Network for Automatic Learning in Chess,” Artificial Neural Networks and Machine Learning-ICANN, 2016: 25th International Conference on Artificial Neural Networks, Barcelona, Spain, pp. 88-96, 2016.
[CrossRef] [Google Scholar] [Publisher Link]
[3] Matthew Lai, “Giraffe: Using Deep Reinforcement Learning to Play Chess,” Arxiv, pp. 1-39, 2015.
[CrossRef] [Google Scholar] [Publisher Link]
[4] Pieter Bijl, and Anh Phi Tiet, “Exploring Modern Chess Engine Architecture,” Bachelor Thesis, Vrije Universiteit Amsterdam, pp. 1-30, 2021.
[Google Scholar] [Publisher Link]
[5] Board Representation, Chess Programming Wiki, 2020. [Online]. Available: https://www.chessprogramming.org/Board_Representation
[6] Larry Kaufman, “Rate Your Own Computer,” Computer Chess Reports, vol. 3, no. 1, pp. 1-24, 1992.
[Google Scholar] [Publisher Link]
[7] Larry Kaufman, “Rate Your Own Computer – Part II,” Computer Chess Reports, vol. 3, no. 2, pp. 1-23, 1992.
[Publisher Link]
[8] Larry Kaufman, “Reprint Problems,” Computer Chess Reports, vol. 4, no. 1, pp. 1-31, 1993.
[Publisher Link]
[9] Matthia Sabatelli et al., “Learning to Evaluate Chess Positions with Deep Neural Networks and Limited Lookahead,” Proceedings of the 7 th International Conference on Pattern Recognition Applications and Methods ICPRAM, Funchal, Madeira, Portugal, vol. 1, pp. 276- 283, 2018.
[CrossRef] [Google Scholar] [Publisher Link]
[10] Arman Maesumi, “Playing Chess with Limited Look Ahead,” Arxiv, pp. 1-11, 2020.
[CrossRef] [Google Scholar] [Publisher Link]
[11] Fenil Mehta et al., “Predicting Chess Moves with Multilayer Perceptron and Limited Lookahead,” Journal of Engineering Research and Application, vol. 10, no. 4, pp. 5-8, 2020.
[Google Scholar] [Publisher Link]
[12] Barak Oshri, and Nishith Khandwala, “Predicting Moves in Chess Using Convolutional Neural Networks,” ConvChess, pp. 1-8, 2016.
[Google Scholar] [Publisher Link]
[13] Chao Gao, Ryan Hayward, and Martin Müller, “Move Prediction Using Deep Convolutional Neural Networks in Hex,” IEEE Transactions on Games, vol. 10, no. 4, pp. 336-343, 2017.
[CrossRef] [Google Scholar] [Publisher Link]
[14] Berk Kapicioglu et al., “Chess2vec: Learning Vector Representations for Chess,” Arxiv, pp. 1-5, 2020.
[CrossRef] [Google Scholar] [Publisher Link]
[15] Joseph Lemley et al., “CWU-Chess: An Adaptive Chess Program that Improves After Each Game,” 2018 IEEE Games, Entertainment, Media Conference (GEM), Galway, Ireland, pp. 1-9, 2018.
[CrossRef] [Google Scholar] [Publisher Link]
[16] Avi Schwarzschild et al., “Can you Learn an Algorithm? Generalizing from Easy to Hard Problems with Recurrent Networks,” Advances in Neural Information Processing Systems, vol. 34, pp. 6695-6706, 2021.
[Google Scholar] [Publisher Link]
[17] Rafał Dreżewski, and Grzegorz Wątor, “Chess as Sequential Data in a Chess Match Outcome Prediction Using Deep Learning with Various Chessboard Representations,” Procedia Computer Science, vol. 192, pp. 1760-1769, 2021.
[CrossRef] [Google Scholar] [Publisher Link]
[18] Zheyuan Fan, Yuming Kuang, and Xiaolin Lin, “Chess Game Result Prediction System,” Stanford University, CS 229 Machine Learning Project Report, pp. 1-5, 2013.
[Google Scholar]
[19] Chris J. Maddison et al., “Move Evaluation in Go Using Deep Convolutional Neural Networks,” Arxiv, pp. 1-8, 2020.
[CrossRef] [Google Scholar] [Publisher Link]
[20] Ilya Sutskever, and Vinod Nair, “Mimicking Go Experts with Convolutional Neural Networks,” Artificial Neural Networks-ICANN, 18th International Conference, Prague, Czech Republic, Proceedings, Part II, pp. 101-11, 2008.
[CrossRef] [Google Scholar] [Publisher Link]