Machine learning in chess

The thesis focuses on the application of algorithms and machine learning methods in game simulation and position analysis. Specifically, it describes the process of data collection, designing models capable of generating chess moves, and analyzing given positions. The potential for combining search algorithms with machine learning algorithms was also explored. The initial part of the thesis presents the theoretical foundations of applying machine learning in strategic games, particularly in chess. Key approaches, such as reinforcement learning and supervised learning of deep neural networks, which form the basis of modern chess systems, are described. The practical part of the thesis involves designing and training machine learning algorithms capable of generating legal chess moves based on position or evaluation functions. Open data sources were utilized, including grandmaster-level games and engine-evaluated positions. The models were trained using the PyTorch framework, and the quality of their generated moves was assessed through simulated games against the current top chess engine. The outcome of the thesis is a set of algorithms capable of playing chess games. The summary lists potential future development steps, advantages, and disadvantages of the algorithms.