Аңдатпа
An adaptation of the AlphaZero model for the game Togyzkumalak, a national Kazakh board game with high combinatorial complexity is presented. The aim of the research was to develop an artificial intelligence capable of independently learning optimal game strategies without using predefined templates. The paper uses an approach based on the modification of the architecture of the AlphaZero neural network, which includes deep learning and the Monte Carlo search method (MCTS). The model was implemented on the PyTorch platform and was trained in the process of self-play, which allowed us to gradually improve strategic skills. The key features of Togyzkumalak's game are taken into account, such as a large number of possible positions on the playing field, and specialized network architectures have been developed for their effective modeling. The results demonstrate the significant potential of combining deep learning and MCTS to solve non-standard game tasks and create effective strategies in Togyzkumalak.