Improvements in Evaluation Functions Based on Endgame Databases of Chinese Dark Chess

Alpha-beta-based search is used in the game of Chinese dark chess, which is stochastic and has perfect information. To efficiently determine a good move, an evaluation function needs to be carefully designed. This paper aims to enhance the evaluation function for material combinations with seven or more pieces by extracting information from 6-piece endgame databases. Firstly, we utilize the existing endgames to evaluate a win/loss ratio of each material combination, establishing a benchmark for endgame scores. We then design the values of each material combination, called the MCV, based on the ratios. Secondly, we extend the MCVs to encompass a larger number of pieces and integrate it into the hybrid evaluation function used by the game-playing program. The experimental results demonstrate that the original evaluation function has been improved, leading to an overall enhancement in playing strength.