Enhancing Chess Engine with a Personalized Quantitative Database

Abstract

Designing chess engines have become a popular topic of computer science studying since 1997, when Deep Blue, invented by the IBM company, defeat Gary Kasparov in a 6-game match. Nowadays, chess engine designers improve their engines with more efficient neural networks, enhanced self-learning methods, complete databases of past games, and so on. However, what these designers ignore is the human’s perspective of playing chess. Different human players have different styles and preferences, and even when their Elo ratings are near, they may choose various options in the same situation. This human’s way of thinking is what modern chess engines do not consider, and I assume that their performance when playing against human players could improve by considering these different styles. What this research is attempting is to set up a personalized quantitative system based on previous games. The author’s and some other chess players’ understanding of chess would be added to create more enhanced standards for this quantitative system. After that, the system will be implemented to Stockfish, the best open-source chess engine in the world, by modifying its source codes. If the modified engine knows the opponents they are facing, it adds those quantitative statistics into analysis with a specific weight. Instead of directly calculates the best moves for a position, the engine chooses the moves that bring the hugest challenge for the opponent, and the moves chosen vary when the opponents change. Then, the performance of Stockfish will be compared to its original version, and verify whether the engine gets better. If the research is done successfully, both the designs of the chess engine and the methods for training human players would get improved.