Enhancing Stockfish: A Chess Engine Tailored for Training Human Players

Stockfish is a highly popular open-source chess engine known for its exceptional strength. The recent integration of a neural network called NNUE has significantly improved Stockfish’s playing ability. However, the neural network lacks the capability to explain the reasoning behind its moves. This poses a challenge for human players who seek moves that align with their playing style. The objective of this paper is to describe some modifications to Stockfish, making the engine more suitable for training human players of all skill levels. We have refactored the move search and evaluation algorithms to selectively analyze potential continuations, incorporating dynamic evaluations based on the specific nature of the position and the player’s training abilities. The engine remains strength is still very high, in some situations even better than the original. We evaluate and discuss the outcomes of these enhancements.