Human chess players rely strongly on patterns of the chessboard and immediate tactical threats while considering every move. This paper claims an improved move ordering method that uses a neural network to learn the relationship between the control of the squares and the influence of the move. Instead of the popular approach of using a history heuristic with capture moves, Greer uses a chessmap (an abstract representation of a chess position) heuristic to mimic the human thought process that orders moves depending on which areas of the chessboard they influence.

Two methods are presented to determine what sectors are relevant to a position: a knowledge-based chessmap heuristic (using a neural network) and an experience-based technique in which the sectors are ordered according to the results of the previous search of the game tree. A simple feedforward neural network was used for training, and the input vectors consisted of 70 elements (64 elements represented the chessmap values, and the remaining six represented the relative positions of the kings). Test results reveal that the neural network technique performs slightly better than the random move ordering. To improve the efficiency of the chessmap heuristic, Greer suggests more precise heuristics (for example, extracting tactical threats to complement the positional evaluation).

Even though the empirical results reveal that the history heuristic searches only slightly fewer nodes than the chessmap heuristic and requires less computation time, it might be of great interest to continue this research and to incorporate more powerful heuristics. This paper presents some interesting ideas, and I hope the author will continue to improve the move-ordering algorithm by incorporating more heuristics and improving the learning capability of the neural network.