Search and Knowledge in Lines of Action

Abstract

This paper describes the design and development of two w0rld-class Lines of Action game-playing programs: YL, a three time Computer Olympiad gold-medal winner, and MaNA, which has dorninated international e-mail correspondence play. The underlying design philosophy of the two programs is very different: the former emphasizes fast and efficient search, whereas the latter focuses on a sophisticated but relatively slow evaluation of each board position. In addition to providing a technical description of each program, we explore some long-standing questions on the trade-offs between search and knowledge. These experimental results confirm the conclusions made by earlier researchers in the domain of chess, thus showing that the trends are not game-specific. In particular, we see dirninishing retums with additional search depth, and observe that the knowledge level of a program has a significant impact on the results of such experiments.Keywords: Lines of Action, search, knowledge IntroductionOne of the most important considerations when designing a strategic gameplaying program is the trade-off between know ledge and search.To decide on the best move continuation, programs typicallx perfonn a lookahead search, evaluate the positions at the leaves of the search tree, and then propagate those values back to the root using the minimax principle. A program that uses a sophisticated but time-consuming board evaluation can more accurately determine the merit of each game-state visited, at the cost of sacrificing some of the look-ahead depth. Conversely, a program that uses a faster but less sophisticated board evaluation method can perform a deeper search, improving its short-term tactica! ability. There is also compensation toward better knowledge, in that each additionallevel of search provides a more refined approximation of the value of each preceding position.The trade-off between knowledge vs. search has spurred a considerable amount of research interest in the past, mainly for the game of chess (Schaeffer, 1986;Berliner et al

Monte Carlo Tree Search in Lines of Action

“…In the following experiments each match data point represents the result of 1,000 games (unless otherwise specified), with both colors played equally. A standardized set of 100 three-ply starting positions [16] is used, with a small random factor in the evaluation function preventing games from being repeated. The thinking time is 5 seconds per move (unless otherwise specified).…”

“…For instance, the four programs MIA (Maastricht In Action) [14], BING [15], YL [16], and MONA [16] are example of strong LOA programs. Since 2000 LOA has been played seven times at the Computer Olympiad, a multi-games event in which all of the participants are computer programs.…”

“…This is an important milestone for MCTS, because up until now the traditional game-tree search approach has been considered to be better suited for LOA, which is a highly-tactical slow-progression game featuring both a moderate branching factor and good state evaluators (the best LOA programs use highly sophisticated evaluation functions). The previously best game-playing programs for this game, MIA [14], BING [15], YL [16], and MONA [16], are all αβ based.…”

GTQ: A Language and Tool for Game-Tree Analysis

“…The search engine and the position evaluator, the two core parts of any such program, become quite sophisticated when all the necessary bells and whistles have been added [8,18,7,1]. To maintain and enhance such complicated software is a challenging task, and the risk of introducing bugs or other unwanted behavior during modifications is substantial.…”

Informed search in complex games

“…For LOA, it is a challenge to build such an evaluation function, since it should incorporate the basic principles of the game and simultaneously increase the profitability. The difficulty lies in the fact that knowledge about LOA evaluation functions is not well developed, although some material on the basic principles has been published recently (Handscomb, 2000a(Handscomb, , 2000b(Handscomb, , 2000cChaunier and Handscomb, 2001;Billings and Björnsson, 2003).…”