Strategy optimization for deductive games

This paper presents novel algorithms for strategy optimization for deductive games. First, a k-way-branching (KWB) algorithm, taking advantage of a clustering technique, can obtain approximate results effectively. Second, a computer-aided verification algorithm, called the Pigeonhole-principle-based backtracking (PPBB) algorithm, is developed to discover the lower bound of the number of guesses required for the games. These algorithms have been successfully applied to deductive games, Mastermind and “Bulls and Cows.” Experimental results show that KWB outperforms previously published approximate strategies. Furthermore, by applying the algorithms, we derive the theorem: 7 guesses are necessary and sufficient for the “Bulls and Cows” in the worst case. These results suggest strategies for other search problems.