论文部分内容阅读
落子预测是通过在所有合法落子位置选择最优落子点,以提高计算机围棋棋力的重要手段.为了提高落子预测的正确率,本文利用基于人工智能的模式相对频率方法从所有落子中选择一部分落子点,通过OWL-QN(Orthant Wise Limited-memory Quasi-Newton)算法对最大熵模型进行训练,使用训练得到的最大熵模型对相对频率较高的一部分落子点进行重排名来获得最优落子点.通过实验获得了20.58%的落子正确率,表明该方法对落子预测有着一定的指导意义.
In order to improve the correctness of the prediction of the accuracy of the LOPs, the method of pattern relative frequency based on artificial intelligence is used to select a part of the LACs from all the LADs , The maximum entropy model is trained by using OWL-QN (Orthant Wise Limited-memory Quasi-Newton) algorithm, and the maximum entropy model obtained by training is used to rearrange a part of razor points with relatively high frequency to obtain the optimal sub-spot. The correct rate of 20.58% was obtained experimentally, which shows that the method is of certain significance for the prediction of rapjon.