基于专家系统与DAN网络的围棋局面判断算法

Situation of Go Position Judgment Based on Expert System and DAN Network

分析围棋局面形势判断特点,选用专家系统模式库、影响函数给出特征值,将特征值传递给深度适配网络(Deep Adaptation Network,DAN),利用已有盘面和结果标签训练网络,生成完整的盘面判断评估模型.通过1个公开数据集及1个私有数据集进行网络训练,得到8层、迭代200次左右的最优评估网络.实验结果证明了本文算法的可行性和有效性,解决了当前围棋评估算法参数难以训练、优化的问题.

This paper analyzes the characteristics of the situation judgment of Go,and selects the patterns and influence function in the expert system to give the characteristic values.Combining the features and passing them to the Deep Adaptation Network(DAN),we use the existing disk and result labels to train the network in a supervised learning mode to generate a complete disk judgment and evaluation model.Through a public data set and a private data set,we can get an optimal evaluation network with 8 layers and 200 iterations is obtained.The experimental results prove the feasibility and effectiveness of the algorithm,and solve the problem that the current Go evaluation algorithm parameters are difficult to train and optimize.