基于效用的结构语法的属性学习

Attributer Learning Based on Structure Grammar and Utility

无论是在机器学习还是在软件设计中,对问题的分析都是假定对概念属性已知的条件下展开的。本文采用假设对象是在结构语法的基础上,通过确定对象的领域和效用,用领域的条件和状态及其边际效用选择出学习的类。在一个对象类中,属性的选择学习用迭代前向逐步插入、迭代向后删除算法。对决策树的学习,设计了决策树遗传归纳算法学习。对于新构造属性,用属性效用的投影、积、差、值、最值、级数算法构造边际效用法。对不同的对象进行不同的表达式学习,并命名新属性,进而进行属性的数据类型有界限定,从而得到新属性。

For machine leaning and software design,dealing with and analyzing concept is based on the attribute of it. This paper hypothesizes the object is a structure grammar, which is determined by the domain and utility of the object, and based on the structure grammar,and chose out for learning by the terms and conditions in its domain and marginal utility. In the structure of a object, the learning chooser of attributes needs to use the Prior to Gradually Insert Iterative Method,Delete Iterative Algorithm Backwards Method and the Decision Tree learning, then design the Decision Tree Algorithm Genetic Summarized Method. For the attributes newly constructed, it learns different expressions to different objects by constructing Marginal Utility Method by projection, multiplication, division, subtraction, extreme value, maximum and minimum,Series method of Attributes＇ Utility, and gives a name to the new attributes. Then it restricts the type of attributes＇ data to get the new attributes.