摘要： 在机器学习领域，特征选择对于提高学习机器的性能和效率具有重要意义，但是当前特征选择算法普遍存在着具体实现独立性强、可扩展性差的问题，使得对多种算法性能的统一对比评估实施困难，算法的替换和扩展比较复杂。以面向对象的设计理念为指导，基于设计模式中的策略模式，提出特征选择算法工具库FSL的设计构想，通过将一些常用的特征选择算法按照策略模式进行包装，以便机器学习算法用户的使用，同时确保其较强的可扩展性。

关键词: 机器学习, 特征选择, 策略模式, FSL-SP

Abstract: Feature selection is very important in improving the performance of learning systems. Various feature selection algorithms greatly facilitate the research of the scientists from different disciplines, there is a common problem that those algorithms are implemented by different researchers. So it is hard for the users to integrate or compare those independent implementations of different programming styles and incompatible designs. The feature selection library on strategy pattern(FSL-SP) is conceived to solve the above problem. The FSL-SP encapsulates many popular feature selection algorithms under unified interfaces, while different strategies of one algorithm could be exchanged conveniently. This library will bring help to those machine-learning algorithms users. The FSL-SP itself has good extensibilities, and new algorithm can be added into the library easily.

Key words: Machine learning, Feature selection, Strategy pattern, Feature selection library on strategy pattern (FSL-SP)

中图分类号: 

• TP181