摘要
提出了一种新的Anytime分类算法,anytime averaged probabilistic under mutual information estimators(AAPMIE)。该分类算法能较好地适用于需要即时响应的在线业务。从信息论的角度认为每个属性所携带的信息量是不同的,对其他属性影响较大的属性应该具有较高的权限被优先选择作为super-parent参加分类计算,这样,有助于提高分类开始阶段的分类准确率,有助于在较少的计算资源下返回更好的分类效率。实验验证了该算法能在anytime分类的早期较好的改善分类效果降低分类的0-1损失错误率,伴随计算资源的增加,算法能进一步得到更好的分类准确率。
A new anytime classification algorithm that is anytime averaged probabilistic under mutual information estimators (AAPMIE) is proposed in this paper. This algorithm can be used in online application. This paper regards that attribute with the highest weight has the biggest average mutual information with other attributes, and this attribute should be super-parent first, because it can get better effect in early stage. The experimental result shows that the AAPMIE can get better result.
出处
《电子测量与仪器学报》
CSCD
2009年第3期99-104,共6页
Journal of Electronic Measurement and Instrumentation
关键词
贝叶斯分类
anytime分类
互信息
Bayes classification
anytime classification
mutual information