在”证明“股票基于价格与成交量的技术分析是不可行的” 一贴中本人提供了证明的思路与文献今天,经长期思考,为它提供一更具实际价值的定理。。鉴于本人都是用英文学习这方面的内容,因此用英文来表述它更正确,
”For any infinite domain set, X, the class of all binary valued functions over X is not a countable union of classes of finite VC-dimension . It follows that whenever the domain is not finite, there exists no nonuniform learner with respect to the class of all binary classifiers (although for each such classifier there exists a trivial algorithm that learns it – ERM with respect to the
hypothesis class that contains only this classifier)."
这有几个重要的概念:“nonuniform learner" 其实质是‘可学习性’概念中最弱的一个。
'VC-dimension' 学习假设的复杂性的测定
”ERM“ 样本误差最小化
很对不起,不再证明它!!