88.txt

This complete matlab for neural network

发信人: GzLi (笑梨), 信区: DataMining
标  题: Re: [转载] 请教：regression和classification有何异
发信站: 南京大学小百合站 (Sun May 25 22:20:30 2003)

【 以下文字转载自 AI 讨论区 】
【 原文由 daniel 所发表 】


【 在 cczhu (congcongzhu) 的大作中提到: 】
: Dr. Z.-H.Zhou 的一片paper提到“regression and classification have distinct cha
: racteristics”。但regression也可以通过classification得到近似解，而classificati
: on更是可以通过regression之后再进行离散化处理得到。这样看来，两类问题在原理上应
: 该是类似的。那Dr.Zhou指的不同在什么地方呢？
: 
: 更请daniel不吝指教，谢谢！


I guess he meant the nature of these two things are different. for example,
classification usually employs zero-one loss while regression does not.
since many good properties are not held by zero-one loss, the mathematical
tools for anlyzing classification seems fewer than that for regression.
it is hard to useclassification to approximate regression, even you come with
some king of smoothing technique, the result won't be encouraging. it is
possible to use regression technique to tackle classifiction problems, but
some kind of thresholding is necessary.

--
宠辱不惊  闲看庭前花开花落
去留无意  漫随天外云卷云舒

※ 来源:．南京大学小百合站 bbs.nju.edu.cn．[FROM: dialup61-118.nju.edu]
--
※ 转载:．南京大学小百合站 bbs.nju.edu.cn．[FROM: 202.120.8.48]