發(fā)信人: ihappy (hungry christmas), 信區(qū): DataMining
標  題: Re: 有些迷惑
發(fā)信站: 南京大學(xué)小百合站 (Sun Dec 29 23:53:01 2002)


hehe, from the perspective of data mining, such tricks are important.

but there is the NFL theorem. In fact, i think before study any specific
ML algorithm, NFL should be mentioned first.


【 在 GzLi (笑梨) 的大作中提到: 】
: really,
: The results are not only depend on the algorithms you used, but also on 
: the preprocess and proprocess of the data and the usage of the algorithm. 
: tricks also in great need.
: So I think a complete system of ML is to be grasped.
: 【 在 daniel (飛翔鳥) 的大作中提到: 】
: :    it should be noted that almost all UCI data sets are from real-world
: : applications. So, it is not the fact that they are 'far away' from
: : real world. Many methods can perform well on most of these data just
: : because these data sets are publicly availabe so that they have been 
: : deeply investigated by researchers around the world.


--

※ 來源:．南京大學(xué)小百合站 bbs.nju.edu.cn．[FROM: 128.61.49.82]