[学术讨论] 求教哪些定理公式或檢驗是被華人命名的？ [复制链接]

whateverittakes
UID: 3477615

初级会员

Rank: 3 Rank: 3

声望: 74
寄托币: 353
注册时间: 2013-10-10
精华: 0
帖子: 47

发表于 2013-12-17 07:20:27 |显示全部楼层

chrishacker 发表于 2013-12-17 06:50
你是说LMS这种不算Nonparametric? 反正我老师认为算. 你认为不算那我没办法。分类这种事，本来就很主观。 ...

Please explain why LTS is nonparametric. Also, please explain how you can use the analysis of robust regression to conjecture the usefulness of nonparametric regression.

26Fall祈福求offer，包灵验！

使用道具举报

chrishacker
UID: 3338447

高级会员

Rank: 5 Rank: 5

声望: 79
寄托币: 1778
注册时间: 2012-7-4
精华: 0
帖子: 432

发表于 2013-12-17 08:32:15 |显示全部楼层

本帖最后由 chrishacker 于 2013-12-17 08:33 编辑

Deleted

双非DIY美硕申请总结

使用道具举报

chrishacker
UID: 3338447

高级会员

Rank: 5 Rank: 5

声望: 79
寄托币: 1778
注册时间: 2012-7-4
精华: 0
帖子: 432

发表于 2013-12-17 08:32:45 |显示全部楼层

本帖最后由 chrishacker 于 2013-12-17 08:41 编辑

whateverittakes 发表于 2013-12-17 07:20
Please explain why LTS is nonparametric. Also, please explain how you can use the analysis of robu ...

In my course website there is title called "nonparametric regression" and two sub-titles: One is called robust regression and another called smoothing.
If you think only smoothing is nonparametric, OK, I can accept that. Names are not important.
You can just simulate a non-normal error to do the simulation. And be sure that sample size shouldn't be too large. Or if CLT applies, then OLS should be good. Then generate another group of data to see which one predicts better or which one points to the correct model more often (these two are different under some circumstance).
LMS, as far as I recall, often do things wrong...

使用道具举报

whateverittakes
UID: 3477615

初级会员

Rank: 3 Rank: 3

声望: 74
寄托币: 353
注册时间: 2013-10-10
精华: 0
帖子: 47

发表于 2013-12-17 09:21:35 |显示全部楼层

本帖最后由 whateverittakes 于 2013-12-17 11:16 编辑

"In my course website there is title called "nonparametric regression" and two sub-titles: One is called robust regression and another called smoothing.
If you think only smoothing is nonparametric, OK, I can accept that. Names are not important.
You can just simulate a non-normal error to do the simulation. And be sure that sample size shouldn't be too large. Or if CLT applies, then OLS should be good. Then generate another group of data to see which one predicts better or which one points to the correct model more often (these two are different under some circumstance).
LMS, as far as I recall, often do things wrong..."

There is robust nonparametric regression (e.g. median coupling) but still I CANNOT believe that your uni Prof treated the whole robust regression subject as if it was a subcategory of nonparametric regression... I am done here.

使用道具举报