由买买提看人间百态

topics

全部话题 - 话题: bic
首页 上页 1 2 3 4 5 6 7 8 9 10 (共10页)
A*******s
发帖数: 3942
1
Subset Selection in Regression这本书讲了不少细节,不过我觉得内容有点老。网上
有电子书。
简单来说GLM的automatic selection都是基于likelihood的,一般来说是likelihood
ratio test with pre-specified significance level,当然也可以用AIC,BIC,
cross validation之类的。

leaps
forward
s*********e
发帖数: 1051
s******a
发帖数: 472
3
没有看懂
m*********n
发帖数: 413
4
我本来挺想仔细看一下的
但code是image的,分辨率低还撞色(黑色蓝色),
我上次提议的时候态度不好,但这次我想再试试
楼主能不能直接给code?在blog里。
s*********e
发帖数: 1051
5
对不起。
我想想办法吧。
w**********y
发帖数: 1691
6
深底色的可以试试这种配色方案:
http://studiostyl.es/schemes/son-of-obsidian
m*********n
发帖数: 413
7
多谢,
统计版还是颇有人愿意好好discuss一些非找工/考证/旋转专业的问题的。
就sas enhanced editor自带的颜色就挺好。
s*********e
发帖数: 1051
8
看看这样好不好一些?
p***r
发帖数: 920
9
is this SAS or matlab? I feel little bit confused
N******n
发帖数: 3003
10

这个是sas,
还有,一般都是选AIC低的,但是多少差别是显著差别,有的人说AIC是k^2分布,和这
个是一样的。谢谢
s*********e
发帖数: 1051
11
是matlab.
y*****w
发帖数: 1350
12
It seems survreg() in R and PROC LIFEREG in SAS run the same type of
survival analysis. However, when I ran both of them on a survival data, I
got different results. Both were set as exponential distribution, and have
right censored data. See below. Could anybody tell me why the results are
different? Did I miss specifying any important parameters in R? Thanks!
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The R code:
survFit <- survreg(Surv(time, event, type="right") ~... 阅读全帖
b*********n
发帖数: 1938
13
来自主题: Statistics版 - 新鲜热乎的面经
我猜是aic , bic之类。
A*******s
发帖数: 3942
14
来自主题: Statistics版 - 请教SAS高手关于lasso...
你这个seed是做cross validation用的么?如果想要确定的结果,数据量大的话就用
aic/bic,数据量不大的话就用press,不一定要用CV吧。
A*******s
发帖数: 3942
15
来自主题: Statistics版 - 请教SAS高手关于lasso...
press难道不是closed form solution of leave-one-out CV?
AIC/BIC也是asymtotically equivalent to CV的
s*********I
发帖数: 22
16
来自主题: Statistics版 - 请教SAS高手关于lasso...
哈哈,没完全明白您的意思,不过看上去很厉害的样子 :)刚学统计一个月,很多东
西还是云里雾里。 也就是说如果如果selection criteria 选择press 或 AIC BIC,
得到的效果类似CV,但是可以产生确定model (不需要seed)。 对吗? 谢谢!
c**i
发帖数: 234
17
来自主题: Statistics版 - help, 关于poisson regression
1.请问除了proc gendom外还有什么proc 可以用?
2.有8个independent variable,但基本都是category variable。我是应该直接用这些
category variable,还是把它们都分成dummy variable (0,1) 好?
3. proc gendom的output中,我应该怎么比较model?我没有看到任何关于AIC, BIC之类
的东西。是不是需要加什么option?
非常感谢!
a****g
发帖数: 8131
18
来自主题: Statistics版 - building prediction models from large dataset
i remember you ever had a piece of blog talking about BIC, can you provide
the link again?
Also, could you spend a few moments to briefly talk about this topic? Thanks.
a****g
发帖数: 8131
19
来自主题: Statistics版 - building prediction models from large dataset
i remember you ever had a piece of blog talking about BIC, can you provide
the link again?
Also, could you spend a few moments to briefly talk about this topic? Thanks.
l******h
发帖数: 855
20
来自主题: Statistics版 - 问个logistic model的面试问题
怎么判别两个logistic model那个好?
这么基本的问题。。。。我说Pass。。。。然后对方就一愣。。。。
另外,linear model的话是不是就根据AIC,BIC, R^2,Adjusted R^2之类的?
a*******y
发帖数: 105
21
来自主题: Statistics版 - 问个logistic model的面试问题
看Aic, bic,如果model 的变量不一样。如果是nested model, 可以做likelihood
ratio test. 参考 Roc. 可以再看看有无rank order break

★ 发自iPhone App: ChineseWeb 7.8
r********n
发帖数: 6979
22
来自主题: Statistics版 - 新人问个matlab统计方面的问题
这不就是做feature selection么
你可以google一下
方法多了去了
forward, backward, AIC, BIC。。。
你说的这个例子其实就是做一个F-test(anova)
如果用更少predictor的那个model能够得到和更多predictor的model相似的fitness
那就应该选那个predictor比较少的
一般来说这样产生的model更robust
如果你只是先快速检验一下那个model可能会更好一点
可以用这个方法
不过我个人不是太喜欢这么做
因为做出来的model在新的数据里面“不一定”更好
如果愿意多花些时间的话
我一般会先做cross validatiaon或者bootstrapping
重复N次
然后比较两个model在test set里面的fitness
然后选择那个fitness error更低的那个
优点是这样做出来的model一般来说更robust
缺点是花的时间可能是几十倍甚至几百倍
如果你的feature比较多的时候
可以先用uni-variate model, forward, backward等方法找到有限的一些... 阅读全帖
A*******s
发帖数: 3942
23
来自主题: Statistics版 - 为啥做了segmentation后模型fit更差?
明白了,其实根源是你不应该拿每个segment单独的performance,和segmentation之前
基于整个sample的performance来做比较,这个根本就是apple vs. orange。
要拿任何一个statistic(r^2, adj r^2, AIC, BIC, AUC, whatever)来指导
model selection,这些statistics都是对同一个sample得出来的才有意义。
f*******n
发帖数: 2665
24
来自主题: Statistics版 - 到底ROC能不能用来比较不同model
记得logistic regression training的时候老师说过C(就是ROC下面的面积)不能由来
比较不同的model。如果要比较不同的model,应该用AIC,BIC等, 可是SAS EM 中的
model comparison node就有ROC。 很困惑,请指点!
a******9
发帖数: 20431
25
方法太多了 aic Bic CV vif lasso/ridge/elasticnet...看你要干嘛了
g****e
发帖数: 1829
26
来自主题: Statistics版 - GBM in R
GBM一般都要cv。你做个regression都要f-test, bic, aic或者cv,何况gbm。有些经验
值,比如说shrinkage可以用0.001,但cv是标准流程。

发帖数: 1
27
OPEN POSITIONS in Statistics, Biostatistics, and Econometrics
Positions in:
Market Demand Analytic Team, Global Data Insight & Analytic, Ford Motor
Company
Location: Dearborn, MI
Position: A contract position through Sencom, Inc.
Position Type: Long term/permanent type
H1B /PR sponsor: Yes
Email: [email protected]/* */
Job Requirement
Position Description :
This position will develop analytical methodologies and specify data
requirements by translating business phenomena into appropriate... 阅读全帖

发帖数: 1
28
OPEN POSITIONS in Statistics, Biostatistics, and Econometrics
Positions in:
Market Demand Analytic Team, Global Data Insight & Analytic, Ford Motor
Company
Location: Dearborn, MI
Position: A contract position through Sencom, Inc.
Position Type: Long term/permanent type
H1B /PR sponsor: Yes
Email: [email protected]/* */
Job Requirement
Position Description :
This position will develop analytical methodologies and specify data
requirements by translating business phenomena into appropriate... 阅读全帖
E**********e
发帖数: 1736
29
来自主题: Statistics版 - three way anova
如果高级项统计上不重要,当然可以drop掉。如果某个高级项重要,但是相应的一次项
不重要,就要保留一次项。这是统计上基本的东西。还有可以看别的一些metric,像
Rsquare。减去或加上某一项,AIC/BIC变化大不大。

:y~ 1+ a+b+c+ab+ac+bc+abc

发帖数: 1
30
OPEN POSITIONS in Statistics, Biostatistics, and Econometrics
Positions in: Ford Motor Company: Market Demand Analytic Team, Global Data
Insight & Analytic,
Location: Dearborn, MI
Position: A contract position through Sencom, Inc.
Position Type: Long term/permanent type
H1B /PR sponsor: Yes
Email: [email protected]/* */
Job Requirement
Position Description :
This position will develop analytical methodologies and specify data
requirements by translating business phenomena into appro... 阅读全帖
a**i
发帖数: 1091
31
来自主题: Guang_Xi版 - 来说说black friday的收获
买了14瓶arm&hammer 45oz~50oz洗衣液,2盒russell stove 24oz巧克力,2把gillette
fusion razor, 2支bic ligter点火枪,花了20块
C***i
发帖数: 486
32
来自主题: DataSciences版 - Twitter Data Scientist 电面题目
已挂。补充下具体内容,希望对版友有些帮助。、
T家先来一轮ONLINE CODING,题目和LEETCODE类似,有时候有复杂度要求,但我运气比
较好写的代码混过去了。
大概过了一个月,本以为没戏了,RECRUITER A 联系我,然后过了2周RECRUITER B又帮
我安排和HIRING MANAGER 电面,大概45分钟。面试前问了下形式,没鸟我。
面试前网上搜了下题目,貌似都是和统计有关的。面试当天,一张嘴就发现是个印度人
,简单胡侃,介绍下简历里项目。问我Monte Carlo Simulation 和 BIC的步骤。接着
老印让我做一个ONLINE TEST,然后问我用没用过collabedit.com,我告诉他没有。于
是他就用印度英语给我讲背景,听的那叫一个费劲,只听懂大概。中途说道一半,
RECRUITER B发来链接 (后来RECRUITER B说面试前没听说要考我CODING,一半如果有
要求会提前发)。
题目是有3列DATA, 分别代表邮件发送者,接收者和发件时间。
SENDER| RECEIVER | TIMESTAMP
A B 56
A A 7
A C 5
... 阅读全帖
P*****6
发帖数: 273
33
如果用glmnet的话, 好像有elastic net. 它比lasso会保留更多的变量
无论lasso or elastic net, they are biased. 如果你的features小于100的话,AIC
and BIC for generally linear regression may be also OK.
H**********f
发帖数: 2978
34
来自主题: DataSciences版 - 一道药厂computational biology的面试题
这是个variable selection问题。这题应该没有标准答案,就看你能不能说出个123,
所以你得对这方面有所了解,但不见得上来就会做。你要说出一些关键词比如variable
selection,一些常见的可以做variable selection的分类方法比如forward/backward
selection,lasso regression,svm等,一些判断分类器好坏的办法比如AIC/BIC,
cross-validation。
如果深究这个问题:
https://en.wikipedia.org/wiki/Feature_selection
办法很多,还有不少专门针对gene expression的方法。

and
l****t
发帖数: 36289
35
来自主题: _Hope版 - 花果山装B指南(ZZ)
晴雨伞装逼指南
冬季雨雪多,一把好伞便成了最应景的装逼良品。国产伞里,杭州天堂算是头牌。
奈何价格便宜,这就违背了“只买贵的,不买对的”装逼原则。对于刚发完年底双薪的
初级逼来说,可考虑Burberry经典的秋冬款格子衬里长柄伞,有绅士范儿。如果是女生
,就选同样经典的格子袖珍伞,收起来刚好可以放在手袋里。考虑到晋江东石制伞业名
扬天下,加上Burberry山寨版早已泛滥成灾,你一定要保留专卖店发票,以备必要时验
明正身。可供选择的一线品牌还有LV、Gucci、CD等等——这只是装逼犯里的普通青年
,作为文艺青年,你还可以选择像是安逸猿 (Bape) 或Anna Sui这种设计师品牌,价位
由一千至三千不等,祝各位被打劫愉快。
如果是初级逼里的2B铅笔,又该选择什么呢?买把Materious设计的Sabre伞吧,这
玩意儿不贵(40美刀),看起来就是一把中世纪骑士用的军刀,可以背在身后。喜欢剑
的可以选Broadsword Handle Umbrella,cosplay一下阿瑟王。初级逼到此为止,他们
的共同特点是,对大牌的追求高于一切,或者沉迷于怪异的外形。
要扮演中级逼,就必须显示... 阅读全帖
l*******r
发帖数: 39279
36
来自主题: _Hope版 - 日本印象之购物篇
先写个简单的购物
过去总觉得日本的东西非常精致,现在看看普通商品相当一般,说明我鳖国跟日本的差
距缩小不少。当然日本好些小东西是made in china, 所以不精致很正常。Made in
Japan的东西最贵,当然质量确实会好,符合一分钱一分货这一颠扑不破的真理。
都说日本的药妆店如何琳琅满目,价廉物美,说实话,真没觉得多么好,我只是看得晕
头转向。在我看来,小日本喜欢把物品细分到变态的地步,搞得过于复杂,其实很多东
西用处不大,反而增加了生活的复杂性。
我自己基本没买啥东西,都是帮别人买的,去日本前还在网上搜了好多日本购物清单,
现在发现基本是bullshit, 照着清单买是多么stupid啊!不过北海道白色恋人确实是好
吃的。连锁店Bic Cameria网罗了所有中国人爱买的东西,应有尽有。
东京的银座让人惊叹,繁华却极其有序,热闹却并不喧嚣,这是日本的特点。三越百货
里有很多中国人疯狂采购护肤品。
哦,我还是买了一些东西,买的都是旅游纪念品,我太喜欢买旅游纪念品了,很可能不
少是made in zhejiang的。京都的旅游纪念品价格还是比较公道的,比机场同类商品便
宜不少,... 阅读全帖
c**********8
发帖数: 92
37
根据尚学路上提供的那个国际汇款地址,我去银行问了,好像需要一个在美国的中间银
行才能进行国际转账
可能每个银行的规矩不一样,我问的银行是每笔12刀的手续费,没有转账上限
支付宝不接受国际信用卡,所以最方便的方法是回国的时候开通一个网银,再用支付宝
支付
Beneficiary Bank: China Everbright Bank, Chao Nei Branch
(SWIFT BIC:EVERCNBJ)
Beneficiary: China Association of Social Work Children’s Hope Welfare
Committee
Beneficiary’s 21 digits Acct. No. 083502120100304558058
Tel:010-65279081
海外捐款方式
w*******y
发帖数: 60932
w*******y
发帖数: 60932
39
Click Here For Details:
https://www.penfed.org/productsAndRates/creditCards/3percentcashback.asp?
cmpid=email-ccestmt-1010
Purchase Consumer Electronics from Retailers such as:
ABC WAREHOUSE
ABT ELECTRONICS
AMERICAN TV
AT&T
BATTERIES PLUS
BERNIE'S AUDIO, VIDEO & APPLIANCES
BEST BUY
BEST DENKI
BIC CAMERA
BOSE
CASIOLANDIA
COMET RADIOVISION
CRUTCHFIELD
CURRYS STORES
DBUYS.COM
DEEPDISCOUNT .COM
DICK SMITH ELECTRONICS
DIGITAL RIVER
DIRECTRON.CO M
DIXONS STORES GROUP
EBAY
ECOST.COM
EL GIGANTEN
ELECT... 阅读全帖
w*******y
发帖数: 60932
w*******y
发帖数: 60932
41
2 DAYS ONLY -- Sunday / Monday...
Computer desks SKU 853580 or 857426 $30 instantly
Culley manager chair $60 instantly
7 Days -- Jan 16 to Jan 22...
Post-its 12 pack $12 100% Staples REWARDS BACK
M writing pads 6-pack $10 100% Staples REWARDS BACK
Staples brand plastic tote $8 100% Staples REWARDS BACK
Papermate 4-pack pens $5 100% Staples REWARDS BACK
Avery 8160 labels 750 count $10 100% Staples REWARDS BACK
Banker Box 10-pack $17 100% Staples REWARDS BACK
25-pack catalog envelopes $8.49 100% S... 阅读全帖
w*******y
发帖数: 60932
42
So I got two $25 GC for MEIJER from the BIC F-12 Subwoofer deal that's
burning a hole in my pocket. Surfed around looking for some deals and found
this bike. Of course it's made in china but for $68.63 shipped, u can't
beat it with a stick. Target has this bike for $129.99. This bike is
perfect as a one time use for that monster jump you always wanted to try.
Here's the deal. Item # S2760
Link:
https://www.meijer.com/user/add_item_to_wishlist.cmd?productId=145422&
productVariantId=181644&t... 阅读全帖
w*******y
发帖数: 60932
43
来自主题: _DealGroup版 - 【$】Staples 99 Cent Deals
99 Cent Promotion from catalog, limit 5 unless noted
-Catalog # next to item, use "Order by Catalog #" to get item
Staples Comfort Stic Grip Ballpoint Pens (606117-NY for Black) (606124-NY
for Blue)
Staples Stickies Page Flags (474649-NY for 125) (634770-NY for 40)
Bic Wite-Out Correction Fluid (521898-NY)
Staples Dual-Dome Pencil Sharpener (935650-NY)
Dixon Oriole Pencils (573932-NY)
Staples Hype Pen-Style Highlighters (412783-NY)
Staples Power Magnets (516098-NY)
Swingline Standard Staples 500... 阅读全帖
w*******y
发帖数: 60932
44
The HSU VTF-3 MK3 subwoofer is on a sale at HSUResearch.com. While it is
only $70 off normal price they only go on sale about 1x a year, and because
it is an internet direct company that doesn't use resellers you won't find a
better deal than this on this sub. This subwoofer will blow the doors off
any other subwoofer you typically see linked around here. Forget Klipsch,
forget energy, forget BIC. This sub is the real deal.
If any of you have heard of craigsub's rankings. He rates this sub ... 阅读全帖
w*******y
发帖数: 60932
45
Excellent sub for the money. Hasn't been in stock for this cheap in a while.
Link:
http://www.shopblt.com/cgi-bin/shop/shop.cgi?action=thispage&th
w*******y
发帖数: 60932
46
I posted this deal previously and it is now available once again, but a few
dollars cheaper.
Here are the details:
ZT-180 1Ghz Android Tablet with HDMI + Free Screen Protector + Free 5 day
Int'l shipping for $107.99:
http://www.mcbub.com/tp/4/?/ItemID=136480
.
Here's the normal page that you find when you search their site:
$134.99 + $30.25 shipping with no extras thrown in:
http://www.mcbub.com/item/UPad-ZT-180-Google-Android-2-2-Froyo-
If you buy it from their sale page, it's $67 less and you ... 阅读全帖
w*******y
发帖数: 60932
47
Many 50 cents items.
Bic Brite Liner $4 - $4 Easy Rebate ( free after rebate)
w*******y
发帖数: 60932
48
Klipsch bundles- RB-41 Theater System + Harman BDS SO/120 5.1 AV Receiver w/
bluray for $919 A/C + more. Receiver alone is $800 elsewhere
AcousticSoundDesign. com has a bunch of Bundles including Klipsch Home
Theater Systems and Harman Kardan BDS 5 SO/120:
http://www.harmankardon.com/EN-US/Products/Pages/ProductDetails 5 SO/120
5.1 AV Receivers with Blu-ray playback. (it has an integrated bluray player
; the cheapest I can find on this receiver alone is $799 direct from Harman)
Most of the bund... 阅读全帖
w*******y
发帖数: 60932
49
Accoustic Sound Design:
http://www.acousticsounddesign.com/company/bf.cfm
Black Friday Deals Today: Premier Acoustic PA-6AW Outdoor Speakers (Pair)
$99, Atlantic Technology System 2200 Home Theater System $629, Harman Kardon
BDS 5 SO/120 5.1 AV Receiver and Blu-Ray Disc Player $299 + More Free
shippingAtlantic Technology System 2200 Home Theater System:
http://www.acousticsounddesign.com/core/view_BigProduct.cfm?pid
$629Use code atdev
Atlantic Technology System 2200C Center Speaker:
http://www... 阅读全帖
w*******y
发帖数: 60932
50
Polk Monitor 40 Series II:
http://www.newegg.com/Product/Product.aspx?Item=N82E16882290202
- $109.99
Polk Monitor 70 Floodstand Speakers Series II:
http://www.newegg.com/Product/Product.aspx?Item=N82E16882290208
- $149.99
Polk CS2 Center Series II
- $99.99
Buy all for a total of $619.99 shipped if you live outside of California.
Pair it with the Bic PL-200 or the klipsch subwoofer.
These are the best speakers you can buy at this price range.
首页 上页 1 2 3 4 5 6 7 8 9 10 (共10页)