h***i 发帖数: 3844 | 1 okay, p: easy to explain, assumption is stronger. np not easy to explain,
assumption is weaker. |
|
|
v*******e 发帖数: 133 | 3 职位是Marketing Analyst
Base 120K差点
Bonus 15%
RSU 105K over 4 years
No sign in bonus
我:不是engineer, 统计的master, 有7年工作经验。 这个Base和其他中小型公司同类
职位比也不算高,一个我不是engineer, 另外工资是比较我目前公司的pay, 当时想先
搬来湾区所以take了目前公司的low pay。
发个面经给去面试的人参考一下时间流程. 从recuiter联系我到offer一共六周时间:
07/21 Apple recruiter发邮件问我对一个职位有没有兴趣
07/22 Apple recruiter phone screen
07/28 phone interview with hiring manager
08/04 On site interview, 一共三小时,包括recruiter面6个人,每人半小时
08/05 hiring manager回复我的thank you letter, 说所有的人给我的评价都很好,
recruiter会联系我
08/11 recui... 阅读全帖 |
|
s******8 发帖数: 102 | 4 Senior Statistical Programmer – Full Time (San Francisco or Greater Bay
Area)
DESCRIPTION:
Serves as a lead programmer for assigned clinical study projects and is held
accountable for all expected deliverables. Responsible for the development,
validation, execution, maintenance, documentation, and archival of analysis
datasets and programming code used in the analysis of clinical trial data
targeted for regulatory submission. All programming must be consistent with
good programming practice and ... 阅读全帖 |
|
C******n 发帖数: 284 | 5 As far as I understand, you have to estimate a parametric model to predict
future survival time. |
|
y**3 发帖数: 267 | 6 Thanks! I got the point! I found out that some frailty models such as in R
package Frailtypack use parametric method to projection into future. not
sure sas yet |
|
w*******9 发帖数: 1433 | 7 其实google一下就会有很多信息。我这给一个茶馆闲聊级别的解释。
Bayesian: 在经典的参数估计里,参数(比如A)被当成一个固定的数,一般用极大似然
来估计;Bayesian的起源是想利用已有的信息(prior information)结合观察到的数据
得到"更“准确的信息。比如在你观察数据前就知道A大概位于3-4之间,那么有理由相
信这个prior information会使得你的估计更精确。具体到实现上,就是你得假设A是个
随机变量并且服从某个分布,比如[3,4]上的均匀分布,在结合数据的conditional
likelihood, 可以算出A的posterior分布。从统计上来讲,你知道了A的分布,你就知
道了A的一切信息。比如可以用posterior distribution的mode or mean作为A的点估计
,也可进一步根据quantile得到credible interval。
Non-parametric Bayesian: 我的理解是在A是个函数时(比如A就是个未知的分布函数),
这时我们要指定这个函数是怎么分布的(比如我提出这个随机函数可能取值于某个大... 阅读全帖 |
|
d******e 发帖数: 7844 | 8 我认为是Linear in predictors.
如果认为predictor不包含高阶项,那Polynomial Regression就可以被看做是
Nonlinear Regression。
如果认为predictor含高阶项,那Polynomial Regression就可以被解释为Linear
Regression。
个人觉得说到底还是文字游戏,这种简单的polynomial regression本质上还是
parametric regression。 |
|
w**********n 发帖数: 55 | 9 breslow estimator for the cumulative baseline hazard. For smoothed baseline
hazard i think you need nonpara smoothing.Or use parametric estimation like
Weilbull. |
|
A*******s 发帖数: 3942 | 10 Any one has experience? I don't have the data yet but just wanna decide the
modeling approach.
How about R's capability on non-parametric regression model?
Many thanks. |
|
n**********0 发帖数: 66 | 11 如果normal,就可以t,如果不normal,就non-parametric.后者是比median |
|
y****d 发帖数: 432 | 12 ****************************************************************************
1)这些资料我以前也分享过,但是因为一些原因我的google博客被关闭了,我就不能
维护下载链接了。现在重新整理出来,免费分享给需要的人。
2)这些资料整理在www.svipbook.com不用担心,都是免费下载的。
3)如果大家需要在人大经济论坛下载任何文件,可以跟帖,我在有时间的情况下给你
下载下来然后传到你能下载的网盘比如dropbox or mediafire
4)如果您认为对您有帮助,请转载这个帖子。谢谢
****************************************************************************
★【1】牛人整理的统计学资料
list
1. AdvancedCalculus with Applications in Statistics
2.A History of Probability and Statistics and Their Applications... 阅读全帖 |
|
J*X 发帖数: 1001 | 13 你叫这个id也会来问这种问题?问的有点笼统,non-parametric bootstrap? |
|
|
m******r 发帖数: 1033 | 15 我原来翻过本书,比较难。Extending the Linear Model with R: Generalized
Linear, Mixed Effects and Nonparametric Regression Models, Second Edition (
Chapman & Hall/CRC Texts in Statistical Science)
你要是从来没用过R, 估计上手比较难。
不过既然对所谓assumption不满, 干脆上树,森林。
你永远不用检查任何assumption |
|
a****e 发帖数: 150 | 16 link function是 link function. GEE其实就是一种semiparametric的方法. |
|
a****e 发帖数: 150 | 17 link function是 link function. GEE其实就是一种semiparametric的方法. |
|
a****e 发帖数: 150 | 18 link function是 link function. GEE其实就是一种semiparametric的方法. |
|
j********t 发帖数: 201 | 19 自学统计硕士可能吗?
最近有不少非统计专业的硕士生,博士问:自学统计以达到统计的硕士水平是否可能?
当然可能。
你有两种选择:
1)网上的课:我记得有几个名学校有这些PROGRAMS, 而且很不错。
2)完全自学 - 这比较难, 但也不是难于上青天。关键是你了解好的统计系对硕士
学位的要求:学哪些课,读哪些书,达到什么样的水平。
以下是我学统计的课程供参考:
A: 必修课:
1)PROBABILITY
2)INFERENCE
3)EXPERIMENT DESIGN
4)APPLIED STATISTICS
5)REGRESSION
B:选修课:
1)Survival Analysis
2)Categorical Analysis
3)Non-parametric
4)Timeseries
5)Quality Control
6)Advanced Inference (Ph.D only)
7)Measure Theory (Ph.D only)
8)?
There are a few professors of statistics in this forum and I'm sure... 阅读全帖 |
|
j********t 发帖数: 201 | 20 自学统计硕士可能吗?
最近有不少非统计专业的硕士生,博士问:自学统计以达到统计的硕士水平是否可能?
当然可能。
你有两种选择:
1)网上的课:我记得有几个名学校有这些PROGRAMS, 而且很不错。
2)完全自学 - 这比较难, 但也不是难于上青天。关键是你了解好的统计系对硕士
学位的要求:学哪些课,读哪些书,达到什么样的水平。
以下是我学统计的课程供参考:
A: 必修课:
1)PROBABILITY
2)INFERENCE
3)EXPERIMENT DESIGN
4)APPLIED STATISTICS
5)REGRESSION
B:选修课:
1)Survival Analysis
2)Categorical Analysis
3)Non-parametric
4)Timeseries
5)Quality Control
6)Advanced Inference (Ph.D only)
7)Measure Theory (Ph.D only)
8)?
There are a few professors of statistics in this forum and I'm sure... 阅读全帖 |
|
G***s 发帖数: 10030 | 21 小弟学心理学(定量方向),基本都在和数据打交道,背景课上过:
比较基础的统计课加上multivariate analysis, non-parametric;
专业课上过一些social science的model:factor analysis, sem, hlm, irt等等;
minor上过的商学院的课database, web programming (.php), system analysis and
design, information system, 商学院讲的特别特别浅,唯一好处就是minor可以写在
cv上。
编程会sas, r, java本科学过2学期,准备重新开始学;
论文计划做优化,属于or的范围,正在学算法。
请问各位前辈。这样的背景,可以准备做ds么? |
|
m***r 发帖数: 359 | 22 机器学习日报 2015-02-27
@好东西传送门 出品, 过刊见
http://ml.memect.com
订阅:给 [email protected]
/* */ 发封空信, 标题: 订阅机器学习日报
更好看的HTML版
http://ml.memect.com/archive/2015-02-27/short.html
1) 【数据挖掘比赛入门,以去年阿里天猫推荐比赛为例】 by @小斯never
关键词:应用, 资源, 行业动态, 推荐系统
针对去年的比赛重新做了个梳理,希望更适合纯新手。写的不好的地方请留言哈,还在
修改的。 通过 @微盘 分享了一个文件: 数据挖掘比赛入门_以去年阿里天猫推荐比赛
为例.docx, 欢迎大家下载分享! [1]
[1] http://vdisk.weibo.com/s/duAzytstNk6eB
2) 【从用户,时间,位置和文本四个方面对tweets建模】 by @AixinSG
关键词:应用, 自然语言处理, 统计, 推荐系统, 信息检索, 袁泉
袁泉的文章 “Who, Where, When and What: a Non-pa... 阅读全帖 |
|
w*****n 发帖数: 375 | 23 就说non-parametric 好了。 你越用专业术语, 知道的人越少。 |
|
g****u 发帖数: 25 | 24 骑驴找马今天面了一个感觉牛逼哄哄的startup, 已挂,满满的挫败感。最大的感受是
平时做项目的时候machine learning的算法都是抓来用用,那个好用用哪个,但对算法
的理解都很肤浅,没有深入思考过。大牛们看看下面几个问题怎么回答才好?
1)第一个问题是我有一个项目用mape来evaluate模型,面试官问我为什么不能用rmse
,我说mape是客户要求的,所以我就没多想,我实在不好意思讲我当时瞎扯了点啥,
太打脸了
2)面试官问我favorite的算法是什么,我说没有,平时logistic regression和random
forest用的比较多,然后面试官就问我什么时候用lg什么时候用rf。虽然我理论上知
道一点各自的特点比如lg对outlier比较敏感,模型interpretability比较高, rf是non
-parametric的所以对outlier不敏感,不用担心colinearity的问题之类的,但用到具
体问题的时候,我就说不上来了,因为其实我是抓来都用用,如果两个算法
performance差的比较大,我就直接用好的那个。我临时就凭感觉说如果featu... 阅读全帖 |
|
发帖数: 1 | 25 文他们要类似data dictionary,如果不能拿到,即使做出model,纯粹只是数字,不能解
释任何商业上的东西。 如果有missing value,不同的feature要依据不同的规则来填补
。 尤其是parametric predicting, 有些算法是不接受null 值。 |
|
w*******y 发帖数: 60932 | 26 This has a FRONT AUX input & a FRONT USB input! The included Bluetooth
adapter uses the REAR AUX input. Ipod control and hands free calls with any
Bluetooth phone!!
This includes FREE Shipping, Crutchfield's lifetime support, detailed
installation instructions specific to your car, install kit including wiring
harness for your car and I even got a free pocket insert, since my 4Runner
is a double DIN.
Use this code when ordering, to get it down $20 more to $99.99, a total
savings of $50: 3A825
L... 阅读全帖 |
|
w*******y 发帖数: 60932 | 27 Spotted this today...
Link:
http://www.sears.com/shc/s/p_10153_12605_05757509000P?prdNo=21&
JVC KDPKR3002 - CD Receiver with 2 pair of 6.5" Speakers
Available Online or In-Store Pickup
Product Description
Jam to your favorite tunes with this package that includes a full feature cd
receiver and 2 pair of 6.5" speakers.
Variable Color Illumination
Front Aux-in
Wireless Remote Control
MP3/WMA Compatible
20 Watts RMS x 4 (MOS-FET) 50W x 4 MAX
Line Output RCA (2.5V) Selectable Rear/Subwoofer Output
3... 阅读全帖 |
|
|
m*z 发帖数: 2356 | 29 这个现象通常是对的,当然解释有多种,下面是两种合理的解释
解释一,任何推动individual stock和market上涨都是有成本的。当market涨到一定程
度后,必须有新的钱投入,才能继续上涨。如果新的钱进不来,经济形势再好,market
的高点也维持不住。2007年底中国A股下挫就是这个原因,虽然经济形势依然好,但是
没有钱接盘了。对于大股,象AAPL和GOOG,因为Mass太大,推动它需要的力量也非常大
(牛顿第二定理),这不是散户和小fund能推的动的,必须是大庄家。大庄家推到一定
高度,再向上推,要很多钱,这对于它们来讲是很大的风险。所以在上升的最后,庄家
要让散户来接盘,这是虽然还有新的钱近来,但是price上不去了。因此,常常是到顶
的特征。这时一般还应该伴随量较小。
小股票则不然,尤其是价钱在5块以下的小股票,一般的mutual fund是不愿意投的,因
为成本太高。因此,通常是散户自己在炒。这些stock,因为market cap不大,散户完
全能炒上去。因此,在大股到顶是,他们还会冲一阵子。因此,就造成了权重大的停滞
,而index还在上涨的矛盾现在。
解释二... 阅读全帖 |
|
f********f 发帖数: 290 | 30 关于3d modelling得。
我们学校没有Geometric Modeling for CAD Applications这本书,下个礼拜面试要问
关
于radial edge structure,还有Parametric-Space ( 2D Space ) Vs Model Space (
3D Space )相关的内容,有人知道哪本书好点么?
我现在手头只有两三篇paper提到了这个。太少了,不够系统。
不胜感激。 |
|
r***l 发帖数: 36 | 31 don't do filtering seperately along two directions.
google for mean curvature flow, it can be done without parameterization.
you can also look for modified version of it, which can keep features better,
especially since your shape is tube-like, it means the two principal
curvatures
are quite different (one being almost 0), so you have to take the
anisotropy into account. there are quite some papers on this.
needed
along
me
and |
|
m*****o 发帖数: 17 | 32 Hi,romel, thanks a lot for your suggestion. I guess the mean curvature flow
you suggested is level-set method related implicit surface. If my guess were
right, would the speed very slow? Thanks again.
better, |
|
r***l 发帖数: 36 | 33 what I was suggesting is mesh-based, but level set doesn't have to be slow
either.
It is indeed an iterative method, but it's not slow.
What you need to compute is a gradient flow, the graident of total surface
area, and on a surface, it means the vertex needs to follow (1/A_x)dA/dx,
you should be able to find the details for mean curvature flow on meshes.
flow
were |
|
m*****o 发帖数: 17 | 34 Hi, Romel, I really appreciate your input. Actually I didn't expect we could
find very detailed suggestions like yours on this board, and yours just
gave us some insights. Here I have some detailed questions. My apologies
that I asked too many questions and occupied you a lot of time. I'm not a
graphics guy and nobody in my group knows too much about it either, but we
have to do something in graphics to keep our jobs. :(
I believe mesh-based algorithm could be very fast, since we've done some
al |
|
r***l 发帖数: 36 | 35 http://www.multires.caltech.edu/pubs/gi2000.pdf
this is an old one, but should be ok.
try equation (13) for \nabla A
for area of a vertex take 1/3 of the area of triangles adjacent to the
vertex.
basically, you add this - \nabla A/A * dt to X_i iteratively, you can do
implicit integration too. but if you take a small dt, explicit method should
be ok too.
could
calculate
for |
|
m*****o 发帖数: 17 | 36 Hi, Romel, thanks a lot for the paper and your detailed suggestions. I'll
take a look at it. Take care.
Regards,
MRITRIO
should |
|