- UID
- 301194
- 热情
- 212
- 人气
- 705
- 主题
- 9
- 帖子
- 196
- 精华
- 3
- 积分
- 575
- 分享
- 0
- 记录
- 0
- 相册
- 0
- 好友
- 1
- 日志
- 0
- 在线时间
- 800 小时
- 注册时间
- 2011-11-2
- 阅读权限
- 20
- 最后登录
- 2016-3-29
升级 15% - UID
- 301194
- 热情
- 212
- 人气
- 705
- 主题
- 9
- 帖子
- 196
- 精华
- 3
- 积分
- 575
- 阅读权限
- 20
- 注册时间
- 2011-11-2
|
必须computing 2门选一:
779 Professional Computing Skills for Statisticians
782 Statistical Computing
我选了782。
两门课今年都有不一样的改变。782评分标准怪,但没有人不合格,779出现有人不合格。
先说说选择的标准:觉得统计只是一门工作的工具的话,选779;想成为专业的统计从业人员,782。
原因十分简单,现今先进的统计方法不断更新,故有的统计软件SAS,SPSS,根本无法兼顾及开发computing的工作辅导这些新的统计方法。782能教你R最核心的部分。782能在学校等部门提供你更多工作机会,例如research assistant,更是挑战783,784必须的支持。
普通办公室文职,或想专心在SAS的软件上的,779足够了。(十分多不同类型的软件介绍,对自身自学能力有信心的朋友还是782吧)
在老师心目中的难度,该是779是782的准备,尽管两门课并restriction。
需要说明的是,两门课都有不同程度的赶不上当今statistics computing的节奏。
Advanced computing:
如果710/723是理论的恶心,那这两门课就是贯彻始终的恶心。难,但值得读。基本上以上所说的所有介绍,但这里都是教大家如何准备挑战这两门课程的。范围极度的广,官网介绍十分不靠铺。在国外不同学校,因应该学校不同擅长范围,而把782的内容与783或784其中一门相结合。简单的说,783偏theory statistics,784偏applied statistics。
具体内容每年都会因应课程时间有变动,详情请联系该课老师。
783 Simulation and Monte Carlo Methods
最轻松准备:731,730,782,320,325,310,maths270 (这样配置应该很轻松了。)
可以称为足够准备:730/731,320/325,maths270,380/782
傻B挑战配置:320,310
授课相关内容就有以上“最轻松准备”的相关内容。那样就会轻松很多,那这门课就是一门教你怎么提高算法效率的computing课程,因为theory都准备好了。老师相当能人服气。学生作业在R走最快1分钟的项目,他设计了33秒。。。从generating random number& test, bootstrapping, model-selection, monte carlo method(从此统计学生再也不用担心不会求复杂function的积分咯。),markov chain monte carlo method find optimal parameters。
当时我是傻B挑战配置,可悲摧了。。。。。。。。难听点形容,别人一个dissertation programming 部分就已经等于一个assignment。这个例子有点夸张,但实际上,上这门课的人,高手云集,phd至少5个(当时),还要ranking,还只有我一个中国学生,完全没有什么抱团抄往年作业。。。没有。。。老师自身对自己教theory十分有信心,本人觉得至少把731上了,上这门课才能学得value 的东西。
784 Statistical Data Mining
一定需要读了782先 。782,784一起读会作业量有点大。
Coreqs: STATS 310/732, 730, 782 绝对bullshit
更加偏向CS以及统计的结合。前部分花相当大笔墨介绍相关的电脑部分。后部分讲data mining中statistics有关的部分,正如题目上写的。
还没读。还不能给出中肯的答案。老师给分十分严格/严谨(第一章就给你扣什么是数据挖掘这些字眼了,能不严谨吗),实际上今年782由yee教闹出了很多问题,明年应该他会卸任,但784的评价则没有任何积极或消极的影响,虽然因为他对我做的某些行为太过分,令我782成绩差,但还是很客观的说,他只是不熟悉782 的评分标准以及我确实太多小问题让他抓,784还是相当值得一读的。但是!如果有第二个老师教,我想我还是会选第二个老师。(例如,选CS那边的data mining)
BIOSCIENCE
BIOINF 704 Statistical Bioinformatics 2
讲述DNA,基因组学,考试大量背书内容。注意里面会有讲解 linear model 当(n《p)。之前的课程我们只会接触n=p,或p>n的线性模型,有志研究这领域的应该听听。
BIOSCI 738 Advanced Biological Data Analysis.
里面有302的部分内容加750的部分内容,而且它教的302的部分更容易理解。
读此门paper ,最体现其价值的是,当生物honours degree(本科阶段没有统计专业)时,读这门加上选2门统计(每个degree都能在项目里选最多2门非本专业课程),就能读统计了。(注意,学校是你能任何paper 的如果达到该paper要求的话,但要拿post的专业最低要求是不能读超过3张非本专业的paper,否则没有证书)这门课就是这样规避了这个风险而能从生物跳入统计。
MEDSTATISTICS
770 Introduction to Medical Statistics
没评价,反正要拿这个专业,都躲不开读的。
773 Design and Analysis of Clinical Trials
没评价,反正要拿这个专业,都躲不开读的。就是临床试验那种得到结果的推导。
读精算的人会想学survival model,里面有些许用到的数学,名叫time missing value。
761 Mixed Models
注意是mixed model 非mixture model 。Regression 及anova 的结合物,个人觉得。
http://en.wikipedia.org/wiki/Mixed_model
其实真的挺深的,但这课教你用软件SAS直接得到结果然后描述结果。但必须肯定的是读了340,750,761就能成为anova专家了,这确实在就业中非常的好,毕竟有多少公司真要求你推导一个新模型。
780 Statistical consulting
705 Official Statistics
只能说好玩,内容题目写得很清楚了。
没有读,但没有任何负面的评价,或任何课需要读了它们才能继续。
终于来到私人珍藏系列,phd 系列课程!!!
负责人其中之一是James Curran
there are two courses, one in each semester that are being offered by the department for those students who are either in their provisional first year of a PhD, or who are intending to go overseas and do a PhD. The courses are STATS 702 - which is statistical theory, and STATS 763 - which is applied statistics. They are 700 level papers as you note, but they are also taught at a higher level, because, as I said they are for people who are doing a PhD.
Stat 736 Advanced Applied Statistics
Aim
This course is an advanced course on the primarily on theory and application of
modern regression techniques. The course is aimed at those students enrolled in
a provisional year of a PhD, or for those intending to undertake further graduate
study outside of New Zealand.
Lecture sessions take place in the departmental meeting space. The lecturing style
will be informal and discussion will be encouraged.
Course Topics
The course covers the following topics:
_ The uses of regression
_ Specialised methods for prediction
_ Smoothing
_ Bootstrapping
_ Censoring
_ Generalized linear models, sandwich estimation
_ Mixed models
_ Applied Bayesian modelling
Date Lecturer Topic
26 July
Lumley Regression - what is it for: Prediction and inference
August 2, 9
Lee Prediction: CART, regularisation/averaging, bagging, lasso, boosting.
August 16
Yee Smoothing: linear and cubic regression splines, psplines as regression
splines with regularisation, kernel smoothing.
August 23
Giurcaneanu Bootstrapping: A general approach to get standard errors for almost everything.
August 30
Lumley Censoring and related issues rightcensoring of time to event, right and/or
left censoring of assays, left-truncation of time to event. Difference between death and censoring . Competing risks and identifiability. The Cox model, parametric models for censored data.
September 20
Curran Generalised linear models (as semiparametric models for E[Y jX]). Logistic
regression, Poisson regression. Sandwich estimator as an approximation to the bootstrap for overdispersion or misspecification. Interpretation of coefficients, especially interactions.
Distinction between confounding and non-collapsibility in nonlinear models.
September 27, October 4, 11
Triggs Mixed models
October 18, 25
Meyer Applied Bayesian modelling
Stats 702 THEORETICAL STATISTICS
In which we learn about basic mathematical statistics, bringing the traditional course up
to date by considering estimating functions more general than the score and by mentioning
some results for dependent data and in_nite-dimensional objects. Some of the results will
be proved in detail, so that the student becomes acquainted with the techniques of proof.
Others will be stated formally and used in proofs. A few results will be stated informally,
because a formal statement would require mathematics beyond that we can assume.
The course will be o_ered in _rst semester. Each week, students will be given reading
and exercises to do, and these will be discussed in the class session. There will be several
instructors. The class will meet on Fridays from 3pm not quite until 5pm, in the 3rd-oor
Statistics conference room (303.310) except on May 24, when we will be in the 4th-oor
conference room (303.412).
There will be a take-home exam.
(1) March 8, David Scott. Of random variables. Distribution functions, probability
mass functions, measure as notation combining discrete and continuous cases, ab-
solute continuity, the existence of densities
(2) March 15, David Scott. Of the modes of convergence. In distribution (de_ned by
convergence of expectations), in mean, in probability, almost surely. Almost sure
representations demonstrated in the univariate case by the Skhorokhod construc-
tion and asserted in general.
(3) March 22, Alan Lee. Of the limit theorems. The LLN and CLT in iid cases.
The Kolmogorov LLN. The worlds simplest ergodic theorem (LLN with summable
correlations). The statement of the Lindeberg{Feller CLT and some motivation.
(4) Easter break
(5) April 5, Alastair Scott. Of transformations. The continuous mapping theorem and
the delta method. Handwaving for in_nite-dimensional parameters.
(6) April 12, Alastair Scott.Of likelihood The Neyman{Pearson lemma. Su_ciency.
The unbiasedness of the score equation. The information equality.
(7) April 19, Alan Lee. Of consistency in one dimension Monotone estimating func-
tions. Convex likelihoods. Local consistency for more general cases
(9) May 3, Alan Lee.Of asymptotic Normality in one dimension The argument from
smoothness, following Cramer but not restricted to likelihoods and with plug-in
LLN and CLT.
(10) May 10, Thomas Lumley. Of e_ciency The Cramer{Rao bound. The e_ciency of
the mle. Statement of the asymptotic su_ciency of the mle. Bias/variance tradeo_
with Bayesian estimators. The James{Stein estimator
(11) May 17,Thomas Lumley. Of asymptotic e_ciency The complications of asymptotic
e_ciency: the Hodges estimator. The statement of the convolution theorem and
the local asymptotic minimax theorem.
(12) Room change. May 24, Thomas Lumley Of consistency in more dimensions
Wald's method. Convex objective functions.
(13) May 31, Thomas Lumley Of asymptotic Normality The classical smoothness ar-
gument, the statement of the empirical process result for functions Lipschitz in the
parameter.
(14) June 7, Alan Lee Of the bootstrap The argument for functions of means
Ok, 总结,很多往届的人会有抱怨奥大统计怎么不match世界潮流,但其实它们一直有在努力,特别是一些非金融统计的领域。由于经常要转换岗位,又或者写着写着别人有问题来问,那就会写着写着跳了或者写着中文又变英文,并非真心我所愿。希望这资讯能帮到大家。需要更多资讯请留言,我会尽量配合采集进行补充。 |
|