k*******d 发帖数: 1340 | 1 呼唤weekendsunny
我觉得要Profit不一定只是predict price movement吧,比如volatility, volume之类
的如果可以predict出来也好啊 |
|
w**********y 发帖数: 1691 | 2 Isn't this the tricky part of Ultra HFT? Welcome to share more insiders'
views :)
When people mentioned R^2, that might be assumed to be a rough estimate of
predictability of the signal/predictive model. Frankly, the said 0.04 R^2
was from 2012 several months'out-of-sample test, with a medium trading
frequency. Impossible to be replicated in real life. Or why I need to put
that into a paper...
lose
ago, |
|
m**********4 发帖数: 774 | 3 It is not out-of sample, of course.
Generally, it is not used for prediction. It is used for explanation.
你说的完全正确。学术界做statistics de人大多数不用自己CLEAN DATA。 我个人认为
,用SVM,NN等等都是次要问题。主要问题是找到features。 比如地震和癌症预测,我
还是相信自然届存在那些FEATURE的,但怎么构造出这些FEATURE现在还没有解决。我看
到这些price prediction 文章用到的features都是我们能想到的。。。
从这点来说,做NLP的人更接近实际一些。他们要crawl data,转换成数据,然后work
on it。
of |
|
k*******d 发帖数: 1340 | 4 呼唤weekendsunny
我觉得要Profit不一定只是predict price movement吧,比如volatility, volume之类
的如果可以predict出来也好啊 |
|
w**********y 发帖数: 1691 | 5 Isn't this the tricky part of Ultra HFT? Welcome to share more insiders'
views :)
When people mentioned R^2, that might be assumed to be a rough estimate of
predictability of the signal/predictive model. Frankly, the said 0.04 R^2
was from 2012 several months'out-of-sample test, with a medium trading
frequency. Impossible to be replicated in real life. Or why I need to put
that into a paper...
lose
ago, |
|
s******e 发帖数: 841 | 6 Thank you for replying.
I am not a stastics major. It is an engineering problem. I think first I
want to reduce the prediction error as much as possible. That's why I wanted
to try regression tree method. But it failed. My question is can I find a
method that can give me small prediction error and it does not matter if it
is hard to interprete the result.
坏. |
|
h*********o 发帖数: 151 | 7 楼上的大大,你是对的,我在fit的时候没有加 data frame, 改正过来以后就没有
问题了,不好意思,我对这structure不熟悉。
不过 predict 的结果是一个矩阵,第一列是x,第二列是z,是不是应该把这两列加
起来,才是fitted value啊?
还有就是stats下面的loess是一样的做法么?它的预测函数是predict(object.lo,data
),我试了一样的做法,把data frame都弄成一样的,但是结果却仍然是1000个Y。这
是怎么回事呢? |
|
n***p 发帖数: 508 | 8 I took the sas predictive modeler with enterprise miner version 6 exam a
couple of weeks ago. I passed with 87% correct, given the fact that I did
not finish a group of 6 questions, just filled in any answers.
Overall I think it is not very difficult, but without good preparation
passing the exam is also difficult.
I do not think this will be very helpful if you already have some working
experience. for new job seeker, this should prove that you have knowledge of
predictive modeling and sasEM.
I |
|
R******d 发帖数: 1436 | 9 做100次prediction就有100个prediction score,平均起来。AUC就是根据这个100个值
的平均
值来算的。
AUC好像还是受skewed data影响的。
这个有出处么,俺的印象中都是AUC和sensitivity和specificity的组合有关, is not
sensitive to imbalances in the data
像
吗? |
|
F****n 发帖数: 3271 | 10 A model that predicts.
Regression is one type of predictive models. |
|
a*******i 发帖数: 67 | 11 本人用SAS做regression model画了"R-square" 图,
但是默认情况下predicted values没保存到output,而只是给了个图。请问如何把
predicted values保存到output以方便进一步的data analysis呢? |
|
l******n 发帖数: 9344 | 12 regression training data set里面,有个categorical variable只有3个level
需要prediction的data里面有一个data,这个categorical variable的值不再这3个
level里面,怎么做prediction?
谢谢 |
|
d******e 发帖数: 551 | 13 精算偏传统的模型,大部分的做法都是 define好的,很多是一元的风险管理;
Predictive modeling以多元回归为主,在保险公司是新的技术。这个听起来简单,但
是具体的做法上有很多的trivia,包括dependent / independent variable的定义,
loss development, premium on-level,missing data permutation, binning,
outliner (usually big loss) processing,需要一些经验。跟精算是互补关系,一般
来说,精算师负责定价,建模的人负责协助(pure premium model),或者另外加一层
定价(loss ratio model)。建模的人还可以建市场相关的 response rate model来做
targeted marketing
Predictive modeling在保险公司地位尴尬,一方面工作清闲不加班,稳定性有保证。
另一方面,地位不如精算师,钱给得相对少而且涨得慢(现在据说多了一些),数据量
不算特别大(所以往往接触不到最新的... 阅读全帖 |
|
j****k 发帖数: 46 | 14 希望大侠们如果有合适的position,帮小弟refer一下,也欢迎各种建议!
本人现在在illilois一个小学校教数学(visiting assistant professor),因为家里
都是老师,所以以前一直是在往学校发展,没有做过实习,现在觉得自己还是不适合做
老师,决定往业界发展。 本科和phd都是数学,研究生有ms applied math, 统计的
minor, 期间也修了不少统计的课,学校也还不错。research比较理论,做pde的,最近
恶补了一下machine learning,上了stanford 的online 课程。
编程主要是matlab,sas, R,最近考了sas advance, SAS Certified Statistical
Business Analyst,对Predictive Modeling Using Logistic Regression,
Experimental Design了解比较多,time series也知道一些。
想找找Statistician/Data Scientist/Predictive Modeler/Ris... 阅读全帖 |
|
m******t 发帖数: 273 | 15 Hi,
I need to solve a problem about predicting values of some numerical vectors
by using other numerical vectors with all these vectors in the same vector
set, which is generated by one or more black box functions.
Given a vector space:
P =[S_1, S_2, …, S_T | Sk is a vector of q numbers, k = 1, ..., T]
Find a sub-group of vectors g = {S_d | d belongs to 1, …, T } and a
function h( S_d )
Such that
Difference between the value of h(S_g) and S_r is minimized
set g + set r = set P
... 阅读全帖 |
|
A*******s 发帖数: 3942 | 16 depends...
for AG models with time dependent variable, you have to also forecast the
future paths of these independent variables.
For frailty models, could plug in the empirical Bayes estimate of the
frailty term into the prediction.
And, more meaningful prediction is on the density rather than survival time.
can
time |
|
s********1 发帖数: 235 | 17 有什么模型能把linear regression model 和 time series model 合起来做
prediction 吗?现在有一些数据,是一堆产品,他们有一些自身的属性的数据,如网
上的ratings 等,他们还有time series 的销售数据,每一种产品有month by month
的销售值,现在要用这些数据做predictive modeling, 想到的方法有对自身属性的数
据可以做linear regression, svm 这样的模型预测,对time series 的销售数据,可
以用 time series 的模型预测,有没有什么模型能把这两种模型结合起来,用一个模
型考虑两方面,进行预测?多谢! |
|
c*****o 发帖数: 1702 | 18 I used to do this before but without taking into personality factor which I
believe is not a factor you can easily figure out. You can go to payscale.
com to find the starting and mid career income and also you can use census
for median income by career, median income by location and also inflation
for the last 15 years. With these data you should be able to get the salary
increase rate and give the prediction for a regularly employee's income. For
those people who go for their own business and... 阅读全帖 |
|
w*********y 发帖数: 7895 | 19 我们社会科学来说,先把可以影响PREDICTION的因素先分析
一遍,看他们是否会影响 PREDICTION,如果会的话,就
做为COVARIATE 来CONTROL。。。
另一种就是,收集数据的时候CONTROL。。。
如果说是怎么调整BIAS,应该是看你具体做什么样的数据分析吧。。。 |
|
w*********y 发帖数: 7895 | 20 我提自己的想法。。
1. 你首先要用GENDER来预计一下它的影响,如果没有影响,就很欢喜。
2. 如果有影响,你分别用男性和女性来运行一下你的PREDICTION,看
那个是更有影响。。
3.或者采用WEIGHT的方法,大概就是计算男女比例,把这个比例放进
PREDICTION中。。。。。(这个方法我用的比较少,所以你得自己看下书,
比例具体怎么算的。我用第二个方法更多)。 |
|
a********e 发帖数: 78 | 21 针对一个数据集建立回归模型后,如何计算 confidence interval 和 prediction
interval. R 里的predict.lm 能够提供这两个指标。这只是针对线性回归的。 但还
有其他
general的模型比如 cart,gam.感觉这两个指标应该是基于bootstrap方法的,但没找
到相关文献。 |
|
w**2 发帖数: 147 | 22 这题考的是evaluation metrics,像precision, recall, f1之类的。给的数据是
target variable是predicted target variable。
从confusion matrix入手。分析一下根据business model,是optimize precison还是
recall。然后具体怎么做。
为什么不是accuracy,你可以说accuracy可以很高,error很低的情况可能classifier
总是predict majority。
如果又多个model,你可以比较他们的auc score, indicating how well a classifier
separate the two classes
不要担心,多面几次就好了。 |
|
t*******1 发帖数: 135 | 23 所以只是看列顺序,而不是看feature name是吗?
另外,predict的时候用原本的data matrix还是用了sparse.model.matrix之后的
sparse matrix呢?我试了一下用这两个结果不一样。我model training的时候是用的
sparse matrix,是不是predict的时候也要用sparse matrix? |
|
m******n 发帖数: 453 | 24 training data一般都需要standardized
建立好模型之后
那么这时候去predict test data
test data也需要一样的标准化处理吗
是不是用原始数据的test data
可是,如果对test data也进行scale的话
predict出来的y,也是标准化的值啊,怎样recover到原来的值去 |
|
G***G 发帖数: 16778 | 25 so huge earthquake.
Japan, with its advanced technology, could predict it.
right? |
|
t**c 发帖数: 7480 | 26 偶尔用来治疗被西医predict搞不定的不也挺好么 |
|
|
p*****e 发帖数: 290 | 28 Other and past predictions:
Biomolecular motors: Vale, Sheetz, Spudich, Brady
Unfolded protein response: Walter, Mori
Soft lithography and microfluidics: Whitesides, Quake
Chaperonins: Horwich, Hartl, Lindquist, Ellis
Polymers: Frechet, Matyjaszewski, Wang, Willson
Electrochemistry/bioinorganic: Bard, Gray, Lippard
Single-molecule spectroscopy: Moerner, Orrit
Solar: Grätzel, Nocera
DNA synthesis: Caruthers
Next-gen sequencing: Webb, Craighead, Klenerman, Church …
Super-resolution optical mi... 阅读全帖 |
|
o**********e 发帖数: 18403 | 29 嗯。一般来说,在任何主流媒体NEWS:
有REPORTER骂老中吃狗,就直接骂捕杀海豚的日本人。
有REPORTER骂老中南海,就直接骂日本人霸占亚洲的罪行累累。
COMFORT WOMEN啊, 明成皇后惨死啊。南海6个PARTY,
多数在建岛,就拿老中说话。
有人REPORTER说老中SPY,就直接骂日本人偷BOEING,汽车
行业的SECRETS。
有人REPORTER说老中CHEAT,就直接骂印度人CHEAT。
有人REPORTER说老中BIRTH TOURISM, 就骂印度人H1B。
有人REPORTER说老中压迫藏独, 就骂印度军队强奸MANIPUR女人。
最后说:这些亚洲人太不像话了。
把这些阴老中的亚洲人全部绑架起来陪老中挨揍,
这样以后他们就知道不黑老中了。
PREDICTION: 只要TPP不过,每两天抓中国人
http://www.mitbbs.com/article_t/Military/43789881.html |
|
o**********e 发帖数: 18403 | 30 LOOKACAR: 西方抵制玉林吃狗肉,为什么不抵制日本吃海豚?
我认为说得很对。 老中要学习反击西方的舆论。
一般来说,在任何主流媒体NEWS:
有REPORTER骂老中吃狗,就直接骂捕杀海豚的日本人。
有REPORTER骂老中南海,就直接骂日本人霸占亚洲的罪行累累。
COMFORT WOMEN啊, 明成皇后惨死啊。南海6个PARTY,
多数在建岛,就拿老中说话。
有人REPORTER说老中SPY,就直接骂日本人偷BOEING,汽车
行业的SECRETS。
有人REPORTER说老中CHEAT,就直接骂印度人CHEAT。
有人REPORTER说老中BIRTH TOURISM, 就骂印度人H1B。
有人REPORTER说老中压迫藏独, 就骂印度军队强奸MANIPUR女人。
最后说:这些亚洲人太不像话了。
把这些阴老中的亚洲人全部绑架起来陪老中挨揍,
这样以后他们就知道不黑老中了。
PREDICTION: 只要TPP不过,每两天抓中国人
http://www.mitbbs.com/article_t/Military/43789881.html |
|
|
K*****2 发帖数: 9308 | 32 现在又应验了,天朝真是predictable啊 |
|
v*******n 发帖数: 8995 | 33 从全球收集了10000多病毒
包括石正丽的蝙蝠病毒
发现可以感染人后
predict项目就停止了
然后北卡小组和gilead以及美国军方进行药物筛选 找到了瑞德席位
★ 发自iPhone App: ChinaWeb 1.1.5 |
|
发帖数: 1 | 34 【 以下文字转载自 USANews 讨论区 】
发信人: ThinkHarder (ThinkHarder), 信区: USANews
标 题: Breaking: As Andrew Yang predicted Mayor Pete Dropped Out! Still Copying Yang
关键字: Andrew Yang,美国大选,候选人,访谈
发信站: BBS 未名空间站 (Sun Mar 1 19:49:08 2020, 美东)
https://www.youtube.com/watch?v=o_-2PK9ScT8 |
|
g********d 发帖数: 4174 | 35 http://www.bisnow.com/dc-legal/2012/12/14/prediction-doma-prop-
Arent Fox litigation partner Hunter Carter, who co-wrote a 171-page report
that helped convince the NY State Bar Association and state legislature to
endorse equal marriage rights, tells us he expects both DOMA and Prop 8 to
be struck down.
Hunter told us yesterday he thinks both Hollingsworth v. Perry and US v.
Windsor will likely go 6-3, finding Prop 8 and DOMA unconstitutional. The
four more liberal justices could be joined by Ro... 阅读全帖 |
|
U*E 发帖数: 3620 | 36 is it hard to predict the flood? |
|
m******n 发帖数: 6327 | 37 ☆─────────────────────────────────────☆
piyaoke (辟谣客-睡教大师) 于 (Tue Nov 18 14:55:24 2008) 提到:
先看看牛人的水平和信用。
Peter Schiff was right
再看看牛人对侯赛因经济的预测。
Peter Schiff predicts doomed economy under obama
☆─────────────────────────────────────☆
altopalo (城市猎人) 于 (Tue Nov 18 16:41:33 2008) 提到:
Best posting this week!
其实各个总统的行动都可以从他们以前的生活和行政轨迹中得到某种预测。奥巴马的参
议员投票记录显示,他是一个最liberal的参议员。再加上他背后主要的大金主工会,
很难想象,他能做出非社会主义的决策。所以美国注定要沉沦4年了。
美国人选举搞了这么多年,但是真正有政治智慧的选民还是太少了。
就让奥巴马做里根二代的垫脚石吧。
☆────────────────────── |
|
l****z 发帖数: 29846 | 38 Predictable: Low-flow toilets cause a stink in San Francisco
from theblogprof by n*****[email protected] (The blogprof)
Just another of the many stories from liberal la-la land where feelings
trump science, logic and reason. From the San Francisco Chronicle via
memeorandum: Low-flow toilets cause a stink in SF
San Francisco's big push for low-flow toilets has turned into a
multimillion-dollar plumbing stink.
Skimping on toilet water has resulted in more sludge backing up inside
the sewer pip... 阅读全帖 |
|
a**t 发帖数: 3833 | 39 I predict GOP will not be even close to presidency in the next 36 years. |
|
d***h 发帖数: 97 | 40 I created a complex method of my own to predict the battleground states on
2012 US presidential election. I used those latest poll data from
realclearpolitics.com, re-analyzed their reliabilities, MOE, independents (
for O, R, or other candidates). Here are my calculation results.
O R
Florida 47.77 50.53
Virginia 47.58 49.79
New Hampshire 48.69 49.04
Wisconsin 49.56 47.84
IOWA 48.81 48.59
Colorado 48.30 49.40
Nevada ... 阅读全帖 |
|
c*****n 发帖数: 14445 | 41 【 以下文字转载自 Seattle 讨论区 】
发信人: iq350 (iq350), 信区: Seattle
标 题: Dick's predictions
发信站: BBS 未名空间站 (Sun Nov 4 22:30:09 2012, 美东)
Mitt 要赢325 票
from foxnews |
|
m*********6 发帖数: 235 | 42 发信人: majia123456 (@ @), 信区: USANews
标 题: I predict O8 win another term
发信站: BBS 未名空间站 (Thu Dec 15 02:41:20 2011, 美东)
Just based on the prospective GOP candidates though O8 did not do very well.
No argument.
Let's just wait and see. |
|
l****z 发帖数: 29846 | 43 by Andrew C. McCarthy
June 30th, 2013 - 10:13 am
The state of Florida’s politically driven decision to charge George
Zimmerman with murder has resulted, as some of us predicted it would, in a
pathetically weak case. It has taken only a few days of trial to collapse of
its own weightlessness – undone, in fact, by the direct testimony of a
prosecution witness, as Bryan Preston relates at the Tatler and Ed Morrissey
details at Hot Air.
Over a year ago, I explained why this would happen:
When Tr... 阅读全帖 |
|
f*g 发帖数: 184 | 44 My common sense prediction is that Donald J. Trump will be the 45th
President of United States of America. 理由很简单:
Trump在初选时的困难可和登天的难度相当:几乎全世界的所有主流媒体,国际上现有
和过去重要领导人、政客们、美国国内两党首脑、参议院和国会代表们,在数亿美元支
持的铺天盖地的反面媒体广告的助攻下,惊人一致地联手反对、攻击、诬陷一个毫无从
政经验的总统候选人。在这样极端挑战的条件下,Trump knocked out 16位共和党参选
人,于5/26/2016提前clinched the GOP nominee。
相反,Hillary几乎一直只有一个竞争对手,到现在也不能seal the deal。稍有常识的
人都知道,到GE,Trump和Hillary一对一竞争时,一定会sweeping victory。一个不太
恰当的比喻,就是体育比赛。Hillary和Trump比较,就是low energy,她都不敢和
Sanders debate,到时遇到Trump,一定... 阅读全帖 |
|
c********i 发帖数: 1489 | 45 This is an excellent post.
I predict all four indices (SPY, QQQ, DOW, IWM) will end in RED at end of 8/
1/2016. |
|
|
k****g 发帖数: 1509 | 47 I predict that Killary will stress so hard during the debates when Trump
calls her out on all her shit , she'll probably have a seizure and end up
flopping all over the stage like a flounder ! |
|
|
|
m*****w 发帖数: 549 | 50 这个Harvard Law Professor, 是CNN主要的legal consultant。个人挺佩服这个人的观
点。一切以法律为准。
当时他看了律师的hearing,他直接表示Trump will lose appear 因为政府的律师表现
很差。没有回答关键问题。
但是他有另外一个prediction, 政府会在最高法院win,主要是Robert 大法官 是绝对
不能容忍standing issue。也是他,建议要有一个新的order。他指出和Trump同样的观
点,这个判决是Political, not Constitutional。
政府应该把辩论的律师作为主要问题。 Simpson案件就是输在这点上。
CNN还有一个一直穿红衣服的律师,真能说,到现在都不明白她的立场到底是什么。 |
|