x***u
发帖数: 6421 | 1 Drude peak的weight？就是光电导的Drude峰的积分。 |
|
on
发帖数: 199 | 2 it's easy to get an exponential weighted moving average:
mean(t) = q * mean( t-1) + (1-q) * x(t)
where x(1), x(2), ... are the data points.
My question is: how to get an "unbiased" estimate for standard devidation? A
naive formula would be:
var(t) = q * var( t-1) + (1-q) * ( x(t) - mean(t) )^2
but I don't think this is an unbiased estimate, and my simulation says it is
not.
any one knows or a link to reference please? many thanks. |
|
l********s
发帖数: 430 | 3 ccs是什么？你这weighting到底要起什么作用？ |
|
h***i
发帖数: 3844 | 4 还不如把解释了多少variation作为weighting更make sense.
factor |
|
m******1
发帖数: 19713 | 5 I have a data set like this:
id x
1 0.254
2 0.564
3 0.875
4 0.689
.......
The weight for each observation is x/sum of x's. What is the best way to
calculate it?
Thanks. |
|
c*******o
发帖数: 3829 | 6 Here you go:
Proc sql;
create table two as
select id, x, x/sum(x) as weight
from one;
quit; |
|
S******y
发帖数: 1123 | 7 I have the following data set,and try to make predictions for the last two
records. PROC REG and weight are used to fit a linear model. |
|
w*******n
发帖数: 469 | 8 The weight var is supposed to work for the model fitting only, not the
prediction. |
|
E**A
发帖数: 85 | 9 自问自答：
options1=glmnetSet;
options1.weights=权重矩阵，或者用函数赋值
fit1=glmnet(X,Y,'gaussian',options1);
哈哈 |
|
D**u
发帖数: 288 | 10 首先纠正一下楼主，应该是在glmnet用penalty.factor option，weight option 是针
对observation的。
再者，有谁能share一下experience，证实用这种方法（ridge output to lasso）,确
实可以提高 model performance的么？
我得到的结果都很糟糕，根本不如直接用Lasso。 |
|
x******o
发帖数: 165 | 11 1),2)两个问题，向各位求教。目前正在做的project每一行数据的布局：
User ID （每个用户登录我们网站都有unique id记录）
Search (用户在我们网站搜索的次数，搜索得越多表明越喜欢用我们的系统）
Clicks (搜索关键词后点击广告的数量，我们按照per clicks向广告商收钱的，用户点
击得越多，我们收入越多）
Conversions (点击广告后继续点诸如“购买”，“付款”，“下载”之类的内容，转
化得越多，广告商收入越多，就越满意我们公司的服务）
Days_since_last_action (用户距离上次使用我们系统的时间，越短表明越喜欢我们的
系统，所以在公式中是减）
1）现在要建立一个quality score的公式来根据每个user最喜欢的搜索内容来安排相应
广告，得用上上述的变量,基本模型如下：
QS=a1*Searches + a2*Clicks + a3*Conversion - a4*Days_Since_Last_Action
但我的问题是不知道如何给a1,a2, a3, a4设值？
目前根据search, clicks, conv... 阅读全帖 |
|
B****k
发帖数: 188 | 12 For the scoring system, assigning a1 = 1 and a2 = 100 means that you value
the clicking 100 times more than just searching. You can have similar
explanation for a3 = 500. However, a4 is another time of data on which I
would focus. A recent date with a large number of searching and so on would
suggest the customer likes the service. Then, we can put customers into
different groups based on their activities. Once you have the group levels,
logistic model would help you build a score system.
Moreov... 阅读全帖 |
|
x******o
发帖数: 165 | 13 哇噻，很好的思路，谢谢你的回答，太感动了！
再追问一下：
对于第一个问题：
比如我根据上次登录我们网站的时间把user分为是否是frequent users (1/0, 是／否
）两类，取名为Y变量，然后该如何用logistic model build a score system? 是不是
Y=a1 search+a2 clicks +a3 conversions? 然后给每个用户分配一个expected
probability, 再根据这个probability排名所对应的关键词？
PCA (principle component) and cluster analysis 如何应该在第一个问题上？
第二个问题：
equal weight如何用在第二个问题上，即知道rank后抽前百分之几的名次，这个％如何
决定？
非常感谢！
would
, |
|
l*********s
发帖数: 5409 | 14 pls/RRR/SEM etc, but really, you don't have to worry about weights if
prediction is the business goal. |
|
s*******w
发帖数: 27 | 15 Thanks a lot! But Weights 也非常重要的。 俺家领导说， 这个模型的一个重要用途
是，客户要做ALTER SITE.意思就是， 比如他们想着重新装修一下， Apparence 从1到
5， 就指望着销售量增加.
:（ |
|
s********e
发帖数: 323 | 16 如果response是probability，加上weight以后不是有>1了？这种情况怎么办？
多谢！ |
|
|
|
p******s
发帖数: 229 | 19 proc means data= yourdata;
weight yourweight;
class yourgroup;
var your value;
output mean= median= out= youroutput;
run; |
|
x**g
发帖数: 807 | 20 Hi, I need some help here.
I used PROC SURVEYMEANS and I got the standard error of mean. How can I get
the standard deviation of mean?
Here is part of the output from PROC SURVEYMENAS.
Number of Clusters 91
Number of Observations 34572
Sum of Weights 34312.5015
The standard error of mean is 0.007918.
(BTW, the standard deviation from PROC MEANS is 0.725688)
Thanks a lot! |
|
|
p***r
发帖数: 920 | 22 what's the best way to choose weighting scheme while not to falsely incease
or decrease the significance level of the result. Any paper relevant ?
Thanks. |
|
p***r
发帖数: 920 | 23 im not doing variable selection. just trying to run a simple logistic
regression with weights. some observations are more important than others,
how can I show this in model properly? |
|
p***r
发帖数: 920 | 24 for PD modeling. Im wondering whats the weighting scheme we should put for
each of following simplified situations (assuming all other info. are
exactly the same)? and if and how the result and significance level going to
be different?
1. 100 loans with no size information.
2. 100 loans with size all equal to $1k
3. 100 loans with size all equal to $1MM
4. 100 loans with size differs between $1k to 1MM. |
|
|
T*****u
发帖数: 7103 | 26 我没太看明白题。weight是做啥的，根据啥cluster啊 |
|
T*****u
发帖数: 7103 | 27 如果是后者的话，我用mixture of gaussians做过一个类似的,数据的前期整理比较麻
烦；后来有些不规则形状分布的，我的做法是每一个点当作一个rbf, weight当做
function的amplitude，train to get width matrix（can be location dependent）
，然后地图上任意一点的值都可以算，得到一个hot map,然后再做segmentation。其实
就是一堆点用rbf smooth成一个面，然后用这个面作为估计。
非专业人士的非专业做法，仅供参考。 |
|
w******e
发帖数: 1621 | 28 你圈圈大小要是给定的，来个K-means应该可以，EM块， E step 注意用weighted mean
，
开始也想说GMM,但是1M行的话，谨慎表示不客观，用gibbs fit过 GMM, 到50K就受不了了
而且企业里用K-MEANs可能性大,
另求内推DS,SDE。本人主业stat phd, data visualization和machine learning 方向。
副业主刷c++ 和副刷java。 |
|
h*****7
发帖数: 6781 | 29 想耍酷，在黎曼空间定义distance
想绕晕，拟合weight，套贝叶斯
想忽悠，直接上MDS |
|
y*****d
发帖数: 82 | 30 简单? 把weight看成第3维(每个点的高度),在3维空间中做K-means, K-medoids 或其他
算法. |
|
c*********r
发帖数: 19468 | 31 这个问题以前说过很多次，weight distribution还是很重要的，但所谓50:50是没有实
际意义的
一部50:50的车，匀速过弯时即没有转向不足倾向也没有转向过度倾向，完全中性
但这样的说法潜在的前提是前后端下压力一样大，前后轮胎相同，前后悬挂因素相同等等
这一系列前提在现实世界里其实都很难成立，比如说，现实的跑车绝大多数前后轮胎首
先就不一样
另外一层意义是，前面说的匀速过弯，现实中也不成立，车要么是在加速，要么是在减速
加速时你希望落在驱动轮上的负载大一些，对于RWD跑车，也就是要求重量分布偏后一些
对于基于RWD的AWD跑车，这个问题没有那么突出，比如GT-R，重量分布偏前
但加速时动态重心会后移，前轮实际的负载就会下降到50％以下，但又不会过低，所以
依然能发挥很大作用
减速时你同样希望重量分布靠后，这样减速导致重心前移后，后轮仍有足够的负载来发
挥一部分制动作用
否则的话，前轮制动的负荷太大，后轮太小，一是影响制动效果，二是前轮制动容易过
热实效，也影响寿命
所以综合来说，即使是引擎在前的高端跑车，也都是通过前中置引擎后transaxle等手
段让重心尽可能靠后些... 阅读全帖 |
|
|
|
|
w*******y
发帖数: 60932 | 35 GUNS: Daniel Defense DDM4V5 LW Light Weight M4 Carbine complete rifle- $1100
+$6 Shipping
Grabagun.com has these back in stock, $1,099.77 + $5.99 shipping, no credit
card fees.
Probably the best deal on a complete mil-spec AR with hammer forged barrel.
Also in the LW profile, which lately has been quite difficult to come by.
Comes with a hard case, 1 30rd magpul pmag, MOE stock, vertical foregrip,
and 3 magpul ladder covers.
Link:
http://grabagun.com/dd-v5-556nato-16-blk-30rd-lw-bbl.html
|
|
w*******y
发帖数: 60932 | 36 Bowflex is having a March Madness sale (got this on e-mail)
Click here:
http://www.bowflexselecttech.com/bowflex_selecttech_us/products SelectTech 552 Dumbbells.jsp
for the 552 dumbells, add promo code MARCH to include the free stand and
free shipping. Seems to be a pretty good deal!
Stand - normal price $139
Shipping can be quite pricy as well.
Amazon is currently at $349 for just the weights.
|
|
D******9
发帖数: 2665 | 37 I do not believe FDA will approve any weight loss drug soon |
|
l*****d
发帖数: 1948 | 38 3 lbs less than original weight. It has been stabilized for a week. 12 lbs
to go |
|
r******m
发帖数: 5550 | 39 我觉得weight loss不难。。保持最难~~~~~~~~~~ |
|
s*****r
发帖数: 126 | 40 faint, weight lifting MM?
scary~~ |
|
f**d
发帖数: 768 | 41 这是一本计算神经科学的优秀著作，全文拷贝这里（图和公式缺），有兴趣的同学可以
阅读
如需要，我可以分享PDF文件（－－仅供个人学习，无商业用途）
From Computer to Brain
William W. Lytton
From Computer to Brain
Foundations of Computational Neuroscience
Springer
William W. Lytton, M.D.
Associate Professor, State University of New York, Downstato, Brooklyn, NY
Visiting Associate Professor, University of Wisconsin, Madison
Visiting Associate Professor, Polytechnic University, Brooklyn, NY
Staff Neurologist., Kings County Hospital, Brooklyn, NY
In From Computer to Brain: ... 阅读全帖 |
|
d**********g
发帖数: 2014 | 42 这个网站挺好的。
http://www.best9mm.com/compact9mm.html
我感觉先从轻的开始挑。
Compact 9mm Pistols for Concealed Carry
We didn't do a "best list" because your personal requirements for a
concealed carry pistol will not always be the same as ours.
The rules for the 9mm CCW list: Weigh 30 ounces or less, barrel 4.1 inches
or shorter, and cost less than $600 from a dealer. Most confrontations will
be at 7 yards or less, so any choice you make from the below list should be
good for concealed carry. Some of these compa... 阅读全帖 |
|
g*2
发帖数: 658 | 43 １ 想要减肥的人多少，真正在付之行动的又有多少呢？
More than two-thirds of adults in the United States are either trying to
lose or maintain weight. However, only 20 percent are both eating fewer
calories and engaging in at least 150 minutes of physical activity during
leisure time each week.
２ ＢＭＩ怎么算的？理想体重是多少呢？
BMI = body weight (in kg) ÷ square of stature (height squared, in
meters)
Overweight is defined as a BMI of 25 to 29.9 kg/m2; obesity as a BMI of ≥30
kg/m2. Severe obesity is defined as a BMI ≥40 kg/m2 (or... 阅读全帖 |
|
c*****3
发帖数: 1655 | 44 应大家的减肥要求，贴一个以前写给友人的读书笔记
我只有一点不同意Anderson，就是晚上吃东西
最近研究表明，至少要给身体10-12小时的fasting时间，否则长肉！
另外，心理上的改变只是减肥的开始，后来的饮食和运动增肌就有各家学派，因人而异了
祝大家减肥愉快，减肥成功！
作者William Anderson是个心理医生，在他二十五岁300磅左右的时候发明了这个
Anderson疗法，减了140磅，然后一直保持了二十多年至今。
有这样几个重点：
1. Mind会直接影响到body.比如抑郁症、疾病不治自愈等。在weight control方面，要
用mind 的conditioning 功能--"We are driven by repulsion, as well as desire
and craving. Conditioning is the primal power source...drives you to repeat
behavior that feels good, even if you wanted to stop".
具体来说，就是我看到并吃到健康的蔬菜... 阅读全帖 |
|
f*****h
发帖数: 4489 | 45 贴着玩玩比较一下 帕萨特的确鹤立鸡群
2017 Subaru Legacy
Curb weight 3,455 lbs
Peak force 18,533 lbs
Strength-to-weight ratio 5.36
2017 Honda Accord 15,907 lbs
Curb weight 3,235 lbs
Peak force 15,907 lbs
Strength-to-weight ratio 4.92
2017 Mazda 6
Curb weight 3,217 lbs
Peak force 16,776 lbs
Strength-to-weight ratio 5.21
2017 Nissan Altima
Curb weight 3,115 lbs
Peak force 16,485 lbs
Strength-to-weight ratio 5.29
2017 Toyota Camry
Curb weight 3,167 lbs
Peak force 15,148 lbs
S... 阅读全帖 |
|
p****u
发帖数: 22764 | 46 原文链接：
http://blog.wired2fish.com/blog/bid/63566/Five-for-5-Five-Bass-
Every-Angler-Should-Know
Learn five ways to rig bass fishing plastics to catch bass in any situaiton
We've had a bunch of requests to do a piece on how to tie several rigs for
bass fishing with soft plastics and when and where to fish them. We'll start
with this overview piece with diagrams on how to rig the plastics. We'll
follow up with underwater videos to show how the rigs look in the water and
we'll expand on these rigs w... 阅读全帖 |
|
w***n
发帖数: 9040 | 47 本文列举了一些人们常忽视三个的与减肥相关的因素。如果你坚持控制饮食与锻炼身体
，却没有获得满意的效果，那么你要做的就是考虑一下这些因素。
每一位试图减肥的读者都知道这个过程是多么艰难。直觉告诉我们，减肥应该很简单：
消耗比你摄入食物含有的更多的热量，但是减肥这个过程并不简单。虽然把握好摄入和
消耗的热量是减肥的基本要素，但是你还需要考虑其他因素。
本文将把目光投向一些人们常忽视的与体重控制相关的因素。这些因素与一个人减肥的
成功息息相关，可是，只有少数人重视它们。如果你坚持控制饮食与锻炼身体，却没有
获得满意的效果，那么你要做的就是考虑一下这些因素。
饮水与减肥
水是减少胃口的天然食品。如果你时常喝水，那么你就不容易经常感到饥饿，这是由于
大脑难以区分饥饿和口渴。除了压制食欲以及给你饱食的感觉，水还可以加快身体新陈
代谢速率。由于每个人对水的需求是不同的，专家通常建议每天摄入64盎司（大概8杯
）水。虽然你可以一整天不喝水，但是新的研究发现餐前喝上两杯水有利于减肥。正因
为人体无法分辨饥饿和口渴，有时你觉得饥饿的时候可以用水代替食品而避免热量的摄
入。
睡眠与减肥
你也许认为睡眠和减肥... 阅读全帖 |
|
x********u
发帖数: 15396 | 48 http://ow.ly/d/19hq
Dear Valued Customer,
Natura Pet is expanding the recall of specific California Natural, Innova,
EVO, Karma and Healthwise formulas as a precautionary measure due to
potential Salmonella contamination. Mother Nature and wet products are not
included in this recall. No other P&G Pet Care brands are impacted by this
recall.
We were alerted to a single case of Salmonella in a 2.2lb package of EVO
Turkey & Chicken Cat Food on March 15th and took the precautionary measure
of reca... 阅读全帖 |
|
m*q
发帖数: 2 | 49 What does California Have?
• Many world famous giant company, Apple, Intel, Google, Qualcom,
etc
• 48 million people
• One of most productive economy region
• Multi-culture environment
What kind of issue does the current California have?
• Multi billion deficit
• Over 13% of unemployment rate
• One third of people in USA with food stamps located in California
• School rated 48th in 50 states
• Highest tax state
How... 阅读全帖 |
|
o***g
发帖数: 2784 | 50 讨论一道题
原帖如下：
6. http://www.mitbbs.com/article_t/JobHunting/32631467.html
发信人: goodbai (八段锦), 信区: JobHunting
标 题: 热腾腾g电面 已挂
发信站: BBS 未名空间站 (Fri Feb 21 00:20:19 2014, 美东)
同胞面试官，上来就gdoc做题。
2d array ＊代表障碍物 ＃代表货物 空白就是正常的路
问
如何找到一个点为出发点 能实现总共取货路径最短？ 每次只能拿一个货物，遇到障碍
需要绕开，拿到以后要放回出发点，然后再取另一个
＊＊＊＊＊＊
＊ ＃ ＊
＊ ＊＊＊ ＊
＊ ＊
＊ ＊＊ ＊
＊ ＃ ＃＊
＊＊ ＊＊＊＊
大牛们有什么好思路？我用的bfs，但因为之前讨论题目要求花了很久，没有写完。。
我还是太弱了，move on
==================================================================
我看了原帖主题里，都在讨论... 阅读全帖 |
|
