b*****e
发帖数: 5 | 1 万分感谢你的回答！那是不是对binomial，每个theta0(我后面用的是pi_i_0)就是相应
的yi/ni，因为与data完全吻合相当与good fit?就是saturated model. 你说的展开是
否为：对每个i, logL(pi_i_0)=logL(pi_i_hat)+0＋1/2*(pi_i_0-pi_i_hat)^2*logL''
(pi_i_hat) +...得到LRT stat 约等于 (pi_i_0-pi_i_hat)^2*-logL''(pi_i_hat), 然
后能得到pearson里的(y_i-n_i*pi_i_hat)^2/(n_i*pi_i_hat*(1-pi_i_hat))？可是我
好像得不到这个结论呢。因为LRT里是-logL''(pi_i_hat)＝yi/pi_i^2 + (ni-yi)/(1-
pi_i)^2,没有expectation代入mle,是observed info; 如果是取了expectation再代入
mle,就是wald's test的分母,倒是可以得到结论。 你说的score test难道是等同于(O_
i-E_i)^2/E... 阅读全帖 |
|
S******y
发帖数: 1123 | 2 I have finally got Hadoop working on my Linux box. Next I would like to try
to see if I could to parallel model estimation for some commonly used models
such as logistic regression.
My question now is - how to paralell gradient descent for logistic model
estimation for real large data set?
Any thoughts would be greatly appreciated. Thanks in advance!
PS. See R code below. If needed, I could rewrite the following code in Java
or Python. But the question is how to decompose the following estimatio... 阅读全帖 |
|
i*****r
发帖数: 1302 | 3 很奇怪,我用matlab写的,算上各种variance,covariance的假设,未知量少算有10来多个.
分别给初始值,然后最大化logL应该就完了.但是问题那个logL是经过一系列矩阵运算再
相加得出来的. 然后就卡在那一直算了,状态一直是busy,也不知道要算多久,10几个
variable其实也不算多,像optimization这种都要几十个了.怎么会算这么久呢? 怎么
Eviews算得这么快呢? |
|
A*********r
发帖数: 564 | 4 版权是啥意思？ 你想出来的题，还是是你有独特解法的题？
这道题让我想起了那个N个数中取出K个数作为子集，使得子集的半径最小的题，可能更
复杂一些。。
可以用DP, 令 F(k,i,j)表示从前i个序列中取出k的数的最小差值，其中i个序列中取第
j个数：
F(k,i,j)= min {x in 序列 i-1 } { F（k-1,i-1,x)+abs(A[i-1][x]-A[i][j]) }
算法复杂度为O(N^2*L), 可以优化为 O(N^2*logL),其中L为最长序列的长度。。 |
|
x***y
发帖数: 633 | 5 At worst, you need to find the divide point with time O(logL), where L is
the length of search row or column.
For this Y table, the worset search is that target is not in the table and
each divide take the most time.
So, to divide until only one element is left, we need (assume the matrix is
M*N)
Log(M-1)+Log(M-2)+...log(2) + log(N-1)+log(N-2)+....log(2)
=O(log(M!)+log(N!)) then using stirling's formula
=O(MlogM+NlogN) in the worse case. |
|
c*****o
发帖数: 1702 | 6 今天收到人生第一个onsite后的据信。这次onsite面试在2周前,看起来还是很干脆的据
信。痛定思痛，决心以后再也不靠运气吃饭了。努力准备！
发面经攒人品：
电面1：
important features of OOP: encapsulation,inheritance and polymophism. what
is abstract class,interface. are we able to inherit from multiple abstract
class or interface. So basically all OOP definitions.
电面2：
still OOP, can C++/Java/C# use multiple inheritances? 然后就一堆废话，后来
知道第一轮都答对，第二轮就很水
onsite 3:早上11点开始，45分中一轮，一共见了6个人，本来要见的COO不在。分别是
director, senior developer,developer,senior developer, archtecter,还有
managin... 阅读全帖 |
|
n*******r
发帖数: 425 | 7 Tues
Earnings: DRI (TBUS), Helicos BioSciences (HLCS), AspenBio Pharma (APPY),
Iteris (ITI), Embotelladora Andina (AKO.A), Embotelladora Andina (AKO.B),
Rediff.Com (REDF), China Medical Technologies (CMED), Saba Software (SABA),
Sanderson Farms (SAFM), Sociedad Quimica y Minera de Chile (SQM), Bank of
Nova Scotia (BNS), Tata Motors (TTM)
Economic Data: Case-Shiller 20-city Index, Consumer Confidence
Wed
Earnings: Optibase (OBAS), Mega Precious Metals (MPRXF), NXT Energy
Solutions (NSFDF), ... 阅读全帖 |
|
