Iterative Reweighted Least Squares
来源:互联网 发布:淘宝小二人工服务电话 编辑:程序博客网 时间:2024/04/28 11:36
- 逼近 approximation
- 加权最小二乘weighted least squared error approximation
- 逼近其他范式
- IRLS逼近 l_p norm
- Overdetermined system N p
- Algebra 方法
- IRLS for Logit Regression
- Modeling
- Overdetermined system N p
- Iterative Least Squares for logistic regression
- IRLS
本文基于这篇paper。
逼近 (approximation)
对于线性的问题,其模型是:
如果
常采用的是
对于
- If N=p, (square and nonsingular), 存在严格解:
x=A−1b - If N > p (over specified):
ATAx=ATb→x=[ATA]−1ATb
ATA∈ℝp×p - If N < p (under specified):
Ax=AAT(AAT)−1b→x=AT[AAT]−1b
加权最小二乘(weighted least squared error approximation)
W 是diagonal matrix。
ax2+bx+c 的极值是x=−0.5a−1b
- 对于N > p:
x=[ATWTWA]−1ATWTWb - 对于N < p:
逼近其他范式
比较有名的就是具有稀疏性意义的
IRLS逼近 lp norm
IRLS(iterative reweighted least squares) allows an iterative algorithm to be built from the analytical solutions of the weighted least squares with squares with an iterative re weighting to converge to the optimal
IRLS使用迭代的方式解决带权重的
Overdetermined system N > p
Algebra 方法
IRLS for Logit Regression
对于二分问题,我们想知道条件概率
我们的链接函数(link function)是logit,也即:
g(p)=logit(p)=logp1−pp=Pr(Y=1|X) 那么我们要对X进行拓展(extension)
ϕ(x)=[1,ϕ1(x),...,ϕp(x)]T
这样用
我们记:
Modeling
对于n个数据集(
这是一个优化问题:
我们对
Hessian Matrix :
采用Newton-Raphson 迭代方式:
令
可以看作是weighted least square.
Iterative Least Squares for logistic regression
我们知道logistic regression模型是:
Link function 是
我们知道,
Logit Regression 的分布是二项分布,
Because the variance of Z changes with X , this is a heteroskedastic regression
problem, the appropriate way of dealing with such a problem is to use weighted least squares, with weights inversely proportional to the variances. This means that, in logistic regression, the weight at x should be proportional to p(1 − p)
IRLS
- Get the data
(x1,y1),...(xn,yn) , and some initial guessesβ .- Until
β0 , β converge:
a) Calculateη(xi) =βxi ,以及相应的p(xi)
b) Transformed responseszi=η(xi)+yi−p(xi)p(xi)(1−p(xi))
c) Calculate the weightswi=p(xi)(1−p(xi))
d) Do a weighted linear regression ofzi onxi with weightswi
- Iterative Reweighted Least Squares
- 171123 IRLS-Iteratively Reweighted Least Squares
- Least Squares
- Least squares
- 最小二乘法(Least Squares Fitting)
- 最小二乘法(least squares)
- Ordinary least squares是什么意思?
- Partial Least Squares Regression (PLS)
- least squares Ax = b
- Probabilistic Interpretation on Least Squares
- Least Squares Method & Gradient Descent
- Modeling Non-linear Least Squares
- Least-squares and linear programming
- 最小二乘法 (Least Squares Method, LS)
- Least Squares Estimation for multiple linear regression
- LSQR: Sparse Equations and Least Squares
- Linear and nonlinear least squares regeressions
- Ordinary Least Squares 最小二乘约束
- [莫比乌斯反演] BZOJ 4816 [Sdoi2017]数字表格
- RMQ问题
- 【转载】零基础入门深度学习(1)
- 获取对话框句柄 对话框指针
- 什么是对象?
- Iterative Reweighted Least Squares
- Dll注入技术之APC注入
- javascript中从字符串中抽取和切割字符串的方法
- c与c++函数堆栈理解(转的地址)
- Game of Taking Stones(威佐夫博奕 待整理)
- 学习笔记
- 进程(2)- 进程队列及管理
- 俄罗斯方块之三 基本类型设计
- minimum-depth-of-binary-tree