R中的Newton Raphson代码涉及集成和贝塞尔函数

时间:2014-08-17 13:13:33

标签: r integration newtons-method bessel-functions

我想估计涉及贝塞尔函数和积分的函数的参数。但是,当我试图运行它时,我收到一条消息“f(x,...)中的错误:找不到函数”BesselI“”。我不知道如何解决这个问题,并对任何相关提案表示赞赏。

library(Bessel)
library(maxLik)
library(miscTools)


K<-300

f  <- function(theta,lambda,u) {exp(-u*theta)*Vectorize(BesselI(2*sqrt(t*u*theta*lambda),1))/u^0.5}   
F  <- function(theta,lambda){integrate(f,0,K,theta=theta,lambda=lambda)$value}
tt <- function(theta,lambda){(sqrt(lambda)*exp(-t*lambda)/(2*sqrt(t*theta)))*
                                   (theta*(2*t*lambda-1)*F(theta,lambda))}
loglik <- function(param) {
   theta <- param[1]
   lambda <- param[2]
   ll <-sum(log(tt(theta,lambda)))
}
t <- c(24,220,340,620,550,559,689,543)
res <- maxNR(loglik, start=c(0.001,0.0005),print.level=1,tol = 1e-08) 
summary(res)

Newton-Raphson最大化 迭代次数:0 返回码:100 初始值超出范围。

我得到了“有50个或更多警告(使用警告()查看前50个警告”)当我使用警告()时,以下是警告。

In t * u : longer object length is not a multiple of shorter object length.

sessionInfo()
R version 2.14.2 (2012-02-29)
Platform: i386-pc-mingw32/i386 (32-bit)
locale:
[1] LC_COLLATE=English_United States.1252 
[2] LC_CTYPE=English_United States.1252  
[3] LC_MONETARY=English_United States.1252
[4] LC_NUMERIC=C                    
[5] LC_TIME=English_United States.1252    

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] maxLik_1.1-2     miscTools_0.6-16   Bessel_0.5-4     Rmpfr_0.5-1  
[5] gmp_0.5-4       
loaded via a namespace (and not attached):
[1] sandwich_2.2-10

2 个答案:

答案 0 :(得分:1)

警告:答案/探索不完整。

让我们把它剥掉一点,试着看看警告的来源,这可能会给你更多的了解。至少我们可以将其作为一种可能性消除。

K <- 300 ## global variable, maybe a bad idea
t <- c(24,220,340,620,550,559,689,543)  ## global/same name as t(), ditto
library(Bessel)
f  <- function(theta,lambda,u) {
    exp(-u*theta)*BesselI(2*sqrt(t*u*theta*lambda),1)/u^0.5}
## Vectorize() isn't doing any good here anyway, take it out ...
F  <- function(theta,lambda){
       integrate(f,0,K,theta=theta,lambda=lambda)$value}

F()有效但仍然提供了矢量化警告:

F(theta=1e-3,lambda=5e-4)
## [1] 3.406913
## There were 50 or more warnings (use warnings() to see the first 50)
f(theta=1e-3,lambda=5e-4,u=0:10)
##  [1]         NaN 0.010478182 0.013014566
##  [4] 0.017562239 0.016526010 0.016646497
##  [7] 0.018468755 0.016377872 0.003436664
## [10] 0.010399265 0.012919646
## Warning message:
## In t * u : longer object length is not a multiple of shorter object length

我们仍然收到警告。我们还可以看到,在0处评估被积函数可能是个问题(我们在分母中有u的平方根...)

看起来我们可能需要对外部产品f()t的所有组合)评估u,然后可能总结{{ 1}}超过f的值?这确实需要解决,因为t具有固定长度(可能是某种数据样本),而集成变量t将具有任意多个值......

您能提供指向对数似然函数原始推导的链接吗?

答案 1 :(得分:0)

这就是我所拥有的。不幸的是,我无法弄清楚如何在t向量上运行它,但是我想您可以做一个for循环。

library(base)
install.packages("maxLik")
library(maxLik)

#setting some initial values to test as I go
K <- 300
u <- 1
theta <- 1
T <- c(1,2,3,4)
lambda <- 1

#I got it to work with besselI instead of BesselI
#I also dropped Vectorize and instead do a for loop at the end
f  <- function ( theta, lambda, u ) {
    exp(-u * theta) * besselI(2 * sqrt(t * u * theta * lambda), 1) / u^0.5
}
#testing to see if everything is working
f(1,1,1)

#making a function with only one input so it can be integrated 
F <- function ( u ) {
    f(theta, lambda, u)
}
F(1)

#testing for integration
t <- T[1]
integrate(F, 0, K)

#entered the integration function inside here at the bottom
tt <- function ( theta, lambda ) {
    (sqrt(lambda) * exp(-t * lambda) / (2 * sqrt(t * theta))) *
        (theta * (2 * t * lambda - 1) * integrate(F, 0, K)$value)
}
tt(1,1)

#took the absolute value of theta and lambda just in case they turn out negative
loglik <- function(param) {
    theta <- param[1]
    lambda <- param[2]
    ll <-sum(log(tt(abs(theta),abs(lambda))))
    return(ll)
}

#example for using the for loop
for ( i in 1 : length(T) ) {
    t <- T[i]
    print(loglik(c(1,1)))
}

maxNR(loglik, start = c(1, 5), print.level = 1, tol = 1e-08)

我不确定您想要什么参数theta和lamba,但是我的设置有效并且使用了t值。就奇异矩阵而言,可能与t和theta / lambda的大小有关。不确定maxNR是否使用广义逆,但我确定确实如此。