R help - Sep 2004 - optim "a log-likelihood function"

Hello,

i know that i have to use optim, but i'm confused how its
possible  maximize  the sum over all l[i] and get the optimized
max(LL), r  and alpha?

LL <- function(trans,time){
   for(i in 1:length(trans){
     l[i] <- log(lgamma(r+trans[i] - 
gamma(r+1)*(alpha/alpha+t[i]))**r)*(t[i]/alpha+t[i]))**trans[i]
}
return(sum(l))
   }

i'm confused how i have to set r and alpha 
and i found no related help in archives?

...in Excel it works with solver but only for ~65.000 rows :-)

#This notation is 1 for trans  and 1  for time instead the Startvalues for r 
and alpha?

optim(c(1,1),-LL)

many thanks  for an easy example or hint 
regards,christian

Christian Schulz wrote:
> Hello,
> 
> i know that i have to use optim, but i'm confused how its
> possible  maximize  the sum over all l[i] and get the optimized
> max(LL), r  and alpha?
> 
> LL <- function(trans,time){
>    for(i in 1:length(trans){
>      l[i] <- log(lgamma(r+trans[i] - 
> gamma(r+1)*(alpha/alpha+t[i]))**r)*(t[i]/alpha+t[i]))**trans[i]
> }
> return(sum(l))
>    }
> 
> i'm confused how i have to set r and alpha 
> and i found no related help in archives?
> 
> ...in Excel it works with solver but only for ~65.000 rows :-)
> 
> #This notation is 1 for trans  and 1  for time instead the Startvalues for
r
> and alpha?
> 
I'm not sure what the above statement means, so I may have 
misinterpretted what you are trying to accomplish.
> optim(c(1,1),-LL)
> 
> many thanks  for an easy example or hint 
> regards,christian
> 
Did you look at the first example in ?optim? There also numerous errors 
in LL: missing parans, time is not used, t is undefined in the function.

LL <- function(x, trans, time) {
   r <- x[1]
   alpha <- x[2]
   ...
   sum(l)
}

optim(c(1, 1), LL, control = list(fnscale = -1),
       trans = trans, time = time)

Some style issues:
1. Break up lines that run too long, especially if you expect others to 
read your code.
2. You don't need an explicit "return" at the end of a function.
3. You should remove the "for" loop in LL and vectorise "l",
which
should be easy.

Hope this is helpful,

--sundar

>From: Sundar Dorai-Raj <sundar.dorai-raj at pdf.com>
>Reply-To: sundar.dorai-raj at pdf.com
>To: Christian Schulz <ozric at web.de>
>CC: r-help at stat.math.ethz.ch
>Subject: Re: [R] optim "a log-likelihood function"
>Date: Wed, 29 Sep 2004 10:41:45 -0700
>
>
>
>Christian Schulz wrote:
>
>>Hello,
>>
>>i know that i have to use optim, but i'm confused how its
>>possible  maximize  the sum over all l[i] and get the optimized
>>max(LL), r  and alpha?
>>
>>LL <- function(trans,time){
>>    for(i in 1:length(trans){
>>      l[i] <- log(lgamma(r+trans[i] - 
>>gamma(r+1)*(alpha/alpha+t[i]))**r)*(t[i]/alpha+t[i]))**trans[i]
>>}
>>return(sum(l))
>>    }
>>
>>i'm confused how i have to set r and alpha and i found no related
help in
>>archives?
>>
>>...in Excel it works with solver but only for ~65.000 rows :-)
>>
>>#This notation is 1 for trans  and 1  for time instead the Startvalues
for
>>r and alpha?
>>
>
>I'm not sure what the above statement means, so I may have
misinterpretted
>what you are trying to accomplish.
>
>>optim(c(1,1),-LL)
>>
>>many thanks  for an easy example or hint regards,christian
>>
>
>Did you look at the first example in ?optim? There also numerous errors in 
>LL: missing parans, time is not used, t is undefined in the function.
>
>LL <- function(x, trans, time) {
>   r <- x[1]
>   alpha <- x[2]
>   ...
>   sum(l)
>}
>
>optim(c(1, 1), LL, control = list(fnscale = -1),
>       trans = trans, time = time)
>
>Some style issues:
>1. Break up lines that run too long, especially if you expect others to 
>read your code.
>2. You don't need an explicit "return" at the end of a
function.
>3. You should remove the "for" loop in LL and vectorise
"l", which should
>be easy.
>
I also use optim, however, for my case, can you show some light on avoiding 
the loop?

There are around 200 sets of (i,j,k) where i<=j<=k. 3 situations exist 
whether "=" hold, I list one for example,

                      l<-i:(k-j+i)
                      s<-rep(0,k)
                      s[l]<-choose(j,i)*choose((k-j),(l-i))/choose(k,l)
                      ss<-sum(s*x0)

then sum all the log(ss) is my log-liklihood function.

One loop from 1 to 200 is inevitable. I have tried to use vector, however, I 
only can simply to this situation.  Thanks.

Regards,

Zhen
>Hope this is helpful,
>
>--sundar
>
>______________________________________________
>R-help at stat.math.ethz.ch mailing list
>https://stat.ethz.ch/mailman/listinfo/r-help
>PLEASE do read the posting guide! 
>http://www.R-project.org/posting-guide.html