R help - Oct 2008 - Help MLE

Dear, 

I'm starting on R language. I would like some help to implement a MLE
function.

I wish to obtain the variables values (alpha12, w_g12, w_u12) that maximize
the function LL = Y*ln(alpha12 + g*w_g12 + u*w_u12).

Following the code:

 rm(list=ls())   
 ls()
 library(stats4)

 Model = function(alpha12,w_g12,w_u12)
 {
 Y = 1
 u = 0.5
 g = -1
 Y*log(alpha12 + g*w_g12 + u*w_u12)
 }

 res = mle(minuslog=Model,start=list(alpha12=0.1,w_u12=0.1,w_g12=0.1),
method = "BFGS")
 
 res

Error message:

 > res = mle(minuslog=Model,start=list(alpha12=0.1, w_u12=0.1, w_g12=0.1),
method = "BFGS")
 Error in optim(start, f, method = method, hessian = TRUE, ...) : 
  non-finite finite-difference value [1]

Thanks,
LF
-- 
View this message in context:
http://www.nabble.com/Help-MLE-tp19904724p19904724.html
Sent from the R help mailing list archive at Nabble.com.

LFRC <feliperiehs <at> yahoo.com.br> writes:
> 
> 
> Dear, 
> 
> I'm starting on R language. I would like some help to implement a MLE
> function.
> 
> I wish to obtain the variables values (alpha12, w_g12, w_u12) that maximize
> the function LL = Y*ln(alpha12 + g*w_g12 + u*w_u12).
> 

  You're running into a problem because your linear combination
goes negative in the course of the optimization, which makes
the logarithm NaN, which crashes the optimizer.  The quick
hack would be to calculate the linear combination and then
set it to max(value,some_small_positive_number) -- this will
(probably) work if the solution is not on the boundary.
Alternatively you can use constrOptim (see below),
because you need to impose a linear inequality constraint.

 rm(list=ls())   
 ls()
 library(stats4)

Model = function(alpha12,w_g12,w_u12)
  {
    Y = 1
    u = 0.5
    g = -1
    comb = alpha12 + g*w_g12 + u*w_u12
    r = Y*log(comb)
    cat(alpha12,w_g12,w_u12,comb,log(comb),r,"\n") ## debug
    r
  }

res = mle(minuslog=Model,start=list(alpha12=0.1,w_u12=0.1,w_g12=0.1),
  method = "BFGS")

model2 <- function(p) {
  do.call("Model",as.list(p))
}

constrOptim(theta=c(0.1,0.1,0.1),f=model2,
            grad=NULL,
            ui = c(1,-1,0.5),
            ci = rep(0,3))
            

  Ben Bolker

Thanks again.

Now I want to complicate the model. Supose that Y is a vector with 30 data.

Y = c(0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0,
1, 0, 0, 0, 0, 1, 0)
Then the r = sum(Y*log(comb))

Is it possible? I want to discover the variables values
(alpha12,w_g12,w_u12) for each Y that optimize the function r.

The new code is:

Model = function(alpha12,w_g12,w_u12) 
  { 
    Y = c(0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0,
0, 1, 0, 0, 0, 0, 1, 0) 
    u = 0.5 
    g = -1 
    comb = alpha12 + g*w_g12 + u*w_u12 
    r = sum(Y*log(comb))
    cat(alpha12,w_g12,w_u12,comb,log(comb),r,"\n") ## debug 
    r 
  } 

res = mle(minuslog=Model,start=list(alpha12=0.1,w_u12=0.1,w_g12=0.1), 
  method = "BFGS") 

model2 <- function(p) { 
  do.call("Model",as.list(p)) 
} 

constrOptim(theta=c(0.1,0.1,0.1),f=model2, 
            grad=NULL, 
            ui = c(1,-1,0.5), 
            ci = rep(0,3)) 

ERROR:

Error in optim(start, f, method = method, hessian = TRUE, ...) : 
  objective function in optim evaluates to length 30 not 1

Best regards,
LFRC


Ben Bolker wrote:
> 
> LFRC <feliperiehs <at> yahoo.com.br> writes:
> 
>> 
>> 
>> Dear Ben Bolker,
>> 
>> Thanks a lot for your help.
>> 
>> I have two more questions:
>> 
>> 1) My goal is maximize the function (>     r = Y*log(comb)) but the
>> parameters found, minimized the function (r = Y*log(comb)).
>> 
> 
>   Oh.  Oops.  Just change the sign ( r = -Y*log(comb)) or
> set fnscale=-1 (see ?optim) 
> 
>> 2) What this function do?
>>   > model2 <- function(p) {
>>   >   do.call("Model",as.list(p))
>>   > }
>> 
>  
>   I got a little carried away, this is R Magic.
> mle() expects the function to be written out with
> a list of parameters ( func(param1,param2,param3) ),
> while optim and constrOptim want the function to
> have a single vector argument ( func(paramvec) ).
> model2() is a wrapper that converts a vector p
> into a list and passes it to the Model function.
> 
>   Ben Bolker
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
> 
> 
-- 
View this message in context:
http://www.nabble.com/Help-MLE-tp19904724p19927405.html
Sent from the R help mailing list archive at Nabble.com.