R help - Apr 2017 - How to solve "nlm() non-finite value supplied by 'nlm'" and how to constrain parameters to (0, 1)?

So I'm getting:

Error in nlm(neglikhood, p = c(0.1, 0.1), hessian = T, x = elinajat) :
   non-finite value supplied by 'nlm'
In addition: There were 50 or more warnings (use warnings() to see the 
first 50)

with the following (neglikelihood of 1-param. Weibull):

neglikhood <- function(theta,x) {
   n <- length(x)
   
-(n*log(theta[2])+n*log(theta[1])+(theta[1]-1)*sum(log(x))-theta[2]*sum(x^theta[1]))
}
> elinajat
  [1]  17.88  28.92  33.00  41.52  42.12  45.60  48.48  51.84  51.96  
54.12  55.56  67.80
[13]  68.64  68.64  68.88  84.12  93.12  98.64 105.12 105.84 127.92 
128.04 173.40

I read somewhere that I might need to constrain my parameters to the 
range (0,1). Is this correct?

For example:
> neglikhood(c(0.1,0.1),elinajat)
[1] 195.3213

nlm doesn't include bounds so you'd need to transform. nlminb does, as
do a number of codes in optimr (and more in R-forge version optimrx). The latter
packages have a common call which simplifies choosing the solver. My own Rvmmin
is an all-R
implementation of the same method (but some details changed) as optim::BFGS but
it includes
bounds constraints. Moreover, it is fairly easy to get inside and add some
checks if needed.

If you can, I recommend supplying analytic gradients too, though that is often
quite a
bit of work. If you do this, remember to check that the code is correct
(numDeriv helps).

Since you have only 2 parameters, I think I'd give nmkb (it is from dfoptim
or in optimrx).
It handles bounds, but you cannot start on the bound because it uses the
transfinite transformation
to handle the bounds. The method is relatively slow and clunky, but for 2
parameters, you are not
going to wait too long.

Best, JN


On 2017-04-02 08:19 AM, mviljamaa wrote:
> So I'm getting:
>
> Error in nlm(neglikhood, p = c(0.1, 0.1), hessian = T, x = elinajat) :
>   non-finite value supplied by 'nlm'
> In addition: There were 50 or more warnings (use warnings() to see the
first 50)
>
> with the following (neglikelihood of 1-param. Weibull):
>
> neglikhood <- function(theta,x) {
>   n <- length(x)
>  
-(n*log(theta[2])+n*log(theta[1])+(theta[1]-1)*sum(log(x))-theta[2]*sum(x^theta[1]))
> }
>
>> elinajat
>  [1]  17.88  28.92  33.00  41.52  42.12  45.60  48.48  51.84  51.96  54.12 
55.56  67.80
> [13]  68.64  68.64  68.88  84.12  93.12  98.64 105.12 105.84 127.92 128.04
173.40
>
> I read somewhere that I might need to constrain my parameters to the range
(0,1). Is this correct?
>
> For example:
>
>> neglikhood(c(0.1,0.1),elinajat)
> [1] 195.3213
>
> ______________________________________________
> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> 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.