[ADMB Users] likelihood profile for elements in vector?

[Poos, Jan Jaap]

You can do just about anything in C++.
Define a class likeprof_vector in  the GLOBALS_SECTION
You may have to move some things in the header file to make them
public.  also this was originally done on dos with 8+3 file names so
keep them short
or they may get truncated.
DATA_SECTION
PARAMETER_SECTION
   init_number logR1
   init_number logR2
   sdreport_number avgR
  !!CLASS likeprof_vector xx(1,20,"my")      // vector of 20
likeprof_numbers
   sdreport_number R1
   sdreport_number R2
   objective_function_value f
PROCEDURE_SECTION
   R1=exp(logR1);
   R2=exp(logR2);
   avgR=(R1+R2)/2.0;
   f=100*square(logR1+logR2);
   f+=square(R1);
   f+=square(R2);
   for (int i=1;i<=20;i++)
   {
     xx(i)=i*R1;
   }

GLOBALS_SECTION
   #include <admodel.h>
   class likeprof_vector
   {
       int index_min;
       int index_max;
       param_likeprof_number * v;
   public:
       int indexmin(){ return index_min;}
       int indexmax(){ return index_max;}
       param_likeprof_number& operator () (int i) {return v[i];}
       likeprof_vector(int lb,int ub,const char * la)
       {
           index_min=lb;
           index_max=ub;
           v=new  param_likeprof_number[index_max-index_min+1];
           v-=index_min;
           for (int i=index_min;i<=index_max;i++)
           {
             adstring label=la + adstring("_") + str(i);
             (*this)(i).allocate(label);
           }
        }
         ~likeprof_vector()
        {
           v+=index_min;
           delete [] v;
           v=0;
        }
   };
                                                                                                                     51,1          Bot
----------------------------------------
Hi,
Thanks for this quick response and the worked out example. Indeed I had to move around two lines in the admodel.h file to make them public members of a class. Also, I had to remove the  " // vector of 20" comment. I did not change the newmodmn.cpp file. It runs, but I have some questions about its behaviour.

I tried this on both my own code and the example provided above, and I get the same results (using ADMB 9.0). First, whenever I add the   

  !! likeprof_vector xx(1,20,"my") 
likeprof_numbers

part to the model, ADMB warns that the covariance matrix may not be pos definite, while without that code, I do not get a warning. The warning is followed by a vector with the length of the estimated parameters. The estimation did not change (in terms of estimated parameters, starting values, data), so why do I get a warning that I did not have previously?

If I run the model with the -lprof flag, both the example and my own code end with
 
"Error standard dev of likeprof parameter is 0
 xdist(0,-8) too small"

I get two files in the directory, one called "s.plt" with only "s:" in it, and one file ending in .prf with 

s:
0
123456 345678 (these numbers being different for the example and my own code)

What I am doing wrong?

Kind regards, Jan Jaap