Re: st: Simple evaluation of lf evaluators in mata...

[jpitblado@stata.com (Jeff Pitblado, StataCorp LP)]

Matthew Baker <matthew.baker@hunter.cuny.edu> want to make direct calls to an
'lf' evaluator that was used with -moptimize()-:

> I am encountering the following problem with trying to evaluate
> objective functions written up under different sorts of evaluators in
> mata. The problem is, one can easily evaluate functions written for d0
> evaluators through direct application of the function, but this
> doesn't appear to be the case. That is, if I've constructed a model
> statement M to go with a d0 evaluator and a function d0eval(), I can
> do:
> d0eval(0,M,params,val=.,g=.,H=.)
> 
> and val contains the value of the function evaluated at the
> parameters. If I try this with a type lf evaluator lfeval(), and do:
> lfeval(0,M,params)
> the function is stuck at a previously determined value. My question:
> is there a way to get a quick function evaluation in the lfeval()
> situation, as in the d0eval() situation? Perhaps the following coded
> example will illustrate the problem:
> 
> (example omitted)
> 
> Any advice? I suspect that the optimization is overwriting some aspect
> of the structure underlying the problem and just uses that once
> overwritten.

This kind of thing is just not possible with the 'lf' evaluator, but it is
possible with 'lf0' evaluators.  Here is a modified version of Matthew's
example using an 'lf0' evaluator:

***** BEGIN:
/* Now, a type lf0 evaluator */
function lregeval_lf0(M,todo,b,crit,s,H)
{
        real colvector p1, p2
        real colvector y1

        p1=moptimize_util_xb(M,b,1)
        p2=moptimize_util_xb(M,b,2)
        y1=moptimize_util_depvar(M,1)
        crit=-(y1:-p1):^2:/(2*exp(p2)):-1/2*p2
}
Q=moptimize_init()
moptimize_init_evaluator(Q,&lregeval_lf0())
moptimize_init_evaluatortype(Q,"lf0")
moptimize_init_depvar(Q,1,"mpg")
moptimize_init_eq_indepvars(Q,1,"price weight displacement")
moptimize_init_eq_indepvars(Q,2,"")
moptimize(Q)
lregeval_lf0(Q,0,J(1,5,0),lnf1=.,s=.,H=.)
sum(lnf1)
lregeval_lf0(Q,0,runiform(1,5),lnf2=.,s=.,H=.)
sum(lnf2)
***** END:

--Jeff
jpitblado@stata.com
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/