Dropping a mata function

[George D]

Hi all,

I have the following the optimization function in mata:

mata
mata set matastrict off

void mvopt(todo, p, fml, g, H) {
fml=253*($m1*invlogit(p[1])+$m2*invlogit(p[2])+$m3*((1-invlogit(p[1])-invlogit(p[2]))))/*
*/-sqrt(253)*sqrt($s11*(invlogit(p[1]))^2+$s22*(invlogit(p[2]))^2+$s33*(1-invlogit(p[1])-invlogit(p[2]))^2 /*
*/+2*$s12*invlogit(p[1])*invlogit(p[2])/*
*/+2*$s13*invlogit(p[1])*(1-invlogit(p[1])-invlogit(p[2]))/*
*/+2*$s23*invlogit(p[2])*(1-invlogit(p[1])-invlogit(p[2])))
}

S = optimize_init()
optimize_init_evaluator(S, &mvopt())
optimize_init_which(S, "max")
optimize_init_params(S, (.3, .3))
(void) optimize(S)
p = optimize_result_params(S)
st_matrix("w",p)

end

After completing the optimization I would like to drop the function using mata drop myopt() but I receive the error:

mvopt() in use (nothing dropped)

I can't seem to figure out why it would say the function is in use. Is there a way I can drop this function after running it?

Thank you!
George

[Sergiy Radyakin]

George, I am guessing, but there must be a reference somewhere to the pointer you supplied:
&mvopt() Which needs to be dropped/cleared first. Best, Sergiy Radyakin

[James Fiedler]

Sergiy is right. S is holding a reference to your function. You can see this with a simpler example taken from the documentation:

mata

void myeval(todo, x,  y, g, H)
{
    y = exp(-x^2 + x - 3)
}

S = optimize_init()
optimize_init_evaluator(S, &myeval())

end

After running the above

: optimize_init_evaluator(S)
  0x21c69c0

: &myeval()
  0x21c69c0

Thus you have to drop S before you can drop your function.

[George D]

Many thanks James and Sergiy! I dropped S and it works now.

I am now trying to impose a constraint using optimize_init_constraints(S,Cc) and I was wondering if it is possible to impose an inequality constraint. I know how to impose constraints such as P₁ = P₂ but I don't know how to do something like P₁ >= 0.

Thank you.
George

[Sergiy Radyakin]

George, start a new thread since this is a totally different question.

If you want to minimize F(x) subject to P1>=0, then you can minimize F(x)+(P1<0)*10^200 unconditionally. See 'Penalty method'.

There is likely a more efficient method. So start a new thread and wait for the crowd.

Best, Sergiy

[George D]

Thanks Sergiy. I moved this to another thread.