topic Re: Optimize a function on a column in SAS Programming

Optimize a function on a column

liyongkai800 — Tue, 04 Dec 2018 16:50:49 GMT

say I have a function f(x,y)=x+y, and the data set is given by

var

Now I want to minimize the the function (x+y-3)^2+(x+y-6)^2+(x+y-5)^2+(x+y-7)^2 with the constraints 0<x<5, 0<y<5.

I know the proc fcmp can be used to define a function like (x+y-3)^2, and optimize it under some constrains, but not sure how to use it on the whole column, any hints? Thanks.

Re: Optimize a function on a column

PeterClemmensen — Tue, 04 Dec 2018 17:00:55 GMT

How does the value var relate to the function you want to minimize?

PROC NLIN would be my first choice

Re: Optimize a function on a column

liyongkai800 — Tue, 04 Dec 2018 18:44:42 GMT

It seems PROC NLIN can estimate the parameters of the model directly, therefore I don't need to construct the function by myself. Thanks for the hint, I will give a try.

Re: Optimize a function on a column

PeterClemmensen — Tue, 04 Dec 2018 20:28:08 GMT

Anytime, glad to help 🙂

Let me know how it goes and please post the final code here for other users to benefit from.

Re: Optimize a function on a column

liyongkai800 — Thu, 06 Dec 2018 20:55:29 GMT

The program is pretty much like this, and you may need to adjust the initial values (in parms statement) if the hessian matrix is singular.

proc nlin data=whatever method=marquardt;
parms x1=1 x2=0 to by 0.1;
bounds x1>=0;
model y=f(x1,x2);
output out=b predicted=yp;
run;