Monday, December 3, 2012

GAMS question

A question was posed at!topic/gamsworld/0qYEcVd9x8I:

set    j  /n1*n5/,
       k  /n5/;

variable pi(j);

scalar ab;
ab=sum (k, sum(j$(ord(j)=ord(k)), pi.L(j) )  );

The user want to pick the 5-th element. This implementation clearly does not work. Some valid answers were given:

Ab= sum(j$sameas(j,“n5“), pi.L(j) )  ;
ab=sum (k, sum(j$sameas(j,k), pi.L(j) )  );

However I would say this is not really good modeling practice. The best approach IMHO would be:

set   j     /n1*n5/,
      k(j)  /n5/;

variable pi(j);

scalar ab;
ab=sum(k, pi.L(k));

There are two advantages of this approach:

  • The set k is now domain checked against j; no elements outside j can be specified here without alarm bells going off. Domain checking is very important when developing large, complex models.
  • We can substantially simplify the assignment to ab and increase readability. The expression conveys immediately what it is supposed to do.

Setting up proper subsets will help you down the road.