Re[3]: [Help-glpk] slow matrix generation compared to AMPL

[Andrew Makhorin]

Here is an example of inefficient coding in MathProg.

Let there be the following constraint declaration:

   s.t. CAP{i in I, j in J: A[i,j] > 0 and (i,j) in S} ... ;

An equivalent pseudo-code used by the MathProg translator to generate
CAP is the following:

   for {i in I}
      for {j in J}
         if A[i,j] > 0 and (i,j) in S then generate CAP[i,j];

where 'if' is executed for *all* doublets (i,j) in I x J.

It is understood that if set S contain much less doublets than I x J,
the following declaration would be much more efficient:

   s.t. CAP{(i,j) in S: A[i,j] > 0} ... ;

because its pseudo-code is the following:

   for {(i,j) in S}
      if A[i,j] > 0 then generate CAP[i,j];

where 'if' is executed only for doublets (i,j) in S, and, moreover,
there is no need to test whether (i,j) belongs to S or not.

Note that (unlike the AMPL system) the MathProg translator always
executes all constructions in the order they are written in the model
description.


Andrew Makhorin