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

[Robert Fourer]

Instead of, say, 

   subject to LB2{(i,j,k,l,m,n,c) in EMPTYMOVES: c = 'SO'}: 
      EMPTY[i,j,k,l,m,n,c]>=0;

you can write

   subject to LB2{(i,j,k,l,m,n,'SO') in EMPTYMOVES}: 
      EMPTY[i,j,k,l,m,n,'SO']>=0;

and similarly in many other places in the model.  This is valid AMPL so it can 
be
expected to work in MathProg as well, maybe with highly favorable consequences 
for
the processing time.

-- Bob Fourer
   address@hidden



> -----Original Message-----
> From: address@hidden [mailto:help-glpk-
> address@hidden On Behalf Of Kendall Bailey
> Sent: Monday, August 16, 2004 2:25 PM
> To: Andrew Makhorin; address@hidden
> Subject: [Fwd: Re: [Help-glpk] slow matrix generation compared to AMPL]
> 
> 
> I am now able to post the full model. The size of the data sets are
> 
> L 185
> T 169
> D 29
> C 6
> LC 639
> LTD 31265
> LTDC 107991
> DEMANDLANES 7409
> DEMANDMOVES 63206
> LOADMOVES 27812
> EMPTYMOVES 527993
> 
> The model represents a time/space network transportation model.
> Hopefully the comments make the meaning clear.  I'm interested in ways
> to speed up the matrix generation phase of MathProg.  I think I
> understand the earlier advice, which applies mostly to the set C.  Since
> the constraints specify c = 'SO', then it should be possible to reduce
> the work by 5/6.  However I'm looking at reducing a 24 hour generation
> phase to a few minutes at most.
> 
> Thanks,
> Kendall
> 
> 
>