Bruce McCarl's GAMS Newsletter Number 19

GAMS 22.2 Release

GAMS Corporation has released versions 22.0, 22.1 and 22.2 which expanded some language capabilities. Cumulative release notes are on the GAMS web site at http://www.gams.com/docs/release/release.htm#22.2. Some of the new features within these releases are discussed below. The software can be downloaded from http://www.gams.com/download/.

The main new features are

• Performance improvements for very large and complicated loop structures.
• GAMS IDE modifications
  • GDX data browser is faster and can sort indices by name vs. entry order
  • A symbol shown in the GDX data browser can be written to an Excel file
  • Navigation tree for the lst file.
  • Ability to generate a chart based on an input file and data in a GDX file or based on an interactive process. Look in the IDE Help under Help Topics>Guided Tour>Charts for details
  • Ability to manually select a table in GDX viewer and send it to EXCEL.
• New versions of BARON, CONOPT, CPLEX, DICOPT, EXAMINER, LGO and XPRESS. Some are minor upgrades but CONOPT has improved scaling, CPLEX has a new version (10) with improved MIP and infeasibility features and XPRESS has a new version with improved performance.
• Elimination of integer variable priorities during a run using X.prior=inf;
• Definition of new variable and equation attributes involving
  • Slack variables (slack,slacklo,slackup)
  • Infeasibility (infeas)
• Ability to use model names in a model equation list
  Model first /eq1,eq2/;
  Model second /first,eq3/;
• Model type MPEC now allows discrete variables and a new model type RMPEC relaxes these to continuous
• $IF check for the existence of a directory $IF DEXIST name
• McCarl GAMS User's Guide
  • Changed to a different format permitting improved navigation
  • Collapsed into a single .pdf and .chm file with working links in pdf
• The release notes also indicate new entries in the model and test model libraries.

Speed and GAMS

I was recently working on a large and slow model. I found great success in speeding up certain types of calculations using a looping strategy. Namely I was computing

     climatepesticide(crop,Aggregateregs,periods,gcm)= 
      		sum(foraggregate(Aggregateregs,subreg),
              		climatepesticide(crop,subreg,periods,gcm)*
             		aggregatewtcr(Aggregateregs,SubReg,CROP));

In trying to speed this up I used the loop

     tclimatepesticide(crop,Aggregateregs,periods,gcm)=0;
     loop((rcrop(crop),allregforaggregate(Aggregateregs,subreg),periods,rgcm(gcm))
            	$(    (not sameas(Aggregateregs,subreg))
              	and climatepesticide(crop,subreg,periods,gcm)),
       		tclimatepesticide(crop,Aggregateregs,periods,gcm)
          			= tclimatepesticide(crop,Aggregateregs,periods,gcm)+
             		   climatepesticide(crop,subreg,periods,gcm)*
             		   aggregatewtcr(Aggregateregs,SubReg,CROP));
     climatepesticide(crop,Aggregateregs,periods,gcm)=
          		tclimatepesticide(crop,Aggregateregs,periods,gcm);
     tclimatepesticide(crop,Aggregateregs,periods,gcm)=0;

• Zeros an array and then adds into it using a loop carefully structured over non zero cases rather than summing.
• Adds into a temporary array different rather than the permanent array as this can be much faster. (GAMS apparently can slow down when jumping around in the sparse data array.) After the computation I then copy the results in all at once and then zero the temporary array.

CPLEX and a Basis

For a year or two I have noted a real degradation in the performance of a large model in repeated solves with CPLEX. The GAMS folks helped me out and found that benefits of an advanced basis can be enhanced using three CPLEX options.

Turn off the dual simplex method	lpmethod 1
Cause more aggressive scaling to occur	scaind 1
Force basis use and suppress presolve	advind 1

The real problem is the presolve and the basis do not always live well together and CPLEX used to suppress the presolve and just use the basis but does not as of verion 9 or so. Thus the last option is critical.

A large time savings (75%) resulted and perhaps others using advanced bases or repeated solves can benefit.

MIP Indicator Constraints in CPLEX

     aX - MY <= 0,  X >= 0,  Y in (0,1),  and M a big number

CPLEX has recently introduced an "indicator constraint" concept in an effort to gain solution speed and GAMS has accommodated this at the moment through the CPLEX option file. Namely an indicator constraint identifies a binary variable/constraint pair where the binary variable turns the constraint on or off. For the time being, the couplings between binary variables and constraints are made through an option file as detailed at http://www.gams.com/solvers/cpxindic.htm.

FeasOpt in CPLEX

Be careful with Execute_Loadpoint

Comparing Files

     $setglobal f1 c:\file1.gms
     $setglobal f2 c:\file2.gms
     $call '"%gams.sysdir%gbin\diff.exe"  %f1% %f2%'

Grid Computing and MIPs

Tom Rutherford Utilities

Courses offered

A 3.5 day AAdvanced GAMS class will be offered in Frisco, Colorado August 14-17, 2006. That course is application independent mainly concentrating on the GAMS language, interfaces, graphing, report writing etc.

Further information and other courses are listed on http://www.gams.com/courses.htm.

Unsubscribe to future issues of this newsletter

This newsletter is not a product of GAMS Corporation although it is distributed with their cooperation.

May 8, 2006