Saturday, July 12, 2008

GAMS 22.8 issues

The new 22.8 beta is out (see http://www.gams.com/). Here are some initial experiences:

  1. I made LS (regression solver) and INVERT (matrix inversion program) available for distribution. It is included but has a few minor issues.
  2. I was hoping that the Cplex interface would allow lazy constraints to be expressed in GAMS models.
  3. Mosek does not crash anymore on some of MIQPs that it does not like. The message is not very user-friendly: Return code - 1500 [MSK_RES_ERR_INV_PROBLEM]. May be in a future release more readable texts will be added.
  4. Mosek still has problems with the model in http://yetanothermathprogrammingconsultant.blogspot.com/2008/06/erratic-miqp-behavior.html. It gives the obscure message: Return code - 1050 [MSK_RES_ERR_UNKNOWN]. This will probably be fixed before the final release. (Update: bug still present in final 22.8).
  5. Some of the IDE issues have not been fixed: http://yetanothermathprogrammingconsultant.blogspot.com/2008/06/gamside-bug_06.html, http://yetanothermathprogrammingconsultant.blogspot.com/2008/06/gamside-bug.html Admittedly they are not very important. (Although the Ctrl-T issue hits me quite often).
  6. This has been fixed: http://yetanothermathprogrammingconsultant.blogspot.com/2008/06/gams-bug-interrupt-does-not-work.html. It is now possible to interrupt a running GAMS model. This will help me a lot.
  7. Qmaker and Cplex QCP interface is still very slow. I passed on a test case. (Update: the cplex QCP interface will be fixed for 22.8).
  8. MIP interrupt in Mosek is largely fixed but the results are not always convincing:
    **** SOLVER STATUS 8 USER INTERRUPT
    **** MODEL STATUS 1 OPTIMAL
    That combination looks suspicious. (This happens only occasionally, probably when hitting Ctrl-C near the end. In most cases it seems to work just fine.).
  9. Tools like gdxcopy, gdx2xls have lost their log. They used write progress messages to stdout with useful info what is happening. These messages are removed.