Originally posted by: JohanLöfberg

Trying to tweak the mip settings in MATLAB with cplex 12.6 (tried 12.5 too), and nothing seem to change. For instance, emphasis

f     = [-1 -2 -3 -1]';
Aineq = [-1  1  1 10; 1 -3  1  0];
bineq = [20 30]';
Aeq   = [0  1  0  -3.5];
beq   =  0;
lb    = [0;    0;   0; 2];
ub    = [40; inf; inf; 3];
ctype = 'CCCI';
options = cplexoptimset('cplex');
options.Display = 'on';
options.emphasis.mip = 2;
[x, fval, exitflag, output] = cplexmilp (f, Aineq, bineq, Aeq, beq,[ ], [ ], [ ], lb, ub, ctype, [ ], options);

Running the code displays

Presolve time = 0.00 sec. (0.00 ticks)
MIP emphasis: balance optimality and feasibility.

i.e,, the default mip emphasis of 0 is kept.

Originally posted by: JohanLöfberg

Bumping the question to get a reply from product officials (as CPLEX seems to lack a more structured way to report bugs). Want to know Iif I am doing something wrong or if this is broken in CPLEX.

Originally posted by: Mark L. Stone

Bumping on behalf of Johan, as I too would like to understand what's going on here.  Thanks.

Originally posted by: RWunderling

Unfortunately, you hit a true bug in the handling of CPLEX parameters within the toolbox functions.  This issue is known and will be fixed for the next release. Sorry about the inconvenience.

For the time being you can work around the issue by using parameter numbers.  Taking the mip emphasis parameter as an example, you can see from the documentation that the internal parameter number for this parameter is 2058.  You can set this using

options.parameter2058 = 2;

Alternatively, you can also use the CPLEX class for matlab, which provides better usability and supports model modifications with restarts.

Originally posted by: Mark L. Stone

In light of the suggestion to use the CPLEX Class API, perhaps someone could answer the post  ' "Class API" much slower than cplexqp (and Matlab's quadprog)' https://www.ibm.com/developerworks/community/forums/html/topic?id=e379fbe3-2362-46f3-82ba-94c359303183&ps=25 .  Thanks.

Originally posted by: EdKlotz

Here are some more details on this, from

https://www.ibm.com/developerworks/community/forums/html/topic?id=7d401a80-4ee0-44a1-bb8e-65305ba81c7a#8aa8b62d-79ff-474b-bc2f-ee2628246896

Here is a slightly more detailed explanation of the workaround, using the MIP gap parameter that is of interest for this particular thread.  The issue is that parameters that don't correspond to MATLAB toolbox parameters won't be properly communicated to CPLEX when using the Toolbox API.   To work around the issue, perform the following steps.

1.  Looking up the numeric identifier of the parameter(s) of interest in the CPLEX documentation.   For the MIP gap parameter, this means the last line in the table below.

relative MIP gap tolerance

Sets a relative tolerance on the gap between the best integer objective and the objective of the best node remaining.

2.   Use this number with your options structure as follows.  For a 5% MIP gap,

options.parameter2009=.05;

Then include the options structure in your cplexmilp() or other toolbox optimization function call.

EdKlotz

213 Posts

Re: How to terminate the process using mip relative gap

Yesterday 5:08 PM  in response to EdKlotz

Let me just add one more bit of information to my previous post.   The problem involves only CPLEX 12.6; older versions don't have this.  

We will do a fixpack that will address this, and we'll update the forum when it is available.

• Log in to reply.

Originally posted by: Mark L. Stone

That day has apparently arrived.

Per https://delivery04.dhe.ibm.com/sar/CMA/WSA/04nlf/0/12.6.0-WS-COS-readme-fp001.html ,

Issues addressed in the fix pack version 12.6.0.1

...

RS01546  Version 12.6 of the MATLAB optimization toolbox ignores CPLEX parameter settings that lack a corresponding toolbox parameter.

....