Changing tolerence

Solver related questions
Post Reply
Ebru
User
User
Posts: 9
Joined: 10 months ago

Changing tolerence

Post by Ebru » 9 months ago

Hello,

I have a mixed integer model. GAMS outputs my binary variables as decimals. For this, I guess I need to change the integrality tolerance value. But I don't know how to do it. Do you have a code?

User avatar
bussieck
Moderator
Moderator
Posts: 859
Joined: 6 years ago

Re: Changing tolerence

Post by bussieck » 9 months ago

There is no GAMS option for this. This is solver specific. Check the solver documentation of the used solved and set the option (if available). For example, for GAMS/Cplex this option is called epint. For most solvers the default is 1e-6 and is usually good enough. Reducing this option (e.g. epint) can have serious impact on performance.

-Michael

Ebru
User
User
Posts: 9
Joined: 10 months ago

Re: Changing tolerence

Post by Ebru » 9 months ago

I'm new to use Gams. Sorry for that, cause I can not understand completely what you mean. My solver is CONOPT (non-linear problem effect).
If there is not a Gams option, how can I fix this tolerance problem?

For example, I have a binary variable. And solver gives a solution as 0.8. Thats a huge range for my problem. I want to change the tolerance as 0.

I will be very very appriaciate if you help me🙏🙏

User avatar
bussieck
Moderator
Moderator
Posts: 859
Joined: 6 years ago

Re: Changing tolerence

Post by bussieck » 9 months ago

Conopt does not solve models with binary variables. -Michael

Ebru
User
User
Posts: 9
Joined: 10 months ago

Re: Changing tolerence

Post by Ebru » 9 months ago

Hi again Michael,

Did I say wrong which solver it is? Because the model works and does not give an error. but the result is not binary. Trying to solve it with MINLP.

So what i have to do for getting a sensible results? :(

User avatar
dirkse
Moderator
Moderator
Posts: 194
Joined: 6 years ago
Location: Fairfax, VA

Re: Changing tolerence

Post by dirkse » 9 months ago

Hello,

Have you verified that the solver completed successfully? If the solver terminates prior to finding a solution (say, because of a time limit or iteration limit, or because it declares the model infeasible) then the level values returned by the solver for binary variables might not be close to 0 or 1.

-Steve

Ebru
User
User
Posts: 9
Joined: 10 months ago

Re: Changing tolerence

Post by Ebru » 9 months ago

Hi Steve,

Thank you very much your attention.

I want to add my output panel here as attachment. I think showing it this way will make it clearer because I don't understand exactly why the iterations are terminated. But the infeasible option seems to be correct.
image.png

Post Reply