[Coin-discuss] CBC applied to maximization problems

[Hart, William E]

> > I'm working on a CBC wrapper, and I'm not sure how to 
> interpret CBC when it is applied to a maximization problem 
> that is read-in from a *.lp file.  It appears that CBC 
> automatically negates the objective and solves the 
> corresponding minimization problem.
> >
> > . Is there a way to force CBC to solve the problem as a 
> maximization problem?
> 
> This has little to do with Cbc. CoinLpIO is flipping the 
> objective function when reading the Lp file. Cbc does not 
> know that the original lp file was a maximization.
> 
> The reason for flipping the objective to get a minimization 
> is that orginially, only MPS files were used as input and MPS 
> files are always minimization problems. To avoid problem with 
> solvers that assumed that the input came from an MPS file, I 
> made LpIO flip the objective function.

Ah.  That makes sense.


> > . Is there a way to force CBC to indicate that it negative 
> the objective?
> >
> > Without these hooks, I'm struggling with how to automate 
> the process of converting the CBC result back into the answer 
> for the original maximization problem...
> 
> Why? Isn't flipping the sign of the optimal value good 
> enough? Is there a constant term in the objective function?

Here's my scenario.  I have a wrapper script that passes a LP or MPS file to CBC, and then parses the output.  The results of this optimization process are then returned to the user.  The confusing thing is that a user sees minimization results when passed a maximization problem.  Logically, it's identical, but this will be surprising and confusing to non-technical users.

If CBC/CoinLpIO simply printed a warning that this problem was being reformulated, that would be enough for me to correctly parse the output.  (Working with the library directly, this wouldn't be an issue because I could check the max/min sense of the original problem...).

--Bill