The technique you described of relaxing all but 1-2 days and fixing the
integer part sounds interesting. Are there any published works implenting
the technique. I am also working on a timetabling/student enrollment
problem at my college.
On Wed, Jun 13, 2018, 9:50 AM Michael Hennebry <
> On Tue, 12 Jun 2018, João Pedro de Sá Moreira wrote:
> > I?m currently working on a timetable model that is supposed to deal if a
> pretty big number of variables.
> > On initial tests i can see it takes a fair time to get to a LP solution
> (about 30m) and then it starts to work on finding the optimal integer
> > The problem is that in this part, it takes about 5h working on analyzing
> the whole tree, but the strange thing is that it keeps showing the same
> value every iteration it does until it gets to the end of the tree. The
> value it shows is the same value that can be seen when it finds the lp
> solution, does it means that the LP solution was the optimal since the
> beginning? Am i able to interrupt the execution and take a look at this
> solution? Or is there a way to speed it up a little bit?
> I suspect the problem is a lot of near-equivalent solutions.
> Best-first can be truly horrible on some such things.
> Try depth-first.
> If you have what amounts to 40 loosely connected subproblems,
> each with five nearly optimal solutions.
> Best-first could grind away for a very long time.
> Another possibility is to do the problem in stages.
> For the first stage,
> require only the first two days' variables to be integer.
> For the second stage,
> fix the first day's integers from the first stage
> and require only the second and third days' variables to be integer.
> If you know the problem has a solution,
> how do you know?
> That might provide a mechanism for getting a solution.
> Help-glpk mailing list
Help-glpk mailing list