Remove constraints in LP solver

[AnirudhNarsipur]

I'm a teaching assistant for a course on constraint programming where we'd like students to implement the Branch&Bound algorithm using OR-Tools as a LP solver backend. To do so it would be nice if OR-Tools exposed an interface to remove constraints.

As far as I can see there's no way to remove constraints from a LP solver. Ideally I'd like to do something like:

    solver = pywraplp.Solver.CreateSolver("GLOP")
    x = solver.NumVar(0, solver.infinity(), "x")  
    c = x <= 10
    solver.Add(c)
    solver.Remove(c) # What I want to do

What's the best (if any) way to accomplish this? I tried to just rebuild the entire model anytime a constraint needed to be removed but that led to seg faults.

I'm using the Python bindings for the latest version of OR-Tools. If a solution is exposed in the Java bindings (nothing that I could see from the documentation) that would be fine as well.

#3926 discusses the same issue but doesn't provide a response as the author of that thread had different objectives.

[lperron]

you can change the bounds of the variables directly, no need to add a constraint.

[skadio]

Thank you, @lperron, for the prompt reply!

The above is just one example, which happens to have a single variable in the constraint. In that case, changing the bound would work (there is no need to add/remove a constraint).

In general, can we add/remove constraints in the Python API?

[rma350]

You can delete variables/constraints with the MathOpt API. However, many sovlers (including GLOP/Gurobi) are not so fast at doing incremental solves in response to a deletion, IIRC (I haven't looked at this in a few years). If you are only deleting a single varaible, it is often faster to force it to be zero instead by changing the bounds. If you have a large number of deletions to do (cut/column management) and memory is an issue, deletion makes more sense (you still may want to batch the deletions, this is solver dependent).

[skadio]

I think, I am tracking what you mean --thank you @rma350.

The example above might be misleading (as it only has a single variable). We are interested in adding/removing constraints in general, if that's possible (understand that incremental solve might not be fast).