[SCIP] Resolving with a different objective function

[Victor Miller]

At the request of Ambros I'm posting the python program below which
forbids me from reoptimizing.  I'm using the scip version 7.0.0 with
the pyscipopt from the github.

In python I do

foo = list(bug(5,5, ntimes=2, vtype='B')

which gives the run time error at the second iteration

[scip_var.c:4546] ERROR: invalid SCIP stage <10>
[scip_prob.c:1150] ERROR: Error <-8> in function call
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/victorsmiller/Programming/Optimization/Reopt/scip_bug.py",
line 20, in bug
    model.chgReoptObjective(quicksum(obj_vec[_] * vars[_] for _ in
range(ndim)), "minimize")
  File "src/pyscipopt/scip.pyx", line 4536, in
pyscipopt.scip.Model.chgReoptObjective
  File "src/pyscipopt/scip.pyx", line 245, in pyscipopt.scip.PY_SCIP_CALL
Exception: SCIP: method cannot be called at this time in solution process!

----------
import numpy as np
from pyscipopt import Model, quicksum

def bug(mdim, ndim, ntimes=1, vtype='B'):

    model = Model()
    amat = np.random.randn(mdim, ndim)
    xvec = np.random.randint(0, 2, ndim) if vtype == 'B' else
np.random.randn(ndim)
    bvec = amat @ xvec
    vars = [model.addVar(vtype=vtype, name="x_%d" % _) for _ in range(ndim)]

    for ind in range(mdim):
        model.addCons(quicksum(amat[ind, _] * vars[_] for _ in
range(ndim)) == bvec[ind])

    model.enableReoptimization()

    for ind in range(ntimes):

        obj_vec = np.random.randn(ndim)
        model.chgReoptObjective(quicksum(obj_vec[_] * vars[_] for _ in
range(ndim)), "minimize")
        model.optimize()

        status = model.getStatus()

        if status == "infeasible":
            raise ValueError("The problem is infeasible")

        soln = np.array([model.getVal(vars[_]) for _ in range(ndim)])
        if vtype == 'B':
            yield soln.round().astype(np.int8)
        else:
            yield soln

On Thu, May 28, 2020 at 2:09 AM Ambros Gleixner <gleixner at zib.de> wrote:
>
> Hi Victor,
>
> Please have a look at
>
>     https://www.scipopt.org/doc/html/REOPT.php
>
> SCIP actually provides an intricate mechanism to do what exactly you
> want and reuse information from the previous solving process.  Whether
> this performs better than solving from scratch depends on how large the
> changes between different objective functions are.  Here is a scientific
> publication that gives more details:
>
>     http://dx.doi.org/10.1007/978-3-319-20086-6_14
>
> To achieve the more plain version of simply solving again from scratch,
> you have to call SCIPfreeTransform(); then you are allowed to modified
> the problem.
>
> Best,
> Ambros
>
>
>
>
> Am 27.05.20 um 23:07 schrieb Victor Miller:
> > I have a MIP which I want to solve a number of times with different
> > objective functions.  Just trying to change the objective after a
> > previous solution doesn't work, so how do I do this?  Do I copy the
> > model, and solve the copy?  If so, is there a standard way of finding
> > out which variables in the copy correspond to the variable in the
> > original?
> >
> > Victor Miller
> > _______________________________________________
> > Scip mailing list
> > Scip at zib.de
> > https://listserv.zib.de/mailman/listinfo/scip
> >
>
> --
> Ambros Gleixner, Research Group Mathematical Optimization Methods at
> Zuse Institute Berlin, http://www.zib.de/gleixner
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip