BOP solver: Using number_of_solvers results in abnormal status

[syxolk]

Hi, I am current trying to use the multi-threading feature of the BOP solver with the pywraplp frontend. I'm using Python 3.7.4 with OR-Tools 7.3.7083.

The following code results in an ABNORMAL result status:

from ortools.linear_solver import pywraplp

def main():
    solver = pywraplp.Solver('test', pywraplp.Solver.BOP_INTEGER_PROGRAMMING)
    solver.EnableOutput()
    solver.SetSolverSpecificParametersAsString("number_of_solvers: 2 synchronization_type: SYNCHRONIZE_ALL")

    x = solver.IntVar(0, 10, "")
    y = solver.IntVar(0, 10, "")
    solver.Add(y == 2*x)
    solver.Maximize(y)

    result_status = solver.Solve()
    print(f"{pywraplp.Solver.OPTIMAL} == {result_status}")
    if result_status == pywraplp.Solver.OPTIMAL:
        print(f"x = {x.SolutionValue()}")
        print(f"y = {y.SolutionValue()}")

main()

Output:

0 == 4

However, when I comment out the solver.SetSolverSpecificParametersAsString line or use number_of_solvers: 1 instead it works as expected:

WARNING: Logging before InitGoogleLogging() is written to STDERR
I0815 11:40:39.133632 20305 bop_portfolio.cc:147]       30 .. inf Portfolio - SATRandomFirstSolution. Time limit: inf -- inf
I0815 11:40:39.133743 20305 bop_portfolio.cc:79] Stats. #new_solutions/#calls by optimizer:
                      SATRandomFirstSolution :   0/1    (  0.00%)  Total gain:      0  Total Dtime: 0.000 score: 0.000001
0 == 0
x = 5.0
y = 10.0

[lperron]

Are you sure this is a supported parameter? I guess not, and the solver tells you it could not parse this string.

[lperron]

And really, you should not use BOP for that, BOP is for Boolean problems. The support for integer variables is slow and experimental.

In fact, you should use the CP-SAT solver directly:

from ortools.sat.python import cp_model

def main():
    model = cp_model.CpModel()

    x = model.NewIntVar(0, 10, "")
    y = model.NewIntVar(0, 10, "")
    model.Add(y == 2*x)
    model.Maximize(y)

    solver = cp_model.CpSolver()

    result_status = solver.Solve(model)
    print(f"status = {solver.StatusName(result_status)}")
    if result_status == cp_model.OPTIMAL:
        print(f"x = {solver.Value(x)}")
        print(f"y = {solver.Value(y)}")

main()

which outputs

status = OPTIMAL
x = 5
y = 10

[lperron]

to add Parallelism, just add the line before the Solve() method

solver.parameters.num_search_workers = 8

[syxolk]

Thanks for the suggestion with CP-SAT but I'm currently comparing different MIP solvers, so I want to use BOP.

Are you sure this is a supported parameter?

Yes, I've discovered it here: https://github.com/google/or-tools/blob/v7.3/ortools/bop/bop_parameters.proto#L184

  // The number of solvers used to run Bop. Note that one thread will be created
  // per solver. The type of communication between solvers is specified by the
  // synchronization_type parameter.
  optional int32 number_of_solvers = 24 [default = 1];

I guess not, and the solver tells you it could not parse this string.

It actually does not give me any error message.

When I do something obviously wrong like:

solver.SetSolverSpecificParametersAsString("foobar: 1")

I get an error message:

[libprotobuf ERROR /home/hans/projects/or-tools/dependencies/sources/protobuf-3.9.0/src/google/protobuf/text_format.cc:324]
Error parsing text-format operations_research.bop.BopParameters:
1:7: Message type "operations_research.bop.BopParameters" has no field named "foobar".

But this works just fine without error message:

solver.SetSolverSpecificParametersAsString("number_of_solvers: 2 synchronization_type: SYNCHRONIZE_ALL")

[syxolk]

So, is this an official feature of BOP or not?