JuMP, GLPK integer decision variable value tolerance

Specific Domains Optimization (Mathematical)
,
1

Hi,

I was using GLPK within JuMP to solve an integer linear program. The value of the decision variables are close to integer values. I can just round them if they are within an accepted tolerance.

However, GLPK optimizer has parameters for integer tolerance.

julia> opt = GLPK.Optimizer()
A GLPK model

julia> opt.intopt_param.tol_int
1.0e-5

ERROR: InexactError: Int64(-101.99999999999999)
Stacktrace:
  [1] Int64
    @ ./float.jl:812 [inlined]
  [2] convert
    @ ./number.jl:7 [inlined]
  [3] _broadcast_getindex_evalf
    @ ./broadcast.jl:670 [inlined]
  [4] _broadcast_getindex
    @ ./broadcast.jl:653 [inlined]
  [5] getindex
    @ ./broadcast.jl:597 [inlined]
  [6] macro expansion
    @ ./broadcast.jl:961 [inlined]
  [7] macro expansion
    @ ./simdloop.jl:77 [inlined]
  [8] copyto!
    @ ./broadcast.jl:960 [inlined]
  [9] copyto!
    @ ./broadcast.jl:913 [inlined]
 [10] copy
    @ ./broadcast.jl:885 [inlined]
 [11] materialize
    @ ./broadcast.jl:860 [inlined]
 [12] ideal(problem::MOIntegerLinearProblem)

Is it possible to get JuMP to return integer values for these variables or rounding is necessary? Is this solver parameter merely used for integrality check?

Thanks

2

Rounding is necessary.

Is this solver parameter merely used for integrality check?

The solver parameter is used to define which solutions are feasible. A tolerance of 1e-5 means that GLPK thinks a value is “integer” if it is within +/- 1e-5 of a true integer. (Tolerances are needed because solvers do not use exact arithmetic.)

1 Like
3

Thanks!

1 Like