Unable to set Integer options to infinity

[joeltay]

Because Inf is floating point, this code in grb_params.jl fails when trying to set something in GRB_INT_PARAMS to infinity.

function set_int_param!(env::Env, name::ASCIIString, v::Integer)
ret = @grb_ccall(setintparam, Cint, (Ptr{Void}, Ptr{Cchar}, Cint),
env, name, v)
if ret != 0
throw(GurobiError(env, ret))
end
end

[IainNZ]

What parameter are you trying to set to Inf?

[joeltay]

SolutionLimit using JuMP.

[joeltay]

E.g.

using JuMP, Gurobi
m = Model(solver=GurobiSolver(SolutionLimit=Inf))
solve(m)

fails with

ERROR: `set_int_param!` has no method matching set_int_param!(::Env, ::ASCIIString, ::Float64)
 in setparam! at /home/joel/.julia/Gurobi/src/grb_params.jl:208
 in GurobiMathProgModel at /home/joel/.julia/Gurobi/src/GurobiSolverInterface.jl:18
 in model at /home/joel/.julia/Gurobi/src/GurobiSolverInterface.jl:29
 in solveLP at /home/joel/.julia/JuMP/src/solvers.jl:210
 in solve at /home/joel/.julia/JuMP/src/solvers.jl:46

[joeltay]

Just to note that there do exist integer parameters which are not set to Infinity by default but which should be able to, e.g. CutPasses.

[IainNZ]

Ah I see, OK.
So, its interesting, because it says at http://www.gurobi.com/documentation/5.6/reference-manual/solutionlimit that Infinity is a valid value, but no such concept exists in C (for integers) nor Julia. What does exist is GRB_INFINITY, which I thought was the floating point value 1e100, so thats not good...

[IainNZ]

Ahah, gurobi_c.h has #define GRB_MAXINT 2000000000

[IainNZ]

So that is fairly easy to fix, we can just create another method that takes floating point arguments, checks for a number bigger than GRB_MAXINT and calls the int version with GRB_MAXINT, otherwise error.

[mlubin]

That seems a bit messy on our side. Maybe just export GRB_MAXINT?

[pajalevi]

Hello from the future!
I'm getting this same error message noted here, when implementing the code for benders cuts found here in the JuliaOpt teaching notebooks

I've copy-pasted the code in the notebook above for the lazy constraints version, including test data. The only changes I've made are to update it to use current JuMP conventions around setting constraints etc (@constraint instead of @addConstraint etc) and current notation for sums (sum( for ) instead of sum{ , }), and to use Gurobi as the solver.

solve(masterProblemModel) returns the following error message:

ERROR: MethodError: no method matching set_int_param!(::Gurobi.Model, ::String, ::Float64)
Closest candidates are:
  set_int_param!(::Gurobi.Model, ::String, ::Integer) at /Users/patricia/.julia/v0.6/Gurobi/src/grb_params.jl:266
  set_int_param!(::Gurobi.Env, ::String, ::Integer) at /Users/patricia/.julia/v0.6/Gurobi/src/grb_params.jl:192
Stacktrace:
 [1] setparam!(::Gurobi.Model, ::String, ::Float64) at /Users/patricia/.julia/v0.6/Gurobi/src/grb_params.jl:319
 [2] setparams!(::Gurobi.GurobiMathProgModel) at /Users/patricia/.julia/v0.6/Gurobi/src/GurobiSolverInterface.jl:58
 [3] #GurobiMathProgModel#14(::Array{Any,1}, ::Type{T} where T, ::Void) at /Users/patricia/.julia/v0.6/Gurobi/src/GurobiSolverInterface.jl:27
 [4] (::Core.#kw#Type)(::Array{Any,1}, ::Type{Gurobi.GurobiMathProgModel}, ::Void) at ./<missing>:0
 [5] LinearQuadraticModel(::Gurobi.GurobiSolver) at /Users/patricia/.julia/v0.6/Gurobi/src/GurobiSolverInterface.jl:68
 [6] #build#119(::Bool, ::Bool, ::JuMP.ProblemTraits, ::Function, ::JuMP.Model) at /Users/patricia/.julia/v0.6/JuMP/src/solvers.jl:353
 [7] (::JuMP.#kw##build)(::Array{Any,1}, ::JuMP.#build, ::JuMP.Model) at ./<missing>:0
 [8] #solve#116(::Bool, ::Bool, ::Bool, ::Array{Any,1}, ::Function, ::JuMP.Model) at /Users/patricia/.julia/v0.6/JuMP/src/solvers.jl:168
 [9] addBendersCut(::Gurobi.GurobiCallbackData) at /Users/patricia/Documents/Google Drive/stanford/second yearpaper/Tutorial II/Julia_scripts/testbenders_lazy.jl:100
 [10] lazycallback(::Gurobi.GurobiCallbackData, ::JuMP.Model, ::Array{JuMP.LazyCallback,1}) at /Users/patricia/.julia/v0.6/JuMP/src/callbacks.jl:78
 [11] (::JuMP.##140#141{JuMP.Model,Array{JuMP.LazyCallback,1}})(::Gurobi.GurobiCallbackData) at /Users/patricia/.julia/v0.6/JuMP/src/callbacks.jl:96
 [12] mastercallback(::Ptr{Void}, ::Ptr{Void}, ::Int32, ::Ptr{Void}) at /Users/patricia/.julia/v0.6/Gurobi/src/GurobiSolverInterface.jl:695
 [13] optimize(::Gurobi.Model) at /Users/patricia/.julia/v0.6/Gurobi/src/grb_solve.jl:5
 [14] optimize!(::Gurobi.GurobiMathProgModel) at /Users/patricia/.julia/v0.6/Gurobi/src/GurobiSolverInterface.jl:417
 [15] #solve#116(::Bool, ::Bool, ::Bool, ::Array{Any,1}, ::Function, ::JuMP.Model) at /Users/patricia/.julia/v0.6/JuMP/src/solvers.jl:175
 [16] solve(::JuMP.Model) at /Users/patricia/.julia/v0.6/JuMP/src/solvers.jl:150
 [17] macro expansion at /Users/patricia/.julia/v0.6/Atom/src/repl.jl:118 [inlined]
 [18] anonymous at ./<missing>:?

I'm having trouble identifying the place where set_int_param! is called with a Float argument instead of an Integer, so I'm not quite sure why this is happening but it seems that it should not be an available option within the Gurobi wrapper.

It looks like the commit referencing this issue was never merged into the master file, which might be why it is still a problem. Any advice for how to work around this problem or fix it would be greatly appreciated!

thanks
Patricia

[joeltay]

This error occurs because the Gurobi model is being initialized with some parameter that only takes integers, but is being set to a float.

In my case, it sufficed to set it to a sufficiently large integer; I don't think there is a solution in Julia for if you really need Inf.

In your case, there are only two cases where the Model() function is called, so it was easy to check. Try changing this line:

subProblemModel = Model(solver=GurobiSolver(Presolve=0.0))

to Presolve=0

http://www.gurobi.com/documentation/8.0/refman/presolve.html#parameter:Presolve

[pajalevi]

Ah, such a simple solution - Thanks so much for your speedy help. It didn't occur to me that the 'int_param' in question was one that I had set in my own code.

I changed it per your suggestion and everything ran perfectly. Excellent!