Callback for customized stopping rule based on Barrier dual bound for LP in Julia

Hi,

I am trying to create a callback function that allows me to stop a Barrier algorithm for a linear program prematurely based on some predefined threshold for the dual bound.

In a larger decomposition algorithm for two-stage stochastic programs, I have a current best objective of, say, 1000. A new candidate solution is sent to a potentially costly LP for evaluation. The evaluation only needs to be precise if the LP objective value is below the current best objective of 1000. Consequently, if the dual bound in the Barrier algorithm solving the LP exceeds 1000, I can be sure that I will not improve the current best with the given iterate. Hence, I want to stop the Barrier prematurely and still obtain the current primal bound and the corresponding primal and dual variable values.

I have been using the syntax provided for the Gurobi.jl wrapper in Julia: https://github.com/jump-dev/Gurobi.jl#callbacks

I have come up with this so far:

# builds a large-scale linear program for energy system optimization in JuMP
m = Model(Gurobi.Optimizer)

define_model!(m,dtr,maxhours=8760) 

set_optimizer_attribute(m,"Method",2)

set_optimizer_attribute(m,"Crossover",0)

c = []  # container collecting all dual objective values (probably not very efficient)

thr = 1000 # this is the threshold as defined in a previous iteration of the decomp algorithm

function absolute_stopping_criterion(cb_data,cb_where)
    objbound =Ref{Cdouble}()
    Gurobi.GRBcbget(cb_data,cb_where,Gurobi.GRB_CB_BARRIER_DUALOBJ,objbound)
    objbound = objbound[]
    push!(c,objbound)
    if maximum(c) > thr
        Gurobi.GRBterminate(backend(m))
    end
end

MOI.set(m,Gurobi.CallbackFunction(),absolute_stopping_criterion)
optimize!(m)

I get the following error once the threshold is reached:

ERROR: MethodError: no method matching unsafe_convert(::Type{Ptr{Nothing}}, ::MathOptInterface.Utilities.CachingOptimizer{MathOptInterface.Bridges.LazyBriingOptimizer{MathOptInterface.Bridges.LazyBridgeOptimizer{Gurobi.Optimizer}, MathOptInterface.Utilities.Unive{Float64}}})rsalFallback{MathOptInterface.Utilities.Model{Float64}}})
Closest candidates are:                                 
  unsafe_convert(::Union{Type{Ptr{Nothing}}, Type{Ptr{Base.Libc.FILE}}}, ::Base.Libc.FILE) at ..\Local\Programs\Julia-1.7.2\share\julia\base\libc.jl:94                              pData\Local\Programs\Julia-1.7.2\share\julia\
  unsafe_convert(::Type{Ptr{T}}, ::SharedArrays.SharedArray{T}) where T at ..\Local\Programs\Julia-1.7.2\share\julia\stdlib\v1.7\SharedArrays\src\SharedArrays.jl:361                ta\Local\Programs\Julia-1.7.2\share\julia\std
  unsafe_convert(::Type{Ptr{T}}, ::SharedArrays.SharedArray) where T at ..\Local\Programs\Julia-1.7.2\share\julia\stdlib\v1.7\SharedArrays\src\SharedArrays.jl:362
  ...

It appears to relate to Gurobi.GRBterminate(). Has there been a change to the command for stopping a solver instance?

Thank you very much!

Best regards,

Felix

Felix Schmidt

September 29, 2023 10:29

Edit: I managed to implement the custom stopping criterion and retrieve the associated primal objective value. But since the TerminationStatusCode is INTERRUPTED and not USERLIMIT I cannot retrieve variable values. How does one do that?

Best,

September 29, 2023 12:46

I have created a reproducible example here: Solver-specific callback for custom termination rule for LP Barrier using Gurobi and JuMP - Specific Domains / Optimization (Mathematical) - Julia Programming Language (julialang.org)

David Torres Sanchez

Gurobi Staff

October 02, 2023 10:01

Hi Felix,

Thanks for the example.

Not sure why but this works in Python.
I will see if there's a workaround.

Cheers,
David

October 02, 2023 10:10

Hi David,

thank you very much.

Best

October 02, 2023 13:03

For reference see:

Discourse (example and hack).
Gurobi.jl#19 (to track fix)

October 02, 2023 13:17

Thank you!

Callback for customized stopping rule based on Barrier dual bound for LP in Julia

Comments

Didn't find what you were looking for?