cbgetlpsolution returns assertionerror

[elchiconuevo]

I'm trying to collect some info using an infocallback illustrated in the callback section of the JuMP document. When I try to access the fractional lp solution at the root node using cbgetlpsolution within the infocallback function, I get an assertion error. This is the function I use to add the infocallback to the model.
addinfocallback(mdl, infocallback, when = :Intermediate)

The objective is to access the fractional solution similar to this. How can I achieve this in JuMP?

[odow]

Can you provide a complete exmaple that demonstrates the problem? See https://stackoverflow.com/help/mcve

[elchiconuevo]

A simple MWE (based on knapsack.jl) of the problem I have. I want to get the fractional LP value of variable x at the root node. However, I get a warning here: Variable value not defined for component of x. Check that the model was solved properly

using JuMP
using CPLEX

type NodeData
  time::Float64  # in seconds since the epoch
  node::Int
  obj::Float64
  bestbound::Float64
  x_val::Array{Float64,1}
end

m = Model(solver=CplexSolver(CPX_PARAM_NODELIM = 0))

@variable(m, x[1:5], Bin)

profit = [ 5, 3, 2, 7, 4 ]
weight = [ 2, 8, 4, 2, 5 ]
capacity = 10

# Objective: maximize profit
@objective(m, Max, dot(profit, x))

# Constraint: can carry all
@constraint(m, dot(weight, x) <= capacity)

bbdata = NodeData[]

function infocallback(cb)
  node      = MathProgBase.cbgetexplorednodes(cb)
  obj       = MathProgBase.cbgetobj(cb)
  bestbound = MathProgBase.cbgetbestbound(cb)
  x_val = getvalue(x)
  push!(bbdata, NodeData(time(),node,obj,bestbound, x_val))
end
addinfocallback(m, infocallback, when = :Intermediate)

solve(m)

In the earlier case, I queried for the lpsolution instead of the values of x. That returned me an assertionerror.

[odow]

I haven't used the callbacks in CPLEX, however it looks like the solution is only available for when=:MIPnode for the infocallback.

https://github.com/JuliaOpt/JuMP.jl/blob/37ab6795eaa269bb762eb11dbba350eb8379bfb1/src/callbacks.jl#L170

You could try addlazycallback(m, infocallback) instead.

The new version of JuMP will not support solver independent callbacks. The behaviour is too solver-dependent.

[elchiconuevo]

I haven't used the callbacks in CPLEX, however it looks like the solution is only available for when=:MIPnode for the infocallback

Is this behaviour independent of the solver? I mean if I switch to another solver, lets say Gurobi, the solution would still not be available?

[odow]

Presumably you're trying to follow https://discourse.julialang.org/t/accessing-the-root-node-solution/10371/7?

This is not a CPLEX.jl issue. The infocallback only supports querying the LP solution in the following states
https://github.com/JuliaOpt/CPLEX.jl/blob/81d7dbb9947960f56c81f7b32dcdf612fa99f7ba/src/CplexSolverInterface.jl#L373-L374

The variable values are not available for when=:Intermediate.

I'm not sure what the story with Gurobi is, but according to
https://github.com/JuliaOpt/JuMP.jl/blob/37ab6795eaa269bb762eb11dbba350eb8379bfb1/src/callbacks.jl#L165-L170
the solution will not be available for when=:Intermediate.

[elchiconuevo]

I tried addlazycallback(m, infocallback) instead of addinfocallback. While it works for the current problem, it shows the same warning for larger problem instances. The reason seems to be that for the simple knapsack problem, the optimal solution was found at node 0 itself.

[odow]

Closing this since it relates to the old JuMP callbacks, which we won't be fixing. The new MOI wrapper implements solver-independent MOI callbacks, and provides access to solver-specific generic CPLEX callbacks.