JuliaOpt/JuMP-dev

Developer-focused chat about JuMP, M̶a̶t̶h̶P̶r̶o̶g̶B̶a̶s̶e̶, MathOptInterface, solver interfaces, and JuMP extensions. Towards JuMP 1.0! Basic usage questions will be answered more quickly at https://discourse.julialang.org/c/domain/opt.

Carleton Coffrin

@ccoffrin

@mlubin, my attempt to distill key points, JuliaOpt/JuMP.jl#2138

Carleton Coffrin

@ccoffrin

@odow, have solver tests for ResultCount made it into MOI? Should this be added to MINLPTests.jl? I just noticed that Juniper managed to avoid implementing this.

Carleton Coffrin

@ccoffrin

Also in the case of infeasibility Ipopt returns, ResultCount()=1, "termination_status" => LOCALLY_INFEASIBLE, "dual_status" => UNKNOWN_RESULT_STATUS, "primal_status" => INFEASIBLE_POINT

Just wanted to check that is the correct semantics.

Benoît Legat

@blegat

have solver tests for ResultCount made it into MOI?

Yes, it should be implemented by most solvers now

Does the dual and primal return something else that NaN for Ipopt ?

Ole Kröger

@Wikunia

ResultCount will be in the next release of Juniper.

Oscar Dowson

@odow

Juniper should implement the MOI tests: https://github.com/lanl-ansi/Juniper.jl/blob/master/test/MOI_wrapper.jl

It currently skips a lot of unit tests, and doesn't run the linear, integer, quadratic, or model tests. See Gurobi.jl for details: https://github.com/JuliaOpt/Gurobi.jl/blob/9ae498c08c4bdcfc9895638758576bc6148d3f89/test/MOI_wrapper.jl#L53-L173

It looks like the reason is because Juniper doesn't support looking up variable/constraints by name? If so, fix that and you will get access to a much broader set of tests. No need to add to MINLPTests

It's likely that Juniper is also failing to implement (or implementing incorrectly) other attributes because it isn't running these tests.

Ole Kröger

@Wikunia

Thanks will have a look into these!

Carleton Coffrin

@ccoffrin

@blegat, in that case Ipopt does populate in primal variables with values from the stationary point that it converged to.

Miles Lubin

@mlubin

@ccoffrin That's the correct semantics. Ipopt returns an infeasible point, which is exactly what the primal status reports.

Benoît Legat

@blegat

@ccoffrin In that case, it's correct what Ipopt returns

Carleton Coffrin

@ccoffrin

Great. Thanks!

Jared Miller

@Jarmill

Hey all. New to Julia and JuMP. I'm trying to export a semidefinite program as formulated by JuMP (and solved by mosek) to file, so I can compare with other solvers. Specifically looking to dump to SDPA. I found this https://github.com/JuliaOpt/MathOptInterface.jl/blob/master/src/FileFormats/FileFormats.jl, but am getting package errors ("SDPA not defined") when I try and use it.

Code is:
' @objective(model, Max, lower)

println("About to Export")
sdpa_model = MathOptInterface.SDPA.Model()
MOI.copy_to(sdpa_model, JuMP.backend(model))
MOI.write_to_file(sdpa_model, "./LR_export.sdpa")'

Thanks in advance

Does there exist functionality in Julia/JuMP to do this? Am I calling it wrong?

Benoît Legat

@blegat

It's MathOptInterface.FileFormats.SDPA

Jared Miller

@Jarmill

so 'sdpa_model = MathOptInterface.FileFormats.SDPA.Model()' ?

Benoît Legat

@blegat

Yes

Jared Miller

@Jarmill

It still says 'ERROR: LoadError: UndefVarError: FileFormats not defined'

am i missing an import or include call?

Oscar Dowson

@odow

It's not a tagged version yet

You'll probably need jump master

Then you can use http://www.juliaopt.org/JuMP.jl/dev/solvers/#File-formats-1

Jared Miller

@Jarmill

Thanks for your help and patience. I'm making edits to a file in a module to perform the export. Is there a way to reload the module without restarting julia? when I do Pkg.update() or Pkg.import() it says 'ERROR: The following package names could not be resolved:...'

Now it is 'ERROR: LoadError: UndefVarError: write_to_file not defined'

Oscar Dowson

@odow

See Revise.jl: https://github.com/timholy/Revise.jl

If you have the latest MOI release (v0.9.9), but not JuMP#master, you can also go:

sdpa_model = MOI.FileFormats.Model(filename = "LR_export.sdpa")
bridged_model = MOI.Bridges.full_bridge_optimizer(sdpa_model, Float64)
MOI.copy_to(bridged_model, JuMP.backend(model))
MOI.write_to_file(sdpa_model, "./LR_export.sdpa")

Jared Miller

@Jarmill

'MathOptInterface.UnsupportedConstraint{MathOptInterface.VectorOfVariables,MathOptInterface.PositiveSemidefiniteConeTriangle}: MathOptInterface.VectorOfVariables-in-MathOptInterface.PositiveSemidefiniteConeTriangle constraint is not supported by the model.'

might be because this is a sum-of-squares program to be processed into an SDP?

Benoît Legat

@blegat

This is unexpected, could you post a MWE as an issue in https://github.com/JuliaOpt/MathOptInterface.jl ?

Oscar Dowson

@odow

Is this with the bridged_model formulation? Or the original.

Ole Kröger

@Wikunia

Can someone confirm that JuMP and MOI handle variable indices in a deterministic way? I currently have a set constraint and the indices in the VectorOfVariables are in a different order than in the @constraint part as well as sometimes they are not only in a different order they seem to be completely different. They seem to be consistent over several runs in the same julia instance but if I restart julia everything is changed. I'm sure it has to be something in my ConstraintSolver as I suppose you implemented it in a deterministic way but before I check every single line of my code it would be nice to hear that it should be indeed deterministic

Ole Kröger

@Wikunia

i.e if I print the indices before I use them in @constraint with x.index.value I get these variables indices: 1,10,19,2,11,20,3,12,21 but in the add_constraint call they are: 1,2,3,4,5,6,7,8,9 when I print Int[vi.value for vi in vars.variables] with vars::MOI.VectorOfVariables

Ole Kröger

@Wikunia

Okay for the latter case the problem is that @variable doesn't actually add variables directly but it adds variables in the order the constraints need them which I find odd in general but ok. Nevertheless the there is something more than just the order of the constraints as I don't change them. Will have a deeper look tmr

Jared Miller

@Jarmill

Bridged model

when i get home I can send you the relevant code

Kaarthik Sundar

@kaarthiksundar

@odow When I create a callback function and invoke JuMP.callback_value(cb, x) I get the error ERROR: LoadError: UndefVarError: callback_value not defined. What is the issue here?

Oscar Dowson

@odow

You probably have an old version of jump

Kaarthik Sundar

@kaarthiksundar

I checked I have 20.1

Hack that worked: x_val = MOI.get(JuMP.backend(JuMP.owner_model(x)), MOI.CallbackVariablePrimal(cb), JuMP.index(x))

Benoît Legat

@blegat

You need to use JuMP master

Miles Lubin

@mlubin

@/all NumFOCUS is again offering to coordinate GSOC for JuMP, and we need to let them know if we're participating. Please start thinking about if you'd like to mentor a project. We'll discuss more during the developer call next week.

Joaquim Dias Garcia

@joaquimg

Awesome!

Joaquim Dias Garcia

@joaquimg

JuMP is reeally close to 1k stars
this kind of graph will be nice for twitter/blog/webpage : https://star-history.t9t.io/#JuliaOpt/JuMP.jl

Oscar Dowson

@odow

https://star-history.t9t.io/#Pyomo/pyomo&Juliaopt/jump.jl

Joaquim Dias Garcia

@joaquimg

nice, cvxpy is outstanding btw
https://star-history.t9t.io/#Pyomo/pyomo&Juliaopt/jump.jl&yalmip/YALMIP&cvxgrp/cvxpy&JuliaOpt/Convex.jl

Where communities thrive

People

Repo info

Activity