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.

BridgingBot

@GitterIRCbot

[slack] <mtanneau> I wouldn’t be surprised if solver companies were accommodating to non-profit as well.

[slack] <pure_interpeter> Mhh, i guess i will write with them didn't dear to ask.

[slack] <mtanneau> https://or.stackexchange.com/questions/4991/combinatorial-problem-in-my-daughter-s-class/4993#4993

[slack] <pure_interpeter> This problem seems very different from what i have in mind. Mine is an relaxation of balanced Minimum Cut with additional conditions.

BridgingBot

@GitterIRCbot

[slack] <pure_interpeter> I am not sure what you are trying to tell me with this link.

BridgingBot

@GitterIRCbot

[slack] <mtanneau> It was a supporting argument: the last answer is from LocalSolver and says “LocalSolver is a commercial product but if you're interested to use it for free to solve this problem, you will be welcome”.

BridgingBot

@GitterIRCbot

[slack] <pure_interpeter> Ah, thanks i wasn't aware of that.

BridgingBot

@GitterIRCbot

[slack] <pure_interpeter> Is there a place for a place where we can list solver not yet tied to JuMP and discuss their merit (for example an GitHub issue documentation file)?

Oscar Dowson

@odow

This channel is an appropriate place to discuss solvers you want to add a JuMP interface to. (I recently clarified this in the documentation. See the note at the bottom of this section: https://jump.dev/JuMP.jl/dev/installation/#Supported-solvers)

BridgingBot

@GitterIRCbot

[slack] <matbesancon> @pure_interpeter I'm a bit late to the party, but another important aspect is that for many problems, proving optimality is what takes a lot of time, an optimal solution has often been found early in the MILP solution process. So with large enough tolerance gaps, this means you will get high quality solutions (often the best one)

BridgingBot

@GitterIRCbot

[slack] <pure_interpeter> There is no being late to the party.

BridgingBot

@GitterIRCbot

[slack] <ExpandingMan> that may not even be the case for all I know, but I do know that academic licenses are a very common thing, and if you go to a website they will have buttons for you to click on that say "academic", but if any of them offer anything for non-profits, I wouldn't even know how to find that out. Who knows, might be worth contacting someone

BridgingBot

@GitterIRCbot

[slack] <pure_interpeter> Is there any reasonable explanation why the same ILP problem has two differently good optimas between Gurobi and CBC?
Ok it is just CBC being drunk. CBC gives a different answer (but saying "Optimal solution found.") while Gurobi always gives the same.

BridgingBot

@GitterIRCbot

[slack] <mtanneau> If they have significantly different objective values, there's a problem.

Gurobi should be run-to-run deterministic, i.e., for the same problem on the same machine, it will return the same answer. That may not be true if, e.g., a time limit is hit (because there is randomness in computing times) or a non-deterministic concurrent optimizer is used.

I don't know whether Cbc is also deterministic. If you use multiple threads or you hit a time limit then it's possible that it's not.

BridgingBot

@GitterIRCbot

[slack] <pure_interpeter> I did ran multi-threaded but there was no time-limit on CBC.
The run-to-run difference with CBC was >10% of the gurobi objective.

[slack] <mtanneau> Then check the validity of both solutions (the solvers may have different tolerances).
Whichever reports an invalid result has a bug 🤷

BridgingBot

@GitterIRCbot

[slack] <pure_interpeter> Yeah I blame CBC because it comes up with different solutions everytime i run it (i mean that by run-to-run difference).

Miles Lubin

@mlubin

@/all Reminder: The monthly developer call will be tomorrow (Thursday) at 2pm US Eastern time on Zoom. Ping me or @odow for an invite.

Ole Kröger

@Wikunia

One small question about the bridges I mentioned yesterday. I was able to simply use the FlipSignBridge for strictly greater than -> strictly less than constraints. Thanks for that. I currently run into a problem when I want to use GreaterThan inside an indicator constraint. The _build_indicator_constraint doesn't exist:

MethodError: no method matching _build_indicator_constraint(::JuMP.var"#_error#68"{Symbol}, ::VariableRef, ::BridgeableConstraint{ScalarConstraint{GenericAffExpr{Float64,VariableRef},ConstraintSolver.GreaterThan{Float64}},UnionAll}, ::Type{MathOptInterface.IndicatorSet{MathOptInterface.ACTIVATE_ON_ONE,S} where S<:MathOptInterface.AbstractScalarSet})

I hoped it would call the indicator constraint with just the already briged ScalarConstraint as a type as

_build_indicator_constraint(
    _error::Function, variable::AbstractVariableRef,
    constraint::ScalarConstraint, ::Type{MOI.IndicatorSet{A}})

exists. How does this currently work with bridges?

Joey Huchette

@joehuchette

I would like to build an MOI model with a collection of variables constrained to lie in Intervals. I would then like to modify some of the values of the upper/lower bounds in these intervals. Is there a straightforward way to do this using modify or something similar? Or do I need to use delete! and then re-add?

Joaquim Dias Garcia

@joaquimg

If the ser type is the same (Interval) you can just:
MOI.set(model, MOI.ConstraintSet(), ci, MOI.Interval{Float64}(lb, ub))

details here: https://jump.dev/MathOptInterface.jl/stable/manual/basic_usage/#Replacements

Joey Huchette

@joehuchette

Perfect, thank you

Benoît Legat

@blegat

BridgeableConstraint is not an AbstractScalarConstraint

We should add

function _build_indicator_constraint(
    _error::Function, variable::AbstractVariableRef,
    constraint::BridgeableConstraint, I::Type{MOI.IndicatorSet{A}}) where A
    return BridgeableConstraint(
        _build_indicator_constraint(_error, variable, constraint.constraint, I),
        constraint.bridge_type
    )
end

Ole Kröger

@Wikunia

Thanks @blegat I think I'm making progress. One more question. If I want to add a completely new bridge: How do I call JuMP.add_bridge ? That needs the ::Model but I only know ::AbstractOptimizer (my own solver structure) Do I get the model somewhere?

Ole Kröger

@Wikunia

If I would build my own bridge for my own type I could use JuMP.build_constraint but currently I would like to bridge IndicatorSet{A,GreaterThan} to IndicatorSet{A,LessThan}(so I don't own the type)

Ole Kröger

@Wikunia

Or is the correct way to use a LazyBridgeOptimizer?

Mathieu Besançon

@matbesancon

if some people are interested, I wrote two quick posts on set distances and projections here:
If interested, I just wrote about some basics of the package here:
http://matbesancon.github.io/post/2020-12-23-chains_sets/
and here
http://matbesancon.github.io/post/2020-12-24-chains_sets2/

the second post covers ChainRules integration and derivatives

Benoît Legat

@blegat

@Wikunia it seems a proper solution would need jump-dev/MathOptInterface.jl#666

Joey Huchette

@joehuchette

Was there a conscious decision to omit a Real <: AbstractScalarSet from MOI? If so, why?

Miles Lubin

@mlubin

It's not a particularly natural set for any solver. If you're looking for the dual of the zeros cone, see Reals.

Joey Huchette

@joehuchette

I'm sketching out a way to model disjunctions through MOI and it would be useful for what I'm currently thinking. Would there be resistance to a PR adding Real to MOI?

Miles Lubin

@mlubin

Can the same functionality be covered by an interval [-Inf, Inf]?

Joey Huchette

@joehuchette

It would be functionally the same, though a bit uglier. Still trying to work out how much I care about that, though.

Benoît Legat

@blegat

Why does it need to be a scalar set? To be used with indicator?

Joey Huchette

@joehuchette

My thought was to create a vector set by stacking some number of scalar sets together

but I'm rethinking the design a bit

I just found it a bit odd there was no scalar counterpart

Ole Kröger

@Wikunia

One more question about bridges: Why does one need to define an inverse_map_set ? Under what circumstances is that called? i.e in the GreaterToLessBridge one wants the map_set when going from >= to <=, right? For the other way we have the LessToGreaterBridge if I'm not mistaken. Is it used for dual variables?

Benoît Legat

@blegat

Getting primal solution and primal start

https://github.com/jump-dev/MathOptInterface.jl/blob/b254c292bbd5f3ac2dff40fe5e8c2cd3ea3ab612/src/Bridges/Constraint/set_map.jl#L101

Getting function

https://github.com/jump-dev/MathOptInterface.jl/blob/b254c292bbd5f3ac2dff40fe5e8c2cd3ea3ab612/src/Bridges/Constraint/set_map.jl#L63

By primal solution, I mean MOI.ConstraintPrimal

These operations are not use extensively so if you don't define the inverse map, you might not notice it directly

Ole Kröger

@Wikunia

Great thanks! @blegat

Ole Kröger

@Wikunia

It's possible to get the name of a variable by the VariableIndex, correct? I see the set and get VariableName methods but don't see where the set method is called as it isn't in my case. I allow copy_names and I have the supports function for it. Anything I'm missing?

Where communities thrive

People

Repo info

Activity