Update to MOI v1 #66

blegat · 2023-06-02T17:17:09Z

Closes #53

Would be helpful to have this working now generic number support is arriving in JuMP: jump-dev/JuMP.jl#3385

ericphanson · 2023-06-02T17:43:50Z

src/file_io.jl

+ optimizer.dualobj = parse(T, objValPrimalstring)
+ optimizer.primalobj = parse(T, objValDualstring)


why are these swapped?

We now use MOI.FileFormats.SDPA which expects the SDP in geometric conic form / image form while before you were doing like CSDP and SDPA which expect it to be in standard conic form / kernel form

ericphanson · 2023-06-02T17:44:55Z

src/objective_function_filter.jl

+ if !iszero(constant)
+ func = copy(func)
+ func.constant = zero(constant)
+ end


could you explain a bit more? It looks like we totally disregard the constant by throwing it away here, is that correct?

It's saved in a field of Optimizer in optimize! and added back when the user asks for the objective value.

ericphanson · 2023-06-02T17:47:44Z

test/MOI_wrapper.jl

+ # FIXME investigate
+ # Expression: isapprox(MOI.get(model, MOI.ObjectiveValue()), T(2), config)
+ # Evaluated: isapprox(5.999999984012059, 2.0, ...
+ r"test_modification_delete_variables_in_a_batch$",
+ # FIXME investigate
+ # Expression: isapprox(MOI.get(model, MOI.ObjectiveValue()), objective_value, config)
+ # Evaluated: isapprox(-2.1881334077988868e-7, 5.0, ...
+ r"test_objective_qp_ObjectiveFunction_edge_case$",
+ # FIXME investigate
+ # Expression: isapprox(MOI.get(model, MOI.ObjectiveValue()), objective_value, config)
+ # Evaluated: isapprox(-2.1881334077988868e-7, 5.0, ...
+ r"test_objective_qp_ObjectiveFunction_zero_ofdiag$",
+ # FIXME investigate
+ # Expression: isapprox(MOI.get(model, MOI.ConstraintPrimal(), index), solution_value, config)
+ # Evaluated: isapprox(2.5058846553349667e-8, 1.0, ...
+ r"test_variable_solve_with_lowerbound$",
+ # FIXME investigate
+ # See https://github.com/jump-dev/SDPA.jl/runs/7246518765?check_suite_focus=true#step:6:128
+ # Expression: ≈(MOI.get(model, MOI.ConstraintDual(), c), T[1, 0, 0, -1, 1, 0, -1, -1, 1] / T(3), config)
+ # Evaluated: ≈([0.3333333625488728, -0.16666659692134123, -0.16666659693012292, -0.16666659692134123, 0.33333336253987234, -0.16666659692112254, -0.16666659693012292, -0.16666659692112254, 0.333333362548654], [0.3333333333333333, 0.0, 0.0, -0.3333333333333333, 0.3333333333333333, 0.0, -0.3333333333333333, -0.3333333333333333, 0.3333333333333333]
+ r"test_conic_PositiveSemidefiniteConeSquare_3$",
+ # FIXME


are these copied from SDPA.jl? Some of them might pass here with the presolve (because we've found that is necessary to function correctly in some cases)

Yes but I have not updated the presolve code yet. It should be udpated to work with MOI.FileFormats.SDPA.Model. Maybe https://github.com/mtanneau/MathOptPresolve.jl could be used ? I haven't checked in details

I would like to take a look at https://github.com/mtanneau/MathOptPresolve.jl so to see if it has everything needed for convex problems and then recommend people to always try using this layer (in addition to the Dualization.jl layer which should always be tried for conic optimization).

it seems like MathOptPresolve doesn't support MOI v1, and also doesn't have docs explaining how to use it

blegat · 2023-06-06T11:14:32Z

@ericphanson Any idea about the Convex.jl failing tests ? Is this only because we removed the presolve ?

ericphanson · 2023-06-06T12:00:47Z

Yeah, I think so. It is really essential to get anywhere close to correct results in many cases.

ericphanson · 2023-06-06T13:26:12Z

Running the tests locally, I see

[ Info: Solving problem sdp_trace_logm_real with variant sdpa and type Double64
sdp_trace_logm_real: Error During Test at /Users/eph/.julia/packages/Convex/tSTAW/src/problem_depot/problems/sdp.jl:2074
  Test threw exception
  Expression: ≈(p.optval, tr(C * log(evaluate(X))), atol = atol, rtol = rtol)
  MethodError: no method matching eigen!(::Hermitian{Double64, Matrix{Double64}}; sortby::typeof(LinearAlgebra.eigsortby))
  
  Closest candidates are:
    eigen!(::Union{Hermitian{T, var"#s967"}, Hermitian{Complex{T}, var"#s967"}, Symmetric{T, var"#s967"}} where var"#s967"<:(StridedMatrix{T} where T), ::AbstractMatrix{T}; sortby) where T<:Number
     @ LinearAlgebra ~/.julia/juliaup/julia-1.9.0+0.x64.apple.darwin14/share/julia/stdlib/v1.9/LinearAlgebra/src/symmetriceigen.jl:167
    eigen!(::Hermitian; tol) got unsupported keyword argument "sortby"
     @ GenericLinearAlgebra ~/.julia/packages/GenericLinearAlgebra/skQkq/src/eigenSelfAdjoint.jl:605
    eigen!(::StridedMatrix{T}; permute, scale, sortby) where T<:Union{Float32, Float64}
     @ LinearAlgebra ~/.julia/juliaup/julia-1.9.0+0.x64.apple.darwin14/share/julia/stdlib/v1.9/LinearAlgebra/src/eigen.jl:149
    ...
  
  Stacktrace:
   [1] kwerr(::NamedTuple{(:sortby,), Tuple{typeof(LinearAlgebra.eigsortby)}}, ::Function, ::Hermitian{Double64, Matrix{Double64}})
     @ Base ./error.jl:165
   [2] eigen(A::Matrix{Double64}; permute::Bool, scale::Bool, sortby::typeof(LinearAlgebra.eigsortby))
     @ LinearAlgebra ~/.julia/juliaup/julia-1.9.0+0.x64.apple.darwin14/share/julia/stdlib/v1.9/LinearAlgebra/src/eigen.jl:237
   [3] eigen
     @ ~/.julia/juliaup/julia-1.9.0+0.x64.apple.darwin14/share/julia/stdlib/v1.9/LinearAlgebra/src/eigen.jl:235 [inlined]
   [4] #eigvecs#98
     @ ~/.julia/juliaup/julia-1.9.0+0.x64.apple.darwin14/share/julia/stdlib/v1.9/LinearAlgebra/src/eigen.jl:270 [inlined]
   [5] eigvecs
     @ ~/.julia/juliaup/julia-1.9.0+0.x64.apple.darwin14/share/julia/stdlib/v1.9/LinearAlgebra/src/eigen.jl:270 [inlined]
   [6] matrixfunction(fn::Function, m::Matrix{Double64})
     @ DoubleFloats ~/.julia/packages/DoubleFloats/3G2HG/src/math/linearalgebra/matrixfunction.jl:3
   [7] log(m::Matrix{Double64})
     @ DoubleFloats ~/.julia/packages/DoubleFloats/3G2HG/src/math/linearalgebra/matrixfunction.jl:33
   [8] macro expansion
     @ ~/.julia/juliaup/julia-1.9.0+0.x64.apple.darwin14/share/julia/stdlib/v1.9/Test/src/Test.jl:478 [inlined]
   [9] sdp_trace_logm_real(handle_problem!::var"#22#25"{String, Symbol, DataType}, #unused#::Val{true}, atol::Float64, rtol::Float64, #unused#::Type{Double64})
     @ Convex.ProblemDepot ~/.julia/packages/Convex/tSTAW/src/problem_depot/problems/sdp.jl:2074

which looks like breakage in GenericLinearAlgebra to me.

Also:

[ Info: Solving problem sdp_trace_mpower_cplx_5_4 with variant sdpa_dd and type BigFloat
sdp_trace_mpower_cplx_5_4: Error During Test at /Users/eph/.julia/packages/Convex/tSTAW/src/problem_depot/problems/sdp.jl:1955
  Test threw exception
  Expression: ≈(p.optval, evaluate(trace_mpower(B, t, C)), atol = atol, rtol = rtol)
  MethodError: no method matching (Schur{Complex})(::GenericLinearAlgebra.Schur{Complex{BigFloat}, Matrix{Complex{BigFloat}}})
  
  Closest candidates are:
    (Schur{CT})(::Schur{<:Real}) where CT<:Complex
     @ LinearAlgebra ~/.julia/juliaup/julia-1.9.0+0.x64.apple.darwin14/share/julia/stdlib/v1.9/LinearAlgebra/src/schur.jl:223
    (Schur{Complex})(::Schur{<:Complex})
     @ LinearAlgebra ~/.julia/juliaup/julia-1.9.0+0.x64.apple.darwin14/share/julia/stdlib/v1.9/LinearAlgebra/src/schur.jl:256
    (Schur{T})(::Schur{T}) where T
     @ LinearAlgebra ~/.julia/juliaup/julia-1.9.0+0.x64.apple.darwin14/share/julia/stdlib/v1.9/LinearAlgebra/src/schur.jl:257
    ...
  
  Stacktrace:
   [1] schurpow(A::Matrix{Complex{BigFloat}}, p::Rational{Int64})
     @ LinearAlgebra ~/.julia/juliaup/julia-1.9.0+0.x64.apple.darwin14/share/julia/stdlib/v1.9/LinearAlgebra/src/dense.jl:502
   [2] ^(A::Matrix{Complex{BigFloat}}, p::Rational{Int64})
     @ LinearAlgebra ~/.julia/juliaup/julia-1.9.0+0.x64.apple.darwin14/share/julia/stdlib/v1.9/LinearAlgebra/src/dense.jl:547
   [3] trace_mpower(A::Matrix{Complex{BigFloat}}, t::Rational{Int64}, C::Matrix{ComplexF64})
     @ Convex ~/.julia/packages/Convex/tSTAW/src/atoms/sdp_cone/trace_mpower.jl:78
   [4] evaluate(atom::Convex.TraceMpower)
     @ Convex ~/.julia/packages/Convex/tSTAW/src/atoms/sdp_cone/trace_mpower.jl:66
   [5] macro expansion
     @ ~/.julia/juliaup/julia-1.9.0+0.x64.apple.darwin14/share/julia/stdlib/v1.9/Test/src/Test.jl:478 [inlined]
   [6] sdp_trace_mpower_cplx_5_4(handle_problem!::var"#22#25"{String, Symbol, DataType}, #unused#::Val{true}, atol::Float64, rtol::Float64, #unused#::Type{BigFloat})
     @ Convex.ProblemDepot ~/.julia/packages/Convex/tSTAW/src/problem_depot/problems/sdp.jl:1955

and

[ Info: Solving problem sdp_trace_logm_cplx with variant sdpa_dd and type BigFloat
sdp_trace_logm_cplx: Error During Test at /Users/eph/.julia/packages/Convex/tSTAW/src/problem_depot/problems/sdp.jl:2102
  Test threw exception
  Expression: ≈(p.optval, tr(C * log(evaluate(X))), atol = atol, rtol = rtol)
  type Schur has no field vectors
  Stacktrace:
   [1] getproperty(F::GenericLinearAlgebra.Schur{Complex{BigFloat}, Matrix{Complex{BigFloat}}}, s::Symbol)
     @ GenericLinearAlgebra ~/.julia/packages/GenericLinearAlgebra/skQkq/src/eigenGeneral.jl:76
   [2] log(A::Matrix{Complex{BigFloat}})
     @ LinearAlgebra ~/.julia/juliaup/julia-1.9.0+0.x64.apple.darwin14/share/julia/stdlib/v1.9/LinearAlgebra/src/dense.jl:822
   [3] macro expansion
     @ ~/.julia/juliaup/julia-1.9.0+0.x64.apple.darwin14/share/julia/stdlib/v1.9/Test/src/Test.jl:478 [inlined]
   [4] sdp_trace_logm_cplx(handle_problem!::var"#22#25"{String, Symbol, DataType}, #unused#::Val{true}, atol::Float64, rtol::Float64, #unused#::Type{BigFloat})
     @ Convex.ProblemDepot ~/.julia/packages/Convex/tSTAW/src/problem_depot/problems/sdp.jl:2102

seem like upstream issues too (not sure if from Base, GenericLinearAlgebra, or Convex)

codecov-commenter · 2023-06-21T08:34:39Z

Codecov Report

Patch coverage: 86.90% and project coverage change: -14.41 ⚠️

Comparison is base (655008f) 88.56% compared to head (5e655f8) 74.16%.

Additional details and impacted files

@@             Coverage Diff             @@
##           master      #66       +/-   ##
===========================================
- Coverage   88.56%   74.16%   -14.41%     
===========================================
  Files           6        7        +1     
  Lines         551      569       +18     
===========================================
- Hits          488      422       -66     
- Misses         63      147       +84

Impacted Files	Coverage Δ
src/SDPAFamily.jl	`75.00% <ø> (ø)`
src/params.jl	`91.17% <ø> (ø)`
src/MOI_wrapper.jl	`79.50% <86.56%> (-3.12%)`	⬇️
src/objective_function_filter.jl	`86.66% <86.66%> (ø)`
src/file_io.jl	`66.91% <100.00%> (-33.09%)`	⬇️

... and 2 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

blegat · 2023-06-27T07:49:10Z

@ericphanson Tests are passing now, the documentation failure seems unrelated. Good to merge and release ? :)

ericphanson

We should also:

breaking version bump
warning in readme that presolve is not enabled, linking to issue
update docs where presolve is discussed

I can do that as a separate PR though.

test/MOI_wrapper.jl

blegat · 2023-06-27T12:19:38Z

Nice, tests are passing with BigFloat as well

blegat · 2023-06-27T12:21:05Z

I added the steps in #67

ericphanson · 2023-06-27T16:03:52Z

I think you were already running them, they were just under the category Float64 in the loop there

blegat added 2 commits June 2, 2023 10:14

Update to MOI v1

156760c

Fixes

2fcc69f

ericphanson reviewed Jun 2, 2023

View reviewed changes

This was referenced Jun 20, 2023

trace_mpower with BigFloat jump-dev/Convex.jl#502

Closed

Add excludes MOI and Convex tests with presolve jump-dev/Convex.jl#503

Closed

blegat added 3 commits June 20, 2023 07:57

Fix tests

e4d75fc

Exclude tests

d94c2bf

Remove failing tests

3e93fb7

blegat mentioned this pull request Jun 21, 2023

Add excluded MOI and Convex tests with presolve #67

Open

5 tasks

Exclude relative entropy 4

e59d026

ericphanson approved these changes Jun 27, 2023

View reviewed changes

test/MOI_wrapper.jl Outdated Show resolved Hide resolved

Update test/MOI_wrapper.jl

5e655f8

ericphanson merged commit 610e665 into ericphanson:master Jun 27, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Update to MOI v1 #66

Update to MOI v1 #66

blegat commented Jun 2, 2023

ericphanson Jun 2, 2023

blegat Jun 2, 2023

ericphanson Jun 2, 2023

blegat Jun 2, 2023

ericphanson Jun 2, 2023

blegat Jun 2, 2023 •

edited

Loading

blegat Jun 2, 2023

ericphanson Jun 6, 2023

blegat commented Jun 6, 2023

ericphanson commented Jun 6, 2023

ericphanson commented Jun 6, 2023 •

edited

Loading

codecov-commenter commented Jun 21, 2023 •

edited

Loading

blegat commented Jun 27, 2023

ericphanson left a comment

blegat commented Jun 27, 2023

blegat commented Jun 27, 2023

ericphanson commented Jun 27, 2023

		optimizer.dualobj = parse(T, objValPrimalstring)
		optimizer.primalobj = parse(T, objValDualstring)

Update to MOI v1 #66

Update to MOI v1 #66

Conversation

blegat commented Jun 2, 2023

ericphanson Jun 2, 2023

Choose a reason for hiding this comment

blegat Jun 2, 2023

Choose a reason for hiding this comment

ericphanson Jun 2, 2023

Choose a reason for hiding this comment

blegat Jun 2, 2023

Choose a reason for hiding this comment

ericphanson Jun 2, 2023

Choose a reason for hiding this comment

blegat Jun 2, 2023 • edited Loading

Choose a reason for hiding this comment

blegat Jun 2, 2023

Choose a reason for hiding this comment

ericphanson Jun 6, 2023

Choose a reason for hiding this comment

blegat commented Jun 6, 2023

ericphanson commented Jun 6, 2023

ericphanson commented Jun 6, 2023 • edited Loading

codecov-commenter commented Jun 21, 2023 • edited Loading

Codecov Report

blegat commented Jun 27, 2023

ericphanson left a comment

Choose a reason for hiding this comment

blegat commented Jun 27, 2023

blegat commented Jun 27, 2023

ericphanson commented Jun 27, 2023

blegat Jun 2, 2023 •

edited

Loading

ericphanson commented Jun 6, 2023 •

edited

Loading

codecov-commenter commented Jun 21, 2023 •

edited

Loading