Quadratic constraints

[amontoison]

trafficstars

@tmigot

[github-actions[bot]]

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

[github-actions[bot]]

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

[codecov[bot]]

Codecov Report

Merging #102 (3d7001c) into main (bb093af) will increase coverage by 0.07%. The diff coverage is 96.21%.

:exclamation: Current head 3d7001c differs from pull request most recent head e141a9c. Consider uploading reports for the commit e141a9c to get more accurate results

@@            Coverage Diff             @@
##             main     #102      +/-   ##
==========================================
+ Coverage   95.00%   95.07%   +0.07%     
==========================================
  Files           3        3              
  Lines         580      690     +110     
==========================================
+ Hits          551      656     +105     
- Misses         29       34       +5     

Impacted Files	Coverage Δ
src/utils.jl	97.23% <93.24%> (-1.67%)	:arrow_down:
src/moi_nlp_model.jl	100.00% <100.00%> (ø)
src/moi_nls_model.jl	88.36% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update bb093af...e141a9c. Read the comment docs.

[github-actions[bot]]

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

[tmigot]

Hi @amontoison , I made some progress with this one. A list of things that we can discuss:

• [x] Improve implementation of jac_nln_coord! https://github.com/amontoison/NLPModelsJuMP.jl/blob/3db790058ce86b88582e7ab740d40d1bd93ee369/src/moi_nlp_model.jl#L160
• [x] Improve implementation of hprod! https://github.com/amontoison/NLPModelsJuMP.jl/blob/3db790058ce86b88582e7ab740d40d1bd93ee369/src/moi_nlp_model.jl#L407
• [x] Add more tests (note that I used hs61 which is implemented as adnlpmodels in https://github.com/JuliaSmoothOptimizers/OptimizationProblems.jl/blob/main/src/ADNLPProblems/hs61.jl )
• [ ] Copy-paste to the NLS variant.

Right now my test was very basic, just testing that all the API returns something.

jump = hs61()
nlp = MathOptNLPModel(jump)
x1 = rand(nlp.meta.nvar)

obj(nlp, x1)
cons(nlp, x1)
jac_structure(nlp)
jac_coord(nlp, x1)
hess(nlp, x1)
hess(nlp, x1, rand(nlp.meta.ncon))
hprod(nlp, x1, x1)

[github-actions[bot]]

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

[tmigot]

We can also use the following script to test problems with quadratic constraints from OptimizationProblems.jl

using ADNLPModels, OptimizationProblems, NLPModels

"""
    test_quadratic_constraints(sample_size = 100)

Return the list of problems with quadratic constraints
"""
function test_quadratic_constraints(sample_size = 10)
  meta = OptimizationProblems.meta[!, :]
  con_pb = meta[meta.ncon .> 0, :name]
  sample_size = max(sample_size, 2)

  list = []
  for pb in con_pb
    nlp = OptimizationProblems.ADNLPProblems.eval(Symbol(pb))()
    std = similar(nlp.meta.x0)
    blvar = similar(nlp.meta.lvar)
    buvar = similar(nlp.meta.uvar)
    for j=1:nlp.meta.nvar
      blvar[j] = nlp.meta.lvar[j] == -Inf ? -10. : nlp.meta.lvar[j]
      buvar[j] = nlp.meta.uvar[j] == Inf ? 10. : nlp.meta.uvar[j]
      std[j] = max(abs(blvar[j]), abs(buvar[j]))
    end
    Iref = collect(1:nlp.meta.ncon)
    for k in Iref
      if k in nlp.meta.lin # substract linear constraints
        Iref[k] = -1
        continue
      end
      y0 = zeros(nlp.meta.ncon)
      y0[k] = 1.0
      ref = hess(nlp, nlp.meta.x0, y0, obj_weight = 0.0)
      for i=1:sample_size
        x = min.(max.((2 * rand(nlp.meta.nvar) .- 1) .* std, blvar), buvar)
        Hx = hess(nlp, x, y0, obj_weight = 0.0)
        if Hx != ref
          Iref[k] = -1
          break
        end
      end
    end
    if findall(x -> x > 0, Iref) != []
      push!(list, (nlp.meta.name, Iref[findall(x -> x > 0, Iref)]))
    end
  end
  return list
end

test_quadratic_constraints()

[github-actions[bot]]

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

[github-actions[bot]]

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

[tmigot]

@amontoison I am checking with the other problems in OptimizationProblems.jl, but it looks good to me.

[github-actions[bot]]

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

[tmigot]

I am done testing the OptimizationProblems.jl ; I added 45 problems with quadratic constraints https://github.com/JuliaSmoothOptimizers/OptimizationProblems.jl/tree/add-quadratic and no error found! So, it's good for me. Up to you.

[github-actions[bot]]

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

[github-actions[bot]]

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

[github-actions[bot]]

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

[github-actions[bot]]

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

[github-actions[bot]]

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

[tmigot]

Superseded by #181