Quadratic constraints by amontoison · Pull Request #102 · JuliaSmoothOptimizers/NLPModelsJuMP.jl

amontoison · 2022-03-10T15:38:31Z

github-actions · 2022-03-10T15:44:12Z

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

github-actions · 2022-05-06T16:44:19Z

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

codecov · 2022-05-06T19:11:05Z

Codecov Report

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

❗ 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%)`	⬇️
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 · 2022-05-06T19:13:00Z

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

tmigot · 2022-05-06T19:43:35Z

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

Improve implementation of jac_nln_coord! https://github.com/amontoison/NLPModelsJuMP.jl/blob/3db790058ce86b88582e7ab740d40d1bd93ee369/src/moi_nlp_model.jl#L160
Improve implementation of hprod! https://github.com/amontoison/NLPModelsJuMP.jl/blob/3db790058ce86b88582e7ab740d40d1bd93ee369/src/moi_nlp_model.jl#L407
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 · 2022-05-06T19:48:48Z

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

tmigot · 2022-05-06T21:11:30Z

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 · 2022-05-10T22:14:33Z

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

github-actions · 2022-05-11T03:14:45Z

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

tmigot · 2022-05-11T03:42:29Z

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

github-actions · 2022-05-11T13:53:12Z

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

tmigot · 2022-05-12T16:16:58Z

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.

amontoison

Major modifications of NLPModels.hess_coord! are required.

amontoison · 2022-05-12T22:13:17Z

+    for j=1:length(qcon.vec)
+      vals[k + j] = qcon.b[qcon.vec[j]]
+    end
+    nnzj = length(qcon.hessian.vals)


nnzj should be the number of nnz of Qᵢx + bᵢ (length(set)) and not Qᵢ.

amontoison · 2022-05-12T22:28:59Z

+    for i = 1:(nlp.quadcon.nquad)
+      qcon = nlp.quadcon[i]
+      nnzh = length(qcon.hessian.vals)
+      vals[(k + 1):(k + nnzh)] .= qcon.hessian.vals .* y[nlp.meta.nlin + i]


We want to avoid that with the hessian_quad function!
If we have 10 quadratic constraints Q_i with same structure, you will store 10 times the required number of nnz for nothing.

amontoison · 2022-05-12T22:30:28Z

    vals[(nlp.obj.nnzh + 1):(nlp.meta.nnzh)] .= 0.0
  end
  if nlp.obj.type == "NONLINEAR"
+    vals .= 0.0


MOI.eval_hessian_lagrangian doesn't overwrite vals?

https://github.com/jump-dev/MathOptInterface.jl/blob/master/src/Nonlinear/ReverseAD/mathoptinterface_api.jl#L261-L273

If vals .= 0.0 is required, we should also add it in the other NLPModels.hess_coord! function.

github-actions · 2022-05-12T23:37:00Z

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

github-actions · 2022-05-12T23:40:52Z

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

github-actions · 2022-05-12T23:46:43Z

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

github-actions · 2022-05-12T23:47:59Z

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

github-actions · 2022-05-13T00:03:54Z

Package name	latest	stable
NLSProblems.jl
OptimizationProblems.jl

tmigot · 2024-06-21T13:47:14Z

Superseded by #181

tmigot force-pushed the quadratic-constraints branch from 536e263 to 4d4f2bd Compare May 6, 2022 16:40

amontoison changed the title ~~[WIP] Quadratic constraints~~ Quadratic constraints May 12, 2022

tmigot approved these changes May 12, 2022

View reviewed changes

amontoison commented May 12, 2022

View reviewed changes

Comment thread src/moi_nlp_model.jl Outdated

amontoison commented May 12, 2022

View reviewed changes

Comment thread src/moi_nlp_model.jl Outdated

tmigot self-requested a review May 12, 2022 22:59

amontoison and others added 8 commits May 12, 2022 19:46

[WIP] Quadratic constraints

7273d2f

add straightforward test

6990308

debug utils functions

2b1574d

add API for quadratic constraints

3bec27b

simplify nquad

0f08e9a

add hs100

af04c2e

store vec

2aaf353

implement hprod and jac_coord properly

90ea72a

tmigot and others added 7 commits May 12, 2022 19:46

add test problem and fix

836a7b9

only evaluate the NLPEval when there is nonlinear cons

494e3a0

[MathOptNLPModel] Fix hprod!

01863f7

[MathOptNLPModel] Fix cases with nlp.eval == nothing

86ccf44

Update test/nlp_problems/hs100.jl

ef95770

Remove hessian_structure function

c64ac74

Optimize hessian_structure function for quadratic constraints

4b24cab

[MathOptNLPModel] hotfix

e141a9c

amontoison force-pushed the quadratic-constraints branch from 8cbdbb6 to e141a9c Compare May 13, 2022 00:00

amontoison mentioned this pull request Jun 4, 2023

Unsupported features should error, not warn #150

Closed

tmigot marked this pull request as draft October 17, 2023 12:38

tmigot mentioned this pull request Dec 16, 2023

Quadratic functions in the constraints are not picked up by MathOptNLPModel #173

Closed

frapac mentioned this pull request Feb 29, 2024

Update JuMP syntax to use the new nonlinear interface JuliaSmoothOptimizers/OptimizationProblems.jl#321

Closed

tmigot closed this Jun 21, 2024

Conversation

amontoison commented Mar 10, 2022

Uh oh!

github-actions bot commented Mar 10, 2022

Uh oh!

github-actions bot commented May 6, 2022

Uh oh!

codecov bot commented May 6, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

github-actions bot commented May 6, 2022

Uh oh!

tmigot commented May 6, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented May 6, 2022

Uh oh!

tmigot commented May 6, 2022

Uh oh!

github-actions bot commented May 10, 2022

Uh oh!

github-actions bot commented May 11, 2022

Uh oh!

tmigot commented May 11, 2022

Uh oh!

github-actions bot commented May 11, 2022

Uh oh!

tmigot commented May 12, 2022

Uh oh!

amontoison left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

amontoison May 12, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

amontoison May 12, 2022

Choose a reason for hiding this comment

Uh oh!

amontoison May 12, 2022

Choose a reason for hiding this comment

Uh oh!

amontoison May 13, 2022

Choose a reason for hiding this comment

Uh oh!

amontoison May 13, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

github-actions bot commented May 12, 2022

Uh oh!

github-actions bot commented May 12, 2022

Uh oh!

github-actions bot commented May 12, 2022

Uh oh!

github-actions bot commented May 12, 2022

Uh oh!

github-actions bot commented May 13, 2022

Uh oh!

tmigot commented Jun 21, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

codecov bot commented May 6, 2022 •

edited

Loading

tmigot commented May 6, 2022 •

edited

Loading

amontoison May 12, 2022 •

edited

Loading

amontoison May 13, 2022 •

edited

Loading