Add support for KNITRO solver by eminyouskn · Pull Request #2873 · cvxpy/cvxpy

eminyouskn · 2025-07-24T20:29:00Z

Description

This PR adds support for the KNITRO solver.
To run the tests in CI, the KNITRO_CI environment variable needs to be set. Artelys can provide a license for this purpose, but I’m not sure whom to contact to arrange that.

Type of change

New feature (backwards compatible)
New feature (breaking API changes)
Bug fix
Other (Documentation, CI, ...)

Contribution checklist

Add our license to new files.
Check that your code adheres to our coding style.
Write unittests.
Run the unittests and check that they’re passing.
Run the benchmarks to make sure your change doesn’t introduce a regression.

…ms (#1) * Add KNITRO solver to the list of available solvers * Add KNITRO QP and Conic solver interfaces * Add KNITRO citation to the citation dictionary * Add KNITRO QP and Conic solver interface implementation * Add KNITRO to doc * Add KNITRO to ci for optional solvers

CLAassistant · 2025-07-24T20:29:06Z

All committers have signed the CLA.

PTNobel · 2025-07-24T20:45:23Z

Hello, you can reach out to me cvxpydevs@gmail.com with info on us getting a license for CI.

eminyouskn · 2025-07-24T20:54:31Z

@PTNobel thanks.

SteveDiamond · 2025-07-29T18:00:44Z

I added the license under the "KNITRO_LICENSE" secret. Unfortunately secrets can't be accessed in MRs.

eminyouskn · 2025-07-29T18:26:02Z

Thanks @SteveDiamond. Let me then rename the secret to KNITRO_LICENSE instead of KNITRO_CI

eminyouskn · 2025-07-30T18:16:08Z

I noticed that KNITRO wasn't installed in the test_optional_solvers jobs. I assume this is because the secret KNITRO_LICENSE is not accessible from pull requests. Do you need me to do anything about that?

P.S.: On macOS, KNITRO is only supported on the ARM architecture, which is not the default configuration tested in your CI. Would it be possible to add this OS configuration to your CI setup?

Transurgeon · 2025-07-30T20:06:03Z

I noticed that KNITRO wasn't installed in the test_optional_solvers jobs. I assume this is because the secret KNITRO_LICENSE is not accessible from pull requests. Do you need me to do anything about that?

I think the secret licenses are only accessible when the pull request gets merged to master, but I am not entirely sure. I suppose it should be similar to how we deal with the MOSEK license.

P.S.: On macOS, KNITRO is only supported on the ARM architecture, which is not the default configuration tested in your CI. Would it be possible to add this OS configuration to your CI setup?

Yes! I definitely think we should add that. Maybe having an extra entry for macOS-latest should do the job.

github-actions · 2025-07-30T20:06:17Z

Benchmarks that have stayed the same:

   before           after         ratio
 [07d05b53]       [19ef24f8]
      941±0ms          1.02±0s     1.08  gini_portfolio.Cajas.time_compile_problem
      1.38±0s          1.47±0s     1.07  matrix_stuffing.ParamConeMatrixStuffing.time_compile_problem
      284±0ms          295±0ms     1.04  matrix_stuffing.ParamSmallMatrixStuffing.time_compile_problem
      570±0ms          585±0ms     1.03  simple_QP_benchmarks.ParametrizedQPBenchmark.time_compile_problem
      2.35±0s          2.39±0s     1.02  simple_LP_benchmarks.SimpleFullyParametrizedLPBenchmark.time_compile_problem
      4.59±0s          4.66±0s     1.02  huber_regression.HuberRegression.time_compile_problem
      22.3±0s          22.6±0s     1.01  sdp_segfault_1132_benchmark.SDPSegfault1132Benchmark.time_compile_problem
      939±0ms          951±0ms     1.01  simple_LP_benchmarks.SimpleScalarParametrizedLPBenchmark.time_compile_problem
      236±0ms          239±0ms     1.01  gini_portfolio.Murray.time_compile_problem
      13.0±0s          13.1±0s     1.01  finance.CVaRBenchmark.time_compile_problem
      5.33±0s          5.38±0s     1.01  semidefinite_programming.SemidefiniteProgramming.time_compile_problem
      5.00±0s          5.03±0s     1.01  optimal_advertising.OptimalAdvertising.time_compile_problem
      344±0ms          346±0ms     1.01  gini_portfolio.Yitzhaki.time_compile_problem
      1.66±0s          1.67±0s     1.00  tv_inpainting.TvInpainting.time_compile_problem
      843±0ms          844±0ms     1.00  simple_QP_benchmarks.LeastSquares.time_compile_problem
      241±0ms          241±0ms     1.00  high_dim_convex_plasticity.ConvexPlasticity.time_compile_problem
      11.4±0s          11.4±0s     1.00  simple_LP_benchmarks.SimpleLPBenchmark.time_compile_problem
      1.86±0s          1.86±0s     1.00  simple_QP_benchmarks.UnconstrainedQP.time_compile_problem
      297±0ms          297±0ms     1.00  slow_pruning_1668_benchmark.SlowPruningBenchmark.time_compile_problem
      253±0ms          253±0ms     1.00  simple_QP_benchmarks.SimpleQPBenchmark.time_compile_problem
      5.22±0s          5.14±0s     0.99  svm_l1_regularization.SVMWithL1Regularization.time_compile_problem
      1.10±0s          1.08±0s     0.98  finance.FactorCovarianceModel.time_compile_problem
      731±0ms          715±0ms     0.98  matrix_stuffing.ConeMatrixStuffingBench.time_compile_problem
      2.94±0s          2.86±0s     0.97  quantum_hilbert_matrix.QuantumHilbertMatrix.time_compile_problem
     45.2±0ms         43.9±0ms     0.97  matrix_stuffing.SmallMatrixStuffing.time_compile_problem

Transurgeon

thanks for making this PR @eminyouskn , I left a few comments/questions above. Other than that the interface is looking pretty good!

I had a related question for you: we are currently working on adding a new IPOPT interface for NLPs and I was wondering if the KNITRO NLP interface was similar to IPOPT's? We would definitely like to support KNITRO at some point down the line!

.github/workflows/test_optional_solvers.yml

cvxpy/reductions/solvers/conic_solvers/knitro_conif.py

eminyouskn · 2025-08-01T00:51:12Z

Thanks @Transurgeon for your comments. I've addressed most of them. Let me know if there's anything else that needs attention.

Regarding the NLP interface, I'm not familiar with the IPOPT implementation. However, if you'd like to add NLP support for KNITRO, we'd be happy to assist. KNITRO, for example, uses callbacks to handle general nonlinear functions.

PTNobel · 2025-07-24T21:12:20Z

cvxpy/reductions/solvers/qp_solvers/knitro_qpif.py

+                solution = Solution(status, obj, primal_vars, dual_vars, attr)
+
+        # Free the Knitro context.
+        kn.KN_free(kc)


Since we're cacheing the context should we be freeing it here?

I agree. However, we're not using the cache yet. I plan to add that in the future, but for now, I think it's safe to free the context here. By the way the same thing apply for the conic interface.

Transurgeon · 2025-08-01T18:02:53Z

@eminyouskn I just merged the master cvxpy branch to your PR, which includes the new macos-14 ARM runners. Once this PR gets merged we should be able to see that it successfully tested the KNITRO solver (with the license).

cvxpy/reductions/solvers/qp_solvers/knitro_qpif.py

.github/workflows/test_optional_solvers.yml

…r classes

eminyouskn · 2025-08-01T18:47:37Z

@eminyouskn I just merged the master cvxpy branch to your PR, which includes the new macos-14 ARM runners. Once this PR gets merged we should be able to see that it successfully tested the KNITRO solver (with the license).

Thanks @Transurgeon for the PR. I tested it on my fork, and the tests pass for KNITRO. However, there's still an error related to SDPA.

Transurgeon · 2025-08-01T20:54:04Z

@eminyouskn I just merged the master cvxpy branch to your PR, which includes the new macos-14 ARM runners. Once this PR gets merged we should be able to see that it successfully tested the KNITRO solver (with the license).

Thanks @Transurgeon for the PR. I tested it on my fork, and the tests pass for KNITRO. However, there's still an error related to SDPA.

yeah that's fine, the error is definitely unrelated.

Transurgeon

Okay, I think this PR is ready to merge. @PTNobel could you have a sanity check please?

PTNobel

Coursery read looks fine

eminyouskn added 2 commits July 29, 2025 18:30

Use KNITRO_LICENSE instead of KNITRO_CI.

062f151

Fix KNITRO installation condition for macOS architecture check

3b82032

Transurgeon requested changes Jul 31, 2025

View reviewed changes

.github/workflows/test_optional_solvers.yml Show resolved Hide resolved

cvxpy/reductions/solvers/conic_solvers/knitro_conif.py Outdated Show resolved Hide resolved

cvxpy/reductions/solvers/conic_solvers/knitro_conif.py Outdated Show resolved Hide resolved

eminyouskn added 4 commits August 1, 2025 00:40

Update copyright in conic_solver.py and qp_solver.py

bedef37

Update copyright information

0e1ddf3

Reverse back the copyright

032c594

Remove unused import statement for knitro in KNITRO solver classes

2bbb42a

eminyouskn added 2 commits August 1, 2025 01:04

Add import statement for Knitro solver in KNITRO classes

1f000fe

Remove unused import statement for Knitro in KNITRO solver classes

fc0af1b

PTNobel reviewed Aug 1, 2025

View reviewed changes

Merge branch 'master' into master

0020194

Transurgeon added solver interfaces PR no backport needed labels Aug 1, 2025

Transurgeon reviewed Aug 1, 2025

View reviewed changes

cvxpy/reductions/solvers/qp_solvers/knitro_qpif.py Show resolved Hide resolved

.github/workflows/test_optional_solvers.yml Outdated Show resolved Hide resolved

eminyouskn added 2 commits August 1, 2025 18:43

Add documentation reference for Knitro status mapping in KNITRO solve…

5f004bf

…r classes

Fix formatting of KNITRO license echo command in CI workflow

9f78d54

Transurgeon approved these changes Aug 1, 2025

View reviewed changes

PTNobel approved these changes Aug 1, 2025

View reviewed changes

PTNobel merged commit 9009cb9 into cvxpy:master Aug 1, 2025
51 of 53 checks passed

Conversation

eminyouskn commented Jul 24, 2025

Description

Type of change

Contribution checklist

Uh oh!

CLAassistant commented Jul 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

PTNobel commented Jul 24, 2025

Uh oh!

eminyouskn commented Jul 24, 2025

Uh oh!

SteveDiamond commented Jul 29, 2025

Uh oh!

eminyouskn commented Jul 29, 2025

Uh oh!

eminyouskn commented Jul 30, 2025

Uh oh!

Transurgeon commented Jul 30, 2025

Uh oh!

github-actions bot commented Jul 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Transurgeon left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

eminyouskn commented Aug 1, 2025

Uh oh!

PTNobel Jul 24, 2025

Choose a reason for hiding this comment

Uh oh!

eminyouskn Aug 1, 2025

Choose a reason for hiding this comment

Uh oh!

Transurgeon commented Aug 1, 2025

Uh oh!

Uh oh!

Uh oh!

eminyouskn commented Aug 1, 2025

Uh oh!

Transurgeon commented Aug 1, 2025

Uh oh!

Transurgeon left a comment

Choose a reason for hiding this comment

Uh oh!

PTNobel left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

CLAassistant commented Jul 24, 2025 •

edited

Loading

github-actions bot commented Jul 30, 2025 •

edited

Loading