Adapt Uno interface to unopy 0.4.0

[cvanaret]

Description

We implemented the following changes between unopy 0.2.7 and unopy 0.4.0:

• compatibility with numpy arrays
• small breaking changes: the Lagrangian sign convention is passed to the model via the dedicated function set_lagrangian_sign_convention instead of set_lagrangian_hessian

See unopy example

Question: can you pin down the dependency to a particular version of unopy, or a minimum required version?

Issue link (if applicable):

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.

[cvanaret]

Ah I see you don't have Uno tests yet. Should I try to merge into #165 instead?

[Transurgeon]

Ah I see you don't have Uno tests yet. Should I try to merge into #165 instead?

no its fine. I will cleanup the #165 and merge it in.

[github-actions]

Benchmarks that have stayed the same:

   before           after         ratio
 [e6df2dda]       [909b6fd0]
     14.6±0ms         15.9±0ms     1.09  simple_LP_benchmarks.SimpleFullyParametrizedLPBenchmark.time_compile_problem
      20.2±0s          21.1±0s     1.04  sdp_segfault_1132_benchmark.SDPSegfault1132Benchmark.time_compile_problem
     42.5±0ms         43.3±0ms     1.02  matrix_stuffing.SmallMatrixStuffing.time_compile_problem
      710±0ms          720±0ms     1.01  matrix_stuffing.ConeMatrixStuffingBench.time_compile_problem
      1.00±0s          1.01±0s     1.01  finance.FactorCovarianceModel.time_compile_problem
      9.98±0s          10.1±0s     1.01  simple_LP_benchmarks.SimpleLPBenchmark.time_compile_problem
      1.76±0s          1.78±0s     1.01  simple_QP_benchmarks.UnconstrainedQP.time_compile_problem
      518±0ms          523±0ms     1.01  semidefinite_programming.SemidefiniteProgramming.time_compile_problem
      12.1±0s          12.2±0s     1.01  finance.CVaRBenchmark.time_compile_problem
      3.02±0s          3.03±0s     1.00  quantum_hilbert_matrix.QuantumHilbertMatrix.time_compile_problem
      1.57±0s          1.57±0s     1.00  tv_inpainting.TvInpainting.time_compile_problem
      5.09±0s          5.09±0s     1.00  optimal_advertising.OptimalAdvertising.time_compile_problem
      3.88±0s          3.88±0s     1.00  huber_regression.HuberRegression.time_compile_problem
      4.41±0s          4.41±0s     1.00  svm_l1_regularization.SVMWithL1Regularization.time_compile_problem
      283±0ms          283±0ms     1.00  matrix_stuffing.ParamSmallMatrixStuffing.time_compile_problem
      238±0ms          237±0ms     1.00  simple_QP_benchmarks.SimpleQPBenchmark.time_compile_problem
      1.04±0s          1.04±0s     0.99  gini_portfolio.Cajas.time_compile_problem
      230±0ms          228±0ms     0.99  gini_portfolio.Murray.time_compile_problem
      1.42±0s          1.41±0s     0.99  matrix_stuffing.ParamConeMatrixStuffing.time_compile_problem
      881±0ms          869±0ms     0.99  simple_LP_benchmarks.SimpleScalarParametrizedLPBenchmark.time_compile_problem
     15.3±0ms         15.1±0ms     0.99  simple_QP_benchmarks.ParametrizedQPBenchmark.time_compile_problem
      276±0ms          271±0ms     0.98  slow_pruning_1668_benchmark.SlowPruningBenchmark.time_compile_problem
      743±0ms          727±0ms     0.98  simple_QP_benchmarks.LeastSquares.time_compile_problem
      137±0ms          134±0ms     0.98  high_dim_convex_plasticity.ConvexPlasticity.time_compile_problem
      347±0ms          318±0ms     0.92  gini_portfolio.Yitzhaki.time_compile_problem

[PTNobel]

This is awesome! Super happy to see the changes. On the CVXPY side we set minimum versions all the time so that should be no trouble.

[Transurgeon]

This is awesome! Super happy to see the changes. On the CVXPY side we set minimum versions all the time so that should be no trouble.

This is awesome! Super happy to see the changes. On the CVXPY side we set minimum versions all the time so that should be no trouble.

actually it seems like the #165 branch got super outdated.. so maybe you can add the changes here?

      - name: Install Uno solver
        run: uv pip install unopy
      - name: Verify Uno import
        run: uv run python -c "import unopy; print('unopy imported successfully')"

I think you would need to add these four lines in the test_nlp_solvers.yml file. This will install the latest one.

[cvanaret]

OK, unopy installs and runs successfully. It does all kinds of crazy on a couple of instances, which I will investigate :)

[Transurgeon]

I'll approve this for now, thanks a lot @cvanaret !
really exciting to make progress on this collaboration.
Will wait for the CI to pass after #181 so that the nlp tests can be actually run there.