Simplify log/exp verification to check is_affine() and add tests #134

Transurgeon · 2026-01-20T20:08:53Z

Simplify _verify_hess_vec_args and _verify_jacobian_args in log.py and exp.py to check self.args[0].is_affine() instead of requiring isinstance(Variable)
Remove _hess_vec methods from both atoms (NLP solver handles derivatives via C diff engine)
Fix _extract_affine_as_linear_op in converters.py to create column vector (n_vars, 1) instead of row vector for compatibility with C new_linear function
Add comprehensive test suite for affine argument handling including gradient verification, Hessian structure tests, edge cases, and constraint tests

Replace the pure-Python Oracles class with a new implementation that wraps the C-based C_problem class from dnlp_diff_engine. This provides automatic differentiation via the compiled C library instead of the Python-based jacobian/hessian computation. Key changes: - Replace Oracles class implementation in nlp_solver.py - Use lazy import to avoid circular dependency at module load time - Convert CSR sparse matrices from diff engine to COO format for solvers - Cache sparsity structures for jacobian and hessian Currently supported atoms: log, exp, sum, AddExpression, NegExpression, Promote. Other atoms (multiply, power, matmul, etc.) need to be added to the diff engine before all NLP tests pass. Co-Authored-By: Claude Opus 4.5 <[email protected]>

- Document all newly added Python bindings (power, trig, hyperbolic, matmul) - Update test status: 2/15 IPOPT tests now pass - Clarify blocking issues for remaining tests (index, norm, C bugs) Co-Authored-By: Claude Opus 4.5 <[email protected]>

…alue extraction - Remove redundant objective_forward/constraint_forward calls from gradient(), jacobian(), and hessian() since NLP solvers guarantee call ordering - Replace matrix densification + zip iteration with direct COO data return for jacobian and hessian value extraction Co-Authored-By: Claude Opus 4.5 <[email protected]>

- Updated implemented atoms list (index, reshape, sqrt, quad_over_lin, rel_entr) - Updated test results summary with current pass/fail status - Added known issues section (segfaults, bivariate matmul, rel_entr scalar) - Added missing atoms by priority Co-Authored-By: Claude Opus 4.5 <[email protected]>

- reshape atom now fully implemented with Python binding (commit 81c5a12) - All 3 circle packing tests now PASS - test_clnlbeam now PASSES (previously segfaulted) - Memory issues with many constraints fixed - Only test_localization still failing (needs broadcast_to) Co-Authored-By: Claude Opus 4.5 <[email protected]>

Full nlp_tests results (2025-01-14): - 259 passed, 20 failed, 64 skipped, 1 xfailed Remaining failures by missing atom: - broadcast_to: 5 tests (localization, broadcast, best_of) - Prod: 9 tests - MulExpression (bivariate matmul): 5 tests - rel_entr scalar variants: 1 test Co-Authored-By: Claude Opus 4.5 <[email protected]>

Examples: X @ Y (both Variables), cos(X) @ sin(Y) Currently supported: A @ f(x) and f(x) @ A where A is constant Co-Authored-By: Claude Opus 4.5 <[email protected]>

- Added Prod to implemented atoms list - Updated test_prod.py results: 13 passing, 1 failing (axis param) - Updated summary: 267 passed, 12 failed (down from 20) Remaining failures: broadcast_to (5), Prod axis (1), MulExpression (5), rel_entr scalar (1) Co-Authored-By: Claude Opus 4.5 <[email protected]>

- Create cvxpy/reductions/solvers/nlp_solvers/diff_engine/ package - Move C_problem wrapper and ATOM_CONVERTERS from dnlp-diff-engine - converters.py: Expression tree conversion from CVXPY atoms to C nodes - c_problem.py: C_problem class wrapping the C problem struct - Update nlp_solver.py to import from new location This separates CVXPY-specific glue code from the pure C autodiff library, eliminating circular dependencies and preparing dnlp-diff-engine for standalone PyPI distribution. Co-Authored-By: Claude Opus 4.5 <[email protected]>

Merge license header and module docstring into single docstring to avoid "module level import not at top of file" errors. Co-Authored-By: Claude Opus 4.5 <[email protected]>

…asting rule. This is now internally done in the diff engine

- Simplify _verify_hess_vec_args and _verify_jacobian_args in log.py and exp.py to check self.args[0].is_affine() instead of requiring isinstance(Variable) - Remove _hess_vec methods from both atoms (NLP solver handles derivatives via C diff engine) - Fix _extract_affine_as_linear_op in converters.py to create column vector (n_vars, 1) instead of row vector for compatibility with C new_linear function - Add comprehensive test suite for affine argument handling including gradient verification, Hessian structure tests, edge cases, and constraint tests Co-Authored-By: Claude Opus 4.5 <[email protected]>

github-actions · 2026-01-20T20:14:35Z

Benchmarks that have stayed the same:

   before           after         ratio
 [2ed0cd09]       [2b198459]
     41.5±0ms         43.8±0ms     1.06  matrix_stuffing.SmallMatrixStuffing.time_compile_problem
      321±0ms          336±0ms     1.05  gini_portfolio.Yitzhaki.time_compile_problem
      13.7±0s          14.3±0s     1.04  finance.CVaRBenchmark.time_compile_problem
      293±0ms          304±0ms     1.04  matrix_stuffing.ParamSmallMatrixStuffing.time_compile_problem
      942±0ms          971±0ms     1.03  gini_portfolio.Cajas.time_compile_problem
      246±0ms          252±0ms     1.02  simple_QP_benchmarks.SimpleQPBenchmark.time_compile_problem
      236±0ms          241±0ms     1.02  gini_portfolio.Murray.time_compile_problem
      148±0ms          151±0ms     1.02  high_dim_convex_plasticity.ConvexPlasticity.time_compile_problem
      1.47±0s          1.50±0s     1.02  matrix_stuffing.ParamConeMatrixStuffing.time_compile_problem
      927±0ms          939±0ms     1.01  simple_LP_benchmarks.SimpleScalarParametrizedLPBenchmark.time_compile_problem
     14.3±0ms         14.4±0ms     1.01  simple_QP_benchmarks.ParametrizedQPBenchmark.time_compile_problem
      1.90±0s          1.92±0s     1.01  simple_QP_benchmarks.UnconstrainedQP.time_compile_problem
      690±0ms          696±0ms     1.01  matrix_stuffing.ConeMatrixStuffingBench.time_compile_problem
      1.07±0s          1.08±0s     1.01  finance.FactorCovarianceModel.time_compile_problem
      3.22±0s          3.23±0s     1.00  quantum_hilbert_matrix.QuantumHilbertMatrix.time_compile_problem
      289±0ms          290±0ms     1.00  slow_pruning_1668_benchmark.SlowPruningBenchmark.time_compile_problem
      4.39±0s          4.40±0s     1.00  huber_regression.HuberRegression.time_compile_problem
      5.10±0s          5.10±0s     1.00  optimal_advertising.OptimalAdvertising.time_compile_problem
      854±0ms          852±0ms     1.00  simple_QP_benchmarks.LeastSquares.time_compile_problem
      548±0ms          546±0ms     1.00  semidefinite_programming.SemidefiniteProgramming.time_compile_problem
      5.15±0s          5.11±0s     0.99  svm_l1_regularization.SVMWithL1Regularization.time_compile_problem
      1.65±0s          1.64±0s     0.99  tv_inpainting.TvInpainting.time_compile_problem
      23.0±0s          22.6±0s     0.98  sdp_segfault_1132_benchmark.SDPSegfault1132Benchmark.time_compile_problem
      11.8±0s          11.5±0s     0.97  simple_LP_benchmarks.SimpleLPBenchmark.time_compile_problem
     15.2±0ms         14.5±0ms     0.95  simple_LP_benchmarks.SimpleFullyParametrizedLPBenchmark.time_compile_problem

Use is_affine() check instead of isinstance(Variable) so that any affine expression (e.g., 2*x + 3) passes through without creating unnecessary auxiliary variables and equality constraints. Co-Authored-By: Claude Opus 4.5 <[email protected]>

Keep AFFINE_ARG_ATOMS special handling for coefficient extraction while incorporating latest diffengine-setup changes. Co-Authored-By: Claude Opus 4.5 <[email protected]>

- Add diag_vec to ATOM_CONVERTERS for cp.diag(vector) support - Add automatic reshape when linear_op flattens matrix expressions Co-Authored-By: Claude Opus 4.5 <[email protected]>

Transurgeon and others added 30 commits January 11, 2026 15:03

added converter to multiply

e8f2f91

Clarify MulExpression: matrix @ matrix where both depend on variables

5fd79b2

Examples: X @ Y (both Variables), cos(X) @ sin(Y) Currently supported: A @ f(x) and f(x) @ A where A is constant Co-Authored-By: Claude Opus 4.5 <[email protected]>

Fix E402 lint errors: combine docstrings

0da23b5

Merge license header and module docstring into single docstring to avoid "module level import not at top of file" errors. Co-Authored-By: Claude Opus 4.5 <[email protected]>

debugging

bac71da

super nasty buggit statusgit status

78c9e44

equivalent treatment of (n, ) as numpy and cvxpy. Very subtle

e129758

multiply

7310fd0

converters

bc7abc5

relative entropy converter

d43570f

clean up oracle jacobian

5b3c073

cleaned up jacobian oracle a bit

452c44e

minor

3bd8b4c

cleaned up oracle class

fb292d3

best of fix and added derivative checker class

6d98f03

removed hacky logic with many reshapes to handle numpy's weird broadc…

e3a87d5

…asting rule. This is now internally done in the diff engine

removed skipping of tests (fixed with new matmul convention)

932466d

prod converter

b1196b7

prod with axis one

c445413

matmul

0649032

stress_tests_diff_engine/

f1e20ca

random initial points

a5939d9

test for sum

8f657cd

dance858 and others added 8 commits January 19, 2026 08:52

added power flow as a test and started on transpose converter

d19a9ec

cleaned up convert_matmul so we take advantage of sparsity

698b4f4

added test for sparse matrix vector

1478159

small edit to test

e3d3d76

cleaned up converter of multiply

a831303

clean up quad form converter

a42b587

changed name of test

5b96f19

Transurgeon marked this pull request as draft January 20, 2026 20:08

Transurgeon and others added 6 commits January 20, 2026 15:35

added derivative checker to all tests

87a678a

added hstack in converter and as test

2ed0cd0

Merge diffengine-setup into coeffextractor

3e25c67

Keep AFFINE_ARG_ATOMS special handling for coefficient extraction while incorporating latest diffengine-setup changes. Co-Authored-By: Claude Opus 4.5 <[email protected]>

Merge remote-tracking branch 'origin/master' into coeffextractor

4ccda83

Add diag_vec converter and improve dimension handling for affine atoms

5421279

- Add diag_vec to ATOM_CONVERTERS for cp.diag(vector) support - Add automatic reshape when linear_op flattens matrix expressions Co-Authored-By: Claude Opus 4.5 <[email protected]>

Transurgeon force-pushed the diffengine-setup branch from 32a4206 to 1d81fb4 Compare January 30, 2026 21:20

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Simplify log/exp verification to check is_affine() and add tests #134

Simplify log/exp verification to check is_affine() and add tests #134

Uh oh!

Transurgeon commented Jan 20, 2026

Uh oh!

github-actions bot commented Jan 20, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Simplify log/exp verification to check is_affine() and add tests #134

Are you sure you want to change the base?

Simplify log/exp verification to check is_affine() and add tests #134

Uh oh!

Conversation

Transurgeon commented Jan 20, 2026

Uh oh!

github-actions bot commented Jan 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

github-actions bot commented Jan 20, 2026 •

edited

Loading