fix: raise ValueError when cp.Parameter receives a sparse value (#2890) by Vivaan-Atharva · Pull Request #3280 · cvxpy/cvxpy

Vivaan-Atharva · 2026-04-05T17:23:12Z

Problem

cp.Parameter silently accepted a scipy sparse matrix as its value,
storing it without error. The failure only surfaced later during
canonicalization with a confusing traceback.

Reproduction

import cvxpy as cp
x = cp.Variable((2, 2), value=[[1, 0], [0, 1]])
grad = (x + 1).grad[x]  # sparse matrix
p = cp.Parameter((4, 4), value=grad)  # should error here, didn't

Fix

Override the value setter in Parameter to check scipy.sparse.issparse()
before delegating to Leaf.value.fset. Catches the error at assignment time
with a clear message pointing to cp.Constant as the alternative.

Tests

Rejects CSC, CSR, COO sparse arrays at construction time
Rejects sparse via .value = setter after construction
Reproduces the exact case from the issue (grad sparse matrix)
Verifies dense, None, scalar still accepted

github-actions · 2026-04-05T17:30:34Z

Benchmarks that have improved:

   before           after         ratio
 [f61ac402]       [6788a1e6]

    3.34±0s          2.95±0s     0.88  quantum_hilbert_matrix.QuantumHilbertMatrix.time_compile_problem

Benchmarks that have stayed the same:

   before           after         ratio
 [f61ac402]       [6788a1e6]
      1.01±0s          1.04±0s     1.03  finance.FactorCovarianceModel.time_compile_problem
      277±0ms          282±0ms     1.02  matrix_stuffing.ParamSmallMatrixStuffing.time_compile_problem
      1.39±0s          1.41±0s     1.02  matrix_stuffing.ParamConeMatrixStuffing.time_compile_problem
      4.56±0s          4.60±0s     1.01  svm_l1_regularization.SVMWithL1Regularization.time_compile_problem
      227±0ms          229±0ms     1.01  gini_portfolio.Murray.time_compile_problem
      12.6±0s          12.7±0s     1.01  finance.CVaRBenchmark.time_compile_problem
      315±0ms          317±0ms     1.01  gini_portfolio.Yitzhaki.time_compile_problem
      526±0ms          530±0ms     1.01  semidefinite_programming.SemidefiniteProgramming.time_compile_problem
      726±0ms          731±0ms     1.01  matrix_stuffing.ConeMatrixStuffingBench.time_compile_problem
      134±0ms          135±0ms     1.01  high_dim_convex_plasticity.ConvexPlasticity.time_compile_problem
      946±0ms          950±0ms     1.00  gini_portfolio.Cajas.time_compile_problem
      3.95±0s          3.96±0s     1.00  huber_regression.HuberRegression.time_compile_problem
     14.2±0ms         14.2±0ms     1.00  simple_LP_benchmarks.SimpleFullyParametrizedLPBenchmark.time_compile_problem
     42.7±0ms         42.6±0ms     1.00  matrix_stuffing.SmallMatrixStuffing.time_compile_problem
      21.0±0s          20.9±0s     0.99  sdp_segfault_1132_benchmark.SDPSegfault1132Benchmark.time_compile_problem
      1.81±0s          1.79±0s     0.99  simple_QP_benchmarks.UnconstrainedQP.time_compile_problem
      239±0ms          236±0ms     0.99  simple_QP_benchmarks.SimpleQPBenchmark.time_compile_problem
      5.29±0s          5.24±0s     0.99  optimal_advertising.OptimalAdvertising.time_compile_problem
      906±0ms          896±0ms     0.99  simple_LP_benchmarks.SimpleScalarParametrizedLPBenchmark.time_compile_problem
      10.4±0s          10.3±0s     0.98  simple_LP_benchmarks.SimpleLPBenchmark.time_compile_problem
      287±0ms          279±0ms     0.97  slow_pruning_1668_benchmark.SlowPruningBenchmark.time_compile_problem
      1.64±0s          1.60±0s     0.97  tv_inpainting.TvInpainting.time_compile_problem
     15.5±0ms         14.9±0ms     0.96  simple_QP_benchmarks.ParametrizedQPBenchmark.time_compile_problem
      783±0ms          745±0ms     0.95  simple_QP_benchmarks.LeastSquares.time_compile_problem

SteveDiamond · 2026-04-06T01:35:10Z

 from __future__ import annotations

 import cvxpy.lin_ops.lin_utils as lu
+import scipy.sparse as sp


please install the pre-commit

Donee, installed pre commit and ran it, ruff fixed the import ordering.

SteveDiamond · 2026-04-06T01:38:26Z

+    def value(self, val) -> None:
+        if sp.issparse(val):
+            raise ValueError(
+                "Setting a cp.Parameter value to a sparse matrix or array "


This error message is not quite right. We have a distinction now between .value and .value_sparse. I think now that Scipy has ND sparse arrays we can pass in sparse inputs via .value.

Updated the message, it no longer claims Parameters are dense only. It now points to the sparsity attribute + .value_sparse for sparse parameters, and cp.Constant() for the constant use case.

…y#2890)

…mport

…rameter

PTNobel · 2026-04-16T18:26:45Z

                "To use a sparse constant, use cp.Constant() instead."
            )
-        Leaf.value.fset(self, val)
+        Leaf.value.fset(self, val)  # type: ignore[misc]


We generally avoid just masking pyright errors; can you explain why we can't fix the underlying issue? I think the problem is we don't have an annotation on val.

ah got it, added an assert to narrow the type instead

PTNobel · 2026-04-16T18:27:56Z

Why is this a new file? If we keep it as a new file it needs the copyright header etc.

Overall I think this is way too many tests

Why is this a new file? If we keep it as a new file it needs the copyright header etc.

moved everything into test_parameters_failures, deleted the new file

PTNobel · 2026-04-16T18:28:18Z

+        p.value = None
+        assert p.value is None
+
+    def test_accepts_scalar(self):


Surely we already have this test somewhere?

yeah removed those, they're already in test_parameters_successes

PTNobel · 2026-04-16T18:29:00Z

+
+    @staticmethod
+    def _make_sparse_csc(n: int = 4) -> sp.csc_array:
+        return sp.eye(n, format="csc")


Suggested change

return sp.eye(n, format="csc")

return sp.eye_array(n, format="csc")

PTNobel · 2026-04-16T18:30:21Z

+
+    def test_rejects_sparse_csc_at_construction(self):
+        with pytest.raises(ValueError, match="sparse"):
+            cp.Parameter((4, 4), value=self._make_sparse_csc(4))


Why are we using a helper method here but not elsewhere?

removed the helper, just inlined sp.eye_array directly

SteveDiamond reviewed Apr 6, 2026

View reviewed changes

fix: raise ValueError when cp.Parameter receives a sparse value (cvxp…

7644bb2

…y#2890)

Vivaan-Atharva force-pushed the fix/parameter-rejects-sparse-value branch from ec120d0 to 7644bb2 Compare April 14, 2026 13:17

Vivaan-Atharva added 3 commits April 16, 2026 17:32

fix: move scipy.sparse import inside value setter to avoid circular i…

567dcf9

…mport

test: update test_parameters_successes to use dense value for diag Pa…

657b739

…rameter

fix: suppress pyright false positive on Leaf.value.fset

1b4b4dc

PTNobel reviewed Apr 16, 2026

View reviewed changes

	return sp.eye(n, format="csc")
	return sp.eye_array(n, format="csc")

Conversation

Vivaan-Atharva commented Apr 5, 2026

Problem

Reproduction

Fix

Tests

Uh oh!

github-actions Bot commented Apr 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

github-actions Bot commented Apr 5, 2026 •

edited

Loading