proxsuite

0.7.2.post1 verified Fri May 01 auth: no python

The ProxSuite library provides an efficient QP solver tailored for robotics and optimization, with implementations of the ProxQP and OSQP algorithms. The current version is 0.7.2.post1, released under a permissive license. It supports Python >= 3.8 and has a bimonthly release cadence.

pip install proxsuite

Common errors

error ModuleNotFoundError: No module named 'proxqp' ↓

cause Direct import of 'proxqp' instead of using 'proxsuite.proxqp'

fix

Use 'from proxsuite import proxqp' or 'import proxsuite.proxqp as proxqp'.

error ValueError: The matrix is not symmetric positive definite. ↓

cause The Hessian matrix H or constraint matrix C has not been properly formatted or the QP is infeasible.

fix

Verify that H is symmetric positive semidefinite (e.g., cast to symmetric: H = (H + H.T)/2). Check problem dimensions.

error AttributeError: module 'proxsuite' has no attribute 'quadprog' ↓

cause Old code using removed quadprog wrapper.

fix

Update to use proxsuite.proxqp. See migration guide in documentation.

error ImportError: cannot import name 'solve' from 'proxsuite.proxqp' ↓

cause Incorrect import path. The solve function is inside a backend submodule (dense/sparse).

fix

Use 'from proxsuite.proxqp.dense import solve' (or sparse).

Warnings

breaking The 'proxsuite.quadprog' module (which wrapped quadprog) was removed in version 0.4.0. Use 'proxsuite.proxqp' instead. ↓

fix Replace 'from proxsuite import quadprog' with 'from proxsuite import proxqp'. The API is different; see migration guide.

breaking The 'warm_start' parameter in solve() was renamed to 'warm_start' in version 0.6.0. The old name 'warm_start' is still accepted but deprecated. ↓

fix Use 'warm_start' (with underscore) consistently. The old spelling may be removed in future releases.

gotcha Do not pass NumPy arrays with dtype=float32. The solver expects float64. Silently truncates precision leading to unexpected results. ↓

fix Ensure arrays are float64: array.astype(np.float64).

deprecated The 'results.x' attribute for primal solution is deprecated in favor of 'results.x_solution' (from v0.7.0). ↓

fix Use qp.results.x_solution instead of qp.results.x.

Imports

proxsuite
```
import proxsuite
```
The main module. Submodules like proxsuite.proxqp are accessed via dot notation.
proxqp
wrong
```
import proxqp
```
correct
```
from proxsuite import proxqp
```
Direct import 'proxqp' is not available; it is a submodule of proxsuite.

Quickstart

Minimal example solving a dense QP with inequality constraints.

import proxsuite
import numpy as np

# Solve a simple QP: min 0.5 x'Hx + g'x s.t. Cx <= d
H = np.eye(2)
g = np.array([-2, -3])
C = np.array([[1, 2], [-1, 0]])
d = np.array([4, 0])

# Create solver and solve
qp = proxsuite.proxqp.dense.QP(H, g, C, d)
proxsuite.proxqp.dense.solve(qp)
print(qp.results.x)