CUTEst loader

In this example, we load a CUTEst problem using the alpaqa.CUTEstProblem class. These problems can be used to run benchmarks.

import contextlib
import gc
import os
from pathlib import Path

import numpy as np

import alpaqa as pa

# Path containing the compiled CUTEst problems
cutest_dir = Path(os.getenv("HOME")) / "opt" / "CUTEst" / "QP"
problem_name = "CBS"

# alpaqa currently only supports one instance of a CUTEst problem at a time 🙃
with contextlib.suppress(NameError):
    del prob  # noqa: F821
gc.collect()
# Load problem
prob = pa.CUTEstProblem(str(cutest_dir / problem_name), sparse=True)

# Extract the problem data
n = prob.num_variables
m = prob.num_constraints
Q, Q_sym = prob.eval_lagrangian_hessian(np.zeros(n), np.zeros(m))
c, q = prob.eval_objective_and_gradient(np.zeros(n))
x_lb = prob.variable_bounds.lower
x_ub = prob.variable_bounds.upper
A, A_sym = prob.eval_constraints_jacobian(np.zeros(n))
g = prob.eval_constraints(np.zeros(n))
g_lb = prob.general_bounds.lower - g
g_ub = prob.general_bounds.upper - g

# You could now pass these matrices to a QP solver, for example.

# Do note the symmetry of the sparse matrices:
# - Q will usually only contain the elements in the upper triangular
#   (as indicated by Q_sym).
# - A will usually be unsymmetric, and store all its elements explicitly
#   (as indicated by A_sym).
print(n, m)
print(A_sym)
print(A.nnz)
print(Q_sym)
print(Q.nnz)