optlang-copt

optlang-copt is a standalone adapter that makes Cardinal Optimizer (COPT) available through optlang and COBRApy without modifying either installed package.

Supported scope

• Linear programming (LP)
• Mixed-integer linear programming (MILP)
• Linear and convex quadratic objectives (QP/MIQP)
• Variable and constraint addition/removal
• Bound, objective-coefficient, and constraint-coefficient changes
• Primal values, objective values, LP dual values, and reduced costs
• COBRApy FBA, pFBA, FVA, and common reaction-bound/knockout workflows
• COBRApy linear and quadratic MOMA
• COBRApy linear/integer ROOM and loopless linear/MILP workflows
• Python 3.7+ with optlang 1.5.2+ and COPT 8.x

Quadratic constraints (QCP/MIQCP), indicator/SOS constraints, callbacks, native COPT model serialization, and advanced basis operations are not part of version 0.2.0.

Installation

COPT must be licensed on the machine. Install the adapter and its Python dependencies from PyPI:

python -m pip install optlang-copt

When coptpy was installed from the official offline COPT package, use:

python -m pip install --no-deps optlang-copt

Automatic discovery and COBRApy usage

The wheel installs a Python startup hook scoped to the current environment. It registers COPT with optlang and COBRApy automatically, so application code does not need import optlang_copt:

import cobra

model = cobra.io.load_model("textbook")
model.solver = "copt"

solution = model.optimize()
print(solution.status, solution.objective_value)

To make new COBRApy models use COPT by default in the current process:

import cobra

cobra.Configuration().solver = "copt"

This registration does not edit COBRApy files and does not remove Gurobi, GLPK, or other interfaces. Set OPTLANG_COPT_AUTO_REGISTER=0 before starting Python to disable the startup hook. Explicit import optlang_copt remains available when auto-registration is disabled.

MOMA

Existing COBRApy calls remain unchanged after selecting COPT:

from cobra.flux_analysis import moma, pfba

model.solver = "copt"
reference = pfba(model)

with model:
    model.reactions.PGI.knock_out()
    linear_result = moma(model, solution=reference, linear=True)

with model:
    model.reactions.PGI.knock_out()
    quadratic_result = moma(model, solution=reference, linear=False)

Version 0.2.0 rebuilds the native COPT model when COBRApy removes QP auxiliary variables during context rollback. This avoids stale native quadratic references in COPT 8.0.6 and allows subsequent FBA, ROOM, or loopless calculations to continue on the same model.

Direct optlang usage

import optlang

copt = optlang.copt_interface

x = copt.Variable("x", lb=0, ub=10)
y = copt.Variable("y", lb=0, ub=10)
model = copt.Model()
model.add([x, y, copt.Constraint(x + y, ub=12, name="capacity")])
model.objective = copt.Objective(3*x + 2*y, direction="max")
print(model.optimize(), model.objective.value)

Development test

pytest -q

This project is an independent compatibility layer and is not an official component of Cardinal Operations, optlang, or COBRApy.