optlang 0.0.3

Formulate optimization problems using sympy expressions and solve them using interfaces to third-party optimization software (e.g. GLPK).

optlang

Vision

optlang provides a common interface to a series of optimization solvers (linear & non-linear) and relies on sympy for problem formulation (constraints, objectives, variables, etc.). Adding new solvers is easy: just sub-class the high-level interface and implement the necessary solver specific routines.

Installation

Install using pip

pip install optlang

Local installations like

python setup.py install

might fail installing the dependencies (unresolved issue with easy_install). Running

pip install -r requirements.txt

beforehand should fix this issue.

Documentation

The documentation for optlang is provided at readthedocs.org.

Dependencies

• sympy >= 0.7.5

• swiglpk >= 0.1.0

• glpk >= 4.45

Example

Formulating and solving the problem is straightforward (example taken from GLPK documentation):

from optlang import Model, Variable, Constraint, Objective

x1 = Variable('x1', lb=0)
x2 = Variable('x2', lb=0)
x3 = Variable('x3', lb=0)

c1 = Constraint(x1 + x2 + x3, ub=100)
c2 = Constraint(10 * x1 + 4 * x2 + 5 * x3, ub=600)
c3 = Constraint(2 * x1 + 2 * x2 + 6 * x3, ub=300)

obj = Objective(10 * x1 + 6 * x2 + 4 * x3, direction='max')

model = Model(name='Simple model')
model.objective = obj
model.add([c1, c2, c3])

status = model.optimize()

print "status:", model.status
print "objective value:", model.objective.value
for var_name, var in model.variables.iteritems():
    print var_name, "=", var.primal

The example will produce the following output:

status: optimal
objective value: 733.333333333
x2 = 66.6666666667
x3 = 0.0
x1 = 33.3333333333

Future outlook

• Gurobi interface (very efficient MILP solver)

• CPLEX interface (very efficient MILP solver)

• Mosek interface (provides academic licenses)

• GAMS output (support non-linear problem formulation)

• DEAP (support for heuristic optimization)

• Interface to NEOS optimization server (for testing purposes and solver evaluation)

• Automatically handle fractional and absolute value problems when dealing with LP/MILP/QP solvers (like GLPK, CPLEX etc.)

Requirements

• Models should always be serializable to common problem formulation languages (CPLEX, GAMS, etc.)

• Models should be pickable

• Common solver configuration interface (presolver, MILP gap, etc.)