[Python] How to automatically convert objective expression from pyomo model to use as...

I use pyomo to formulate optimization problems and tyically use solvers like for example IPOPT. Now I would like to apply metaheuristic solvers to those optimization problems. I have already heared of frameworks like pymoo or pygad that could be used for such purposes.

However, one always has to formulate the objective and constraints by hand. I would like to find an automated way to fetch the objective or constraints formulations from the pyomo ConcreteModel instance and use it to formulate the fitness_func for pygad.

For my MWE I just create a simple model with a simple objective (and try to pass this objective as a fitness_func to pygad GA), no constraints yet.

Minimum (not as desired) Working Example:

from pyomo import environ as pe
import pygad

model = pe.ConcreteModel()
model.set1 = pe.Set(
initialize=[i for i in range(10)]
)
model.param1 = pe.Param(
model.set1,
initialize={
i: (i+1)**2 for i in model.set1
},
mutable=True
)
model.var = pe.Var(
model.set1,
domain=pe.PositiveReals,
bounds=(2, 5)
)

def obj_rule(model):
return sum(
model.param1 * model.var
for i in model.set1
)

model.obj = pe.Objective(
rule=obj_rule,
sense=pe.maximize
)

solver = pe.SolverFactory('IPOPT')
solver.solve(model)

solution_ipopt = model.obj.expr()

#=== pyomo model done, now comes everything pygad related ======================
n_vars = len(model.set1)

def fitness_func(instance, solution, solution_idx):
# desired: something automated like
# return model.obj.expr
# (but that would be too easy and doesn't work)
return sum(
(i+1)**2 * solution for i in range(1, n_vars)
) # yes, that works but is coded manually, I need something automated

ga_instance = pygad.GA(
num_generations = 100,
num_parents_mating = 10,
fitness_func = fitness_func,
sol_per_pop = 100,
num_genes = n_vars,
gene_type = float,
gene_space = [{'low': 2., 'high': 5.} for _ in range(n_vars)],
parent_selection_type = "tournament",
crossover_type = "single_point",
mutation_type = "random",
mutation_percent_genes = 50,
parallel_processing= ["thread", 8]
)

ga_instance.run()
solution, solution_fitness, solution_idx = ga_instance.best_solution()


In fitness_func I had to manually type sum((i+1)**2 * solution for i in range(1, n_vars)) to build the fitness function according to the model.obj objective function. And I need an automated way (for example I found out, that model.obj.expr (without paranthesis!) gives me something that looks promising, but of course that would be too easy and does not work).

Is there a way to automatically convert the objective function of a pyomo optimization model to an expression suitable to pass to the fitness_func for pygad.GA?

Continue reading...