Repository navigation
Enhancing stop_criteria for Multi-Objective Optimization #314
Description
In the current PyGAD implementation, the stop_criteria parameter references the fitness function's return value to determine when the desired fitness has been achieved. For example:
def fitness_func(ga_instance, solution, solution_idx):
output = numpy.sum(solution * equation_inputs)
fitness = 1.0 / (numpy.abs(output - desired_output) + 0.000001)
return fitness
ga_instance = pygad.GA(
num_generations=200,
sol_per_pop=10,
num_parents_mating=4,
num_genes=len(equation_inputs),
fitness_func=fitness_func,
stop_criteria=["reach_127.4", "saturate_15"])
In this case, the reach_127.4 is pretty clear with what it does. However, the docs do not specify how stop_criteria works in the context of multi-objective optimization, where the fitness function returns multiple values like here:
def fitness_func(ga_instance, solution, solution_idx):
...
return [fitness1, fitness2, ..., fitnessN]
From a quick look at the code I deduced that pygad actually supports it, but it's just not mentioned in the docs. So if I'm correct something like this works just fine for multi-objective optimization:
ga_instance = pygad.GA(
num_generations=200,
sol_per_pop=10,
num_parents_mating=4,
num_genes=len(equation_inputs),
fitness_func=fitness_func,
stop_criteria=["reach_127.4", "reach_130.1", "saturate_15"])
If that's the case, it would be great to have docs updated to reflect this mechanism
Reacted by Ahmed Gad
Activity
ahmedfgad commentedon Feb 6, 2025
Thanks so much!
The documentation is updated and some more bugs are fixed: fd0e18f
Will be available in the next release!
ahmedfgad commentedon Oct 8, 2026
Multi-objective reach criteria now support thresholds for individual objectives, including negative thresholds, and the expected syntax is documented in the stop-criteria documentation. I verified that a run stops when its per-objective thresholds are reached. I am closing this as implemented.