varAnd question

[karma-riuk]

Hello,

I'm currently trying to do some mutations over a list of str. The way I set up deap is with the following code:

creator.create("Fitness", base.Fitness, weights=(-1.0,))
creator.create("Individual", list, fitness=creator.Fitness)
toolbox = base.Toolbox()
toolbox.register("attr_str", fuzzer.random_individual)
toolbox.register("individual", tools.initRepeat, creator.Individual, toolbox.attr_str, n=len(types)) # len(types) can be any int from 1 to 3
toolbox.register("population", tools.initRepeat, list, toolbox.individual)
toolbox.register("evaluate", fitness.get_fitness_of_f(globals()[func]))
toolbox.register("mate", deap_crossover)
toolbox.register("mutate", deap_mutate)
toolbox.register("select", tools.selTournament, tournsize=TOURNSIZE)

for i in range(REPS):
    population = toolbox.population(n=NPOP)
    algorithms.eaSimple(population, toolbox, CXPROB, MUPROB, NGEN, verbose=False)

As you can notice, my Individual is a list, that can contain any number (from 1 to 3) of strings.
Additionally, let me give you the deap_mutate function, as it will be useful for later:

def deap_mutate(ind):
    return creator.Individual(fuzzer.mutate(ind))

fuzzer.mutate takes in a list of strings and returns a copy of that list, with a one element mutated.

For some reason, my code kept crashing after a mutation was triggered. I was really confused why so I started digging in the code of the library, and found this: inside the varAnd function, with this piece of code that handles the mutations:

    for i in range(len(offspring)):
        if random.random() < mutpb:
            offspring[i], = toolbox.mutate(offspring[i])
            del offspring[i].fitness.values

My question is related to that little , in the third line, namely the offspring[i],. Why is it there? I am very confused by it's existence and am probably missing something.

The issue it gave me was the following, given that my individuals are a lists, when the list has one element, that element would get unpacked (because of that ,) and cause an error saying

    del offspring[i].fitness.values
        ^^^^^^^^^^^^^^^^^^^^
AttributeError: 'str' object has no attribute 'fitness'

If however my list contains more than one element, then the error would come from the line where the mutation happen, saying

    offspring[i], = toolbox.mutate(offspring[i])
    ^^^^^^^^^^^^^
ValueError: too many values to unpack (expected 1)

Now, there are two fixes I found for this issue. One on my side and one on the side of the library. The one on my side is simply to wrap the individual returned in a list, in order for the mutated individual to be unpacked correctly inside the library:

def deap_mutate(ind):
    return [creator.Individual(fuzzer.mutate(ind))]

The one on the side of the library is to remove that , in the varAnd function:

    for i in range(len(offspring)):
        if random.random() < mutpb:
            offspring[i] = toolbox.mutate(offspring[i])
            del offspring[i].fitness.values

Hence my question, why are you unpacking the result of the mutation?

[fengyangsun]

I'm not sure. But I guess the mutate function is to align with the mate function or crossover function, which returns a tuple of two variables, i.e two child individuals. Mutate only has one mutated individual, so the second returned variable is left empty, But still takes a position. I think the simplest way to solve the issue is, you make the deap_mutate function also return a tuple of two variables, and leave the second empty.

[fmder]

@fengyangsun You are right. Plus theoretically, mutation could return an infinite number of individuals like in evolution strategies.