Variable length chromosome

[RomekRJM]

Hi,

I was checking the docs and could not find any details of the module support for variable-length chromosomes.
Is it doable with PyGAD? Can you provide code samples?

Thanks!

[ahmedfgad]

Hi @RomekRJM,

The GitHub repository is updated to support changing the the chromosome length from one generation to another. This will be available in the next release 3.3.0.

But still the individual chromosomes must all have the same length in the same generation.

[leonardolombardi07]

Hi!

First and foremost, I want to express my gratitude for this incredible library and its clear documentation. Congratulations @ahmedfgad on the excellent work.
Thanks!

Now, version 3.3.0 is released... 👀 maybe we already have the variable chromossome feature? 👀

If not, my approach would simply have a big length gene_space and consider some special values (like -1) as the absent of that gene (I'm using "gene" as sinonym of "chromossome" here, but I know there's a small conceptual difference).

In my specific case, my optimization involves positioning weights in certain positions (which must be integers in a given range).
I don't know before hand how many weights I need, but I know it won't be bigger than, let's say, 5 weights. So I could have:

gene_space = [
range(-1, max_weight),
range(-1, max_weight),
range(-1, max_weight),
range(-1, max_weight),
range(-1, max_weight),
]

(I know I could just use gene_space=range(-1, max_weight),, but was to clarify that each individual would have 5 genes).

Then, on my fitness function I would have something like:

import pygad
from typing import Tuple, Literal


WeightPositionAsNone = Literal[-1]
MaxWeightPosition = Literal[5]
WeightPosition = Literal[WeightPositionAsNone,
                         0, 1, 2, 3, 4, MaxWeightPosition]

Solution = Tuple[
    WeightPosition,
    WeightPosition,
    WeightPosition,
    WeightPosition,
    WeightPosition,
]


def fitness_func(
        ga_instance: pygad.GA,
        solution: Solution,
        solution_idx: int
):
    filtered_solution = [x for x in solution if x != -1]

    # Calculate fitness solution for the problem
    # desconsidering the weights that are not present
    fitness = sum(filtered_solution)

    return fitness

Does this seem like a valid approach?

[ahmedfgad]

This is absolutely a good idea. Still the feature is not fully merged. It is a major change.