README.md

Flexible List-based Evolutionary Algorithm

Author : Yingkai Liu

updated : Aug.2017

---

NOTE

This project has fulfilled my expectation, so I will let it rest for a while This is document for populationclass.py in /old, new documentation will be coming out

THIS WINTER VACATION

---

status: still incomplete, even this documentation is incomplete

1. code overview

This code is written in python3 compatible with python2. Used packages only includes easygui radom itertools to utilize maximum compatibility.

It has a GUI and you can also disable it with minimum modification.

This code is designed to perform evolutionary algorithm on a population whose gene are in the form of lists. The fitness function is an arbitrary user defined function. It can be as simple as sum or any other complicated function such as the how similar of a graph A compared to B.

This code features a relatively wide range of evolutionary operators. mutate hybrid select are three basics ones. To prevent pre-mature, there are two other functions migrate and niche.

The whole package is included in a single class called population. you can add new functions to this class as well, such as adapt (which I am planning to add later) .

Also you might find that sometimes you get one more or one less choromosomes than you expected, that is because I used int() everything to make sure interger. I assume it won't be a big problem for a large population. This will be fixed in later releases.

2. functions and implementation

2.1 class attributes

    def __init__(self,gene_set,gene_range,population_size):
        self.gene_set = gene_set
        self.gene_size = len(gene_set)
        self.gene_range = gene_range

        self.population_size = population_size

        self.chromosomes = self.initialize_chromosomes(self.population_size)

        self.generation_number = 0

2.2 functions

2.2.1 assign_fitness(self,gene)

This is user defined function

2.2.2 initialize_chromosome(self)

Use attributes and assign_fitness to generate a chromosome in the form of

$$chromosome = [ [ gene[site] ] , fitness ]$$

2.2.3 initialize_chromosomes(self,population_size)

This function use initialize_chromosome to generate a list of chromosomes in the form of

$$chromosomes = [ chromosome ]$$

2.2.4 roulette_fitness(self)

This function will use fitness as a bias to choose a chromosome

2.2.5 mutate(chromosome, mutate_strength=1)

mutate mutate_strength sites in the chromosome

2.2.6 migrate(self,immigration_size)

same as initialize chromosomes

2.2.7 hybrid(father,mother)

generate a child using genes randomly chosen from father and mother. If the site of gene are odd, then mother have one more contribution than father. Else choose half of gene_site randomly from father and mother

2.2.8 population_crossover(self,roulette=False,crossover_ratio=0.1)

2.2.9 assign_simiarity(self,pair)

This function takes in chromosomes and extract thier gene and then return the distance of genes as similarity, the distance is simply absulute difference.

you can make it use euclidean distance

The similairty is averaged normalized in range (0,1) the code need to use gene_range[site] to normalize and gene_size to average

similarity = difference / gene_range

the similarity is then averaged over all sites

2.2.10 niche(self, niche_ratio=0.0, niche_criterion=0.0)

This function use assign_similarity and

2.2.11 select(self, select_ratio=0.0, select_criterion=0.0)

2.2.12 evolve(self, niche_ratio=0.1, niche_criterion=0.0)

3.complex ideas

3.1 multiple hybrid

like hybrid using genes from 3 parents. This can be done but is of little use since generating a purmutation list of three parents are slow, and multiple parents is basically the same if you tune up the crossover_ratio and set roulettel = True

3.2 hybrid strictly according to fitness

3.3 make a method ranks the fitness / number each chromosome /...

later

4.acknowledgement

special thanks to Kai Wu for initial discussions, Yifei Liu for brilliant ideas, Qiang Zhu and Xiangfeng Zhou for instruction. USPEX inspired this script