geneticalpy 0.1.1

GeneticAlPy - Genetic Algorithms with Python

GeneticAlPy

Genetic Algorithms with Python

Yet another package for lightweight applications of GA in Python.

This package provides utilities for implementation of Genetic Algorithm (Holland 1962) for multivariate, multimodal optimization problems. Check out an introduction from a biological perspective in my paper, Nayak & Saha (2022)

By default, the binary representation is chosen as the genotype. The package provides roulette wheel, stochastic universal sampling and rank selection for choosing parents at each generation, uniform crossover, and bit-flip mutation. Elitist and liberal strategies may also be employed optionally.

Quickstart

Install the package with pip

pip install geneticalpy

Imports

from geneticalpy import genetical, examples
import numpy as np
import matplotlib.pyplot as plt

Define a fitness function from your cost function

cost = examples.ackley # replace this with your objective

def fitness(x): # to be maximized, by definition
    return 1/(1 + cost(x)) # this also depends loosely on your objective

Set initial GA hyperparameters

decimals   = 4
n_var      = 2
var_ranges = np.array([[-10,]*n_var,[10,]*n_var])
n_bits_segment    = len(format(int(max(var_ranges[1] - var_ranges[0])*10**decimals), 'b'))
n_bits_chromosome = n_bits_segment * n_var
offset     = -var_ranges[0]
init_popsize = 400

Initialize GA

PopGen = genetical.PopGenetics(
    fitness_func = fitness,
    n_var = n_var,
    decimal_acc = decimals,
    n_bits_chromosome = n_bits_chromosome,
)
pop_bin = PopGen.initialize_population(
    popsize = init_popsize, 
    var_ranges = var_ranges, 
    return_genotype = True,
    offset = offset,
)

Run evolution

evol_rec = PopGen.evolve(
    pop_bin, 
    n_gen    = 50, 
    n_pairs  = 800,
    elitist  = True,
    n_elites = 3,
    liberal  = False,
    n_runts  = 0,
    switch_selection = 5,
    prob_mut = 0.02, 
    prune    = True,
    pruning_cutoff = 800,
    verbose  = True,
    n_workers = 1,
)

Look at the results

print(f"Best solution is {evol_rec['fittest_individual']} with fitness {evol_rec['best_overall_fitness']}.")

plt.figure(figsize = (5,3))
gens = np.arange(len(evol_rec['best_fitness_per_generation']))+1
plt.plot(gens, evol_rec['best_fitness_per_generation'], color = 'green', label = 'Best')
plt.plot(gens, evol_rec['mean_fitness_per_generation'], color = 'blue', label = 'Mean')
plt.plot(gens, evol_rec['median_fitness_per_generation'], color = 'indigo', label = 'Median')
plt.legend()

plt.xlim(gens[0], gens[-1])
plt.xlabel('Generations')
plt.ylabel('Fitness')
plt.show()

Tutorial

Check out the notebook at tutorial/tutorial.ipynb for a lightning tutorial of GeneticAlPy!

Get in touch

Drop me an email at [e-mail] in case of any questions or to request more functionality!