Advanced usage

Genetic algorithms in their original structure have a very simple workflow: create population, selection, crossover, mutation, new population, and repeat it over and over again. But as genetic algorithms started to receive a lot of academic attention, they started to be more and more complex with numerous tweaks in every part of the workflow.

GeneticAlgos implements some algorithm tweaks and modifications which can be adapted and easily tested. It is important to say that even GeneticAlgos has default parameters which might suit for generic use, but default values are definitely not suitable for all cases. For this reason, it is possible to modify genetic algorithm simulation with the desired parameters.

Change default workflow methods

In GeneticAlgos, we can tweak different methods for GA workflow building blocks:

Change parameters during instantiation

When instantiating GeneticAlgo objects, we can modify the following parameters:

gene_intervals: is a mandatory parameter when creating a GeneticAlgo object. Parameter expect a 2-dimensional numpy.ndarray which specifies:
- chromosome length
- chromosome encoding
- value range for each gene - lower bound is inclusive and upper bound is exclusive. Example: integer chromosome with gene range [0, 3] can have values 0, 1, 2.

import geneticalgos as ga
import numpy as np

# 10 gene binary chromosome
binary_chromosome = np.array([[0, 2]] * 10)
bin_chromosome_type = "int"

# 5 gene int chromosome and every gene values from -10 to 10
num_chromosome_1 = np.array([[-10, 11]] * 5)
num_chromosome_1_type = "int"

# 4 gene float chromosome:
#  - gene 1 values from 0 to 1
#  - gene 2 - 4 have can have values from 0 to 5
num_chromosome_2 = np.array([[0, 1], [0, 5], [0, 5], [0, 5]])
num_chromosome_2_type = "float"

ga_model = ga.GeneticAlgo(
    fitness_function=sum,
    gene_intervals=binary_chromosome,
    chromosome_type=bin_chromosome_type
)

objective_goal: optional parameter specifies whether our fitness function is trying to find a maximum or minimum value for a given problem. Supported values are only maximize and minimize. The default value is maximize and can easily be changed to minimize.

import geneticalgos as ga
import numpy as np

ga_model = ga.GeneticAlgo(
    fitness_function=sum,
    gene_intervals=np.array([[0, 10]] * 5),
    objective_goal="minimize",    # default value is "maximize"
)

population_size: specifies population size. The default value is 100 chromosomes (solutions). It can be changed:

import geneticalgos as ga
import numpy as np

ga_model = ga.GeneticAlgo(
    fitness_function=sum,
    gene_intervals=np.array([[0, 10]] * 5),
    population_size=30,   # default is 100
)

chromosome_type: specifies the type of Encoding. Supported encodings are int and float. The default value is float. It can be changed:

import geneticalgos as ga
import numpy as np

ga_model = ga.GeneticAlgo(
    fitness_function=sum,
    gene_intervals=np.array([[0, 10]] * 5),
    chromosome_type="int",   # default is float
)

Simulation parameters

At the moment, GeneticAlgos supports only one end criterion for simulation and it is a finite number of iterations (generation cycles). Evolution ends when the n_iterations parameter is reached. The default value is 100. The number of iterations can be changed in the simulate method in the form of a parameter to this method:

ga_model.simulate(n_iterations=250)   # default is 100

Miscellaneous parameters

n_pairs: specifies the number of pairs selected to reproduce. By default, it is 1/3 of the population size. We can change the number of pairs with the following command:

ga_model.n_pairs = 40

Hint

When the number of n_pairs * 2 + n_elite is greater than population_size then ValueError is raised with guidance to which interval is allowed for n_pairs.

Explore simulation results

The best solution fitness value and chromosome gene values can be accessed after simulate() is finished.

import geneticalgos as ga
import numpy as np

# trivial GA model:
#   search maximum sum in float chromosome
ga_model = ga.GeneticAlgo(
    fitness_function=sum,
    gene_intervals=np.array([[0, 100]] * 10),
)
ga_model.simulate()

# best chrosomome gene values
print(ga_model.best_chromosome)

# fitness value for best chromosome
print(ga_model.best_fitness)

It is possible to visualize the evolving best fitness value during the evolution process. We use best_chromosomes_history attribute which stores the best fitness values after each generation cycle.

First we have to install matplotlib library.

$ pip install matplotlib

best_chromosomes_history is a List which stores the best fitness values after each generation cycle.

import matplotlib.pyplot as plt

plt.plot(ga_model.best_chromosomes_history)
plt.show()