def evolve(self, popsize=300, ngen=40):
"""run back test with genious algorithm to optimize
:param popsize: population size of genious algorithm
:param ngen: number of generations"""
low = -1
high = 1
prototype = [
np.random.randint(low, high) for _ in range(self.index_rows_number)
]
bounds = [[low, high] for _ in range(self.index_rows_number)]
# self.test(prototype)
p = Population(prototype=prototype,
gene_bounds=bounds,
fitness_func=self.test)
p.populate(popsize=popsize)
p.evolve(ngen=ngen)
print(p.best.genes)
from tinyevolver import Population import random random.seed(1234) proto = [1.0 for _ in range(100)] bounds = [(0, random.uniform(0, 10)) for _ in range(100)] pop = Population(proto, bounds, sum) pop.populate(500) pop.evolve(100)
from tinyevolver import Population
"""
In this example we consider individuals consisting entirely of random
booleans. The more 'True's the better (fitter) the individual.
"""
# Our prototype individual is a list of (arbitrary) boolean geans
prototype = [False for _ in range(100)]
# This statement initialises a population class with appropriate methods.
p = Population(prototype=prototype, gene_bounds=None, fitness_func=sum)
# Create the desired number of individuals in the class:
p.populate(popsize=300)
# Evolve! It's as simple as that.
p.evolve(verbose=True)
bounds = [(0.0, 1.0), (0, 3), (0, 5.0)]
# How fit an individual is will depend on how well it approximates the
# data. So let's cook up some data:
times = range(20)
data = [0.5 * time ** 2 + 1.0 + random.uniform(0, 10) for time in times]
def fitness(ind):
curve = [ind[0] * time ** ind[1] + ind[2] for time in times]
square_error = [(f - d) ** 2 for f, d in zip(curve, data)]
# More error = less fit
try:
return 20.0 / sum(square_error)
except ZeroDivisionError:
return float('inf')
# Now to populate and evolve:
p = Population(prototype, bounds, fitness)
p.populate()
p.evolve()
# Let's see how we did:
if plt:
best_ind = p.best
best_fit = [best_ind[0] * time ** best_ind[1] + best_ind[2] for time in times]
plt.plot(times, data)
plt.plot(times, best_fit)
plt.show()
prototype = [0.0 for _ in range(6)] # this goes up to x^5
bounds = [(-1.0, 1.0) for _ in range(6)]
# Construct the noisy sine curve to be approximated
xs = np.linspace(0, 3)
ys = np.sin(xs) + np.random.normal(0.0, 0.1, 50)
# An individual with a geater mean-absolute-error is less fit:
def fitness(ind):
return -np.mean(np.abs(ys - sum(ind[n] * xs ** n for n in range(6))))
# Now we're good to go:
p = Population(prototype, bounds, fitness)
p.populate(500)
p.evolve(ngen=50)
# If you have matplotlib installed, we can plot the best result:
try:
import matplotlib.pyplot as plt
best_ind = p.best
plt.plot(xs, ys)
plt.plot(xs, sum(best_ind[n] * xs ** n for n in range(6)))
plt.show()
except ImportError:
pass