def search(self, n_iterations=0, report=False, log=False):
offsprings = []
while len(offsprings) < len(self.population):
off1, off2 = p1, p2 = self.selection(self.population,
self._random_state,
self.pressure)
if self._random_state.uniform() < self.p_c:
self._crossover(p1, p2)
if self._random_state.uniform() < self.p_m:
off1 = self._mutation(off1)
off2 = self._mutation(off2)
if not (hasattr(off1, 'fitness') and hasattr(off2, 'fitness')):
self.problem_instance.evaluate(off1)
self.problem_instance.evaluate(off2)
#self.select_the_offpring(p1, p2, off1, off2, offsprings)
offsprings.extend([off1, off2])
while len(offsprings) > len(self.population):
offsprings.pop()
#print(np.std([offspring.fitness for offspring in offsprings]))
if np.std([offspring.fitness for offspring in offsprings]) < 0.02:
self.selection = uls.parametrized_tournament_selection(0.2)
else:
self.selection = uls.parametrize_roulette_wheel_w_pressure(0.2)
self.population = np.asarray(self.sort_populations(offsprings))
self.best_solution = self._get_elite(self.population)
if report:
self._verbose_reporter_inner(self.best_solution, n_iterations)
if self.flag:
if self.best_solution.fitness - uls.calculate_media_solution(
self.population) < 0.02:
self.selection = uls.parametrize_rank_selection(self.pressure)
else:
self.selection = uls.parametrize_roulette_wheel_w_pressure(
self.pressure)
# restrictions:
# - 5000 f.e./run
# - 50 f.e./generation
# - use at least 5 runs for benchmarks
# ++++++++++++++++++++++++++
n_gen = 100
ps = 50
p_c = .9
p_m = 0.4
radius = .9
pressure = .2
p_migration = 0
ga1 = GeneticAlgorithm(problem_instance=ann_op_i,
random_state=random_state,
population_size=ps,
selection=uls.parametrized_tournament_selection(0.2),
crossover=uls.geometric_semantic_crossover,
p_c=p_c,
mutation=uls.parametrized_ball_mutation(radius),
p_m=p_m,
pressure=pressure)
ga2 = GeneticAlgorithm(
problem_instance=ann_op_i,
random_state=random_state,
population_size=ps,
selection=uls.parametrize_roulette_wheel_w_pressure(0.2),
crossover=uls.geometric_semantic_crossover,
p_c=p_c,
mutation=uls.parametrized_ball_mutation(radius),
p_m=p_m,
#++++++++++++++++++++++++++
# THE PROBLEM INSTANCE
# - optimization of ANN's weights is a COP
#++++++++++++++++++++++++++
ann_op_i = ANNOP(search_space=(-2, 2, n_weights), fitness_function=ann_i.stimulate,
minimization=False, validation_threshold=validation_threshold)
#++++++++++++++++++++++++++
# THE SEARCH
# restrictions:
# - 5000 offsprings/run max*
# - 50 offsprings/generation max*
# - use at least 5 runs for your benchmarks
# * including reproduction
#++++++++++++++++++++++++++
ga1 = GeneticAlgorithm2(ann_op_i, random_state, ps, uls.parametrized_tournament_selection(pressure),
uls.one_point_crossover, p_c, uls.parametrized_ball_mutation(radius), 0.2)
ga2 = GeneticAlgorithm(ann_op_i, random_state, ps, uls.parametrized_tournament_selection(pressure),
uls.two_point_crossover, p_c, uls.parametrized_ball_mutation(radius), 0.1)
sa1 = SimulatedAnnealing(ann_op_i, random_state, ps, uls.parametrized_ball_mutation(radius), control, update_rate)
search_algorithms = [ga1, ga2]
# initialize search algorithms
[algorithm.initialize() for algorithm in search_algorithms]
# execute search
[algorithm.search(n_iterations=n_gen, report=False) for algorithm in search_algorithms]
#++++++++++++++++++++++++++
# TEST
# - test algorithms on unseen data
def search(self, n_iterations, report=False, log=False):
elite = self.best_solution
offsprings = []
while len(offsprings) < len(self.population):
off1, off2 = p1, p2 = [
self.selection(self.population, self.problem_instance.minimization, self._random_state,
self.presure)
for _ in
range(2)]
if self._random_state.uniform() < self.p_c:
off1, off2 = self._crossover(p1, p2)
if self._random_state.uniform() < self.p_m:
off1 = self._mutation(off1)
off2 = self._mutation(off2)
if not (hasattr(off1, 'fitness') and hasattr(off2, 'fitness')):
self.problem_instance.evaluate(off1)
self.problem_instance.evaluate(off2)
offsprings.extend([off1, off2])
while len(offsprings) > len(self.population):
offsprings.pop()
elite_offspring = self._get_elite(offsprings)
elite = self._get_best(elite, elite_offspring)
# Elitism
selectTheBestOffsprings = pd.DataFrame(np.asanyarray([[offspring, offspring.fitness] for offspring
in offsprings]))
selectTheBestPopulation = pd.DataFrame(np.asanyarray([[individual, individual.fitness] for individual
in self.population]))
selectTheBestOffsprings.rename(index=str, columns={0: "Offspring", 1: "Fitness"}, inplace=True)
selectTheBestPopulation.rename(index=str, columns={0: "Individual", 1: "Fitness"}, inplace=True)
selectTheBestOffsprings.sort_values(ascending=False, inplace=True, by="Fitness")
selectTheBestPopulation.sort_values(ascending=False, inplace=True, by="Fitness")
selectTheBestOffspringsList = selectTheBestOffsprings['Offspring'].tolist()
selectTheBestPopulationList = selectTheBestPopulation['Individual'].tolist()
newPopulation = selectTheBestOffspringsList[:24] + selectTheBestPopulationList[:1]
self.population = newPopulation
if elite.fitness == self.best_solution.fitness:
self.repetition = self.repetition+1
else:
self.best_solution = elite
self.repetition = 0
if self.repetition > 3:
if self.selection == uls.parametrize_roulette_wheel(self.presure):
self.selection = uls.parametrized_tournament_selection(self.presure)
self.repetition = 0
else:
self.selection = uls.parametrize_roulette_wheel(self.presure)
self.repetition = 0
if self.repetition >8:
self.p_m = 0.9
if self.p_m == 0.9 and self.repetition<4:
self.p_m = 0.3
#++++++++++++++++++++++++++
# THE OPTIMIZATION
# restrictions:
# - 5000 f.e./run
# - 50 f.e./generation
# - use at least 5 runs for benchmarks
#++++++++++++++++++++++++++
n_gen = 100
ps = 10
p_c = .8
p_m = .8
radius = .002
pressure = .5
ga1 = GeneticAlgorithm(ann_op_i, random_state, ps,
uls.parametrized_tournament_selection(pressure),
uls.cicle_crossover, p_c,
uls.parametrized_gaussian_member_mutation(radius), p_m,
pressure)
ga2 = GeneticAlgorithm(ann_op_i, random_state, ps,
uls.parametrized_tournament_selection(pressure),
uls.cicle_crossover, p_c,
uls.parametrized_gaussian_member_mutation(radius), p_m,
pressure)
ga3 = GeneticAlgorithm(ann_op_i, random_state, ps,
uls.parametrized_tournament_selection(pressure),
uls.cicle_crossover, p_c,
uls.parametrized_gaussian_member_mutation(radius), p_m,
pressure)
import os
import datetime
import logging
import numpy as np
import matplotlib.pyplot as plt
from sklearn import datasets
from sklearn.metrics import accuracy_score
from sklearn.model_selection import train_test_split
import utils as uls
from problems.ANNOP import ANNOP
from ANN.ANN import ANN, softmax, sigmoid
from algorithms.genetic_algorithm_Elitism import GeneticAlgorithm
# setup logger
file_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), "LogFiles/" + (str(datetime.datetime.now().date()) + "-" + str(datetime.datetime.now().hour) + \
"_" + str(datetime.datetime.now().minute) + "_log.csv"))
logging.basicConfig(filename=file_path,
level=logging.DEBUG,
format='%(name)s,%(message)s')
#++++++++++++++++++++++++++
# THE DATA
# restrictions:
# - MNIST digits (8*8)
# - 33% for testing
# - flattened input images
#++++++++++++++++++++++++++
# import data
digits = datasets.load_digits()
flat_images = np.array([image.flatten() for image in digits.images])
print(flat_images.shape)
print(digits.target_names)