def evolutionary_algorithm(fitness_fn,
objective_fn,
logger,
pop_size=100,
cx_prob=0.8,
mut_prob=0.05,
mut_flip_prob=0.1,
generations=100):
pop = create_population(pop_size)
for G in range(generations):
fit_obj = list(
map(lambda pair: utils.FitObjPair(pair[0], pair[1]),
zip(map(fitness_fn, pop), map(objective_fn, pop))))
evals = G * pop_size
logger.add_gen(fit_obj, evals)
# log.append((G, max(fits), sum(fits)/100, G*POP_SIZE))
#print(G, sum(fits), max(fits)) # prints fitness to console
mating_pool = selection(pop, [f.fitness for f in fit_obj])
offspring = operators(mating_pool, mut_prob, mut_flip_prob, cx_prob)
#pop = offspring[:-1]+[max(pop, key=fitness)] #SGA + elitism
pop = offspring[:] #SGA
logger.write_files()
return pop
# for i in range():
# random.seed(i)
# pop,log = evolutionary_algorithm(fitness)
# logs.append(log)
# fits = list(map(fitness, pop))
# # pprint.pprint(list(zip(fits, pop)))
# # print(sum(fits), max(fits))
# # pprint.pprint(log)
#
# # extract fitness evaluations and best fitnesses from logs
# evals = []
# best_fit = []
# for log in logs:
# evals.append([l[3] for l in log])
# best_fit.append([l[1] for l in log])
#
# evals = np.array(evals)
# best_fit = np.array(best_fit)
#
# # plot the converegence graph and quartiles
# import matplotlib.pyplot as plt
# _, ax = plt.subplots()
#
# ax.plot(evals[0,:], np.median(best_fit, axis=0), label = "median")
# ax.fill_between(evals[0,:], np.percentile(best_fit, q=25, axis=0),
# np.percentile(best_fit, q=75, axis=0), alpha = 0.2, label = "Interquartile range")
#
# legend = ax.legend(loc="upper center", shadow = True, fontsize = 'large')
#
#
# plt.show()
def fitness(ind, cities):
# quickly check that ind is a permutation
num_cities = len(cities)
assert len(ind) == num_cities
assert sum(ind) == num_cities*(num_cities - 1)//2
dist = 0
for a, b in zip(ind, ind[1:]):
dist += distance(cities[a], cities[b])
dist += distance(cities[ind[-1]], cities[ind[0]])
return utils.FitObjPair(fitness=-dist,
objective=dist)
def fitness(ind, weights):
bw = bin_weights(weights, ind)
return utils.FitObjPair(fitness=1 / (max(bw) - min(bw) + 1),
objective=max(bw) - min(bw))
def fitness(ind, train_data, test_data):
return utils.FitObjPair(fitness=accuracy(ind, train_data),
objective=1 - accuracy(ind, test_data))
def fitness(ind, weights):
bw = bin_weights(weights, ind)
for i in range(len(bw)):
return utils.FitObjPair(fitness=1 / (average(bw) - min(bw)),
objective=(average(bw) - min(bw)))
def best_fitness(ind, weights):
bw = bin_weights(weights, ind)
M = sum(bw) / len(bw)
S = sum(map(lambda w: abs(w - M), bw))
F = 0.99**S
return utils.FitObjPair(fitness=F, objective=S)
def better_fitness(ind, weights):
bw = bin_weights(weights, ind)
M = sum(bw) / len(bw)
S = sum(map(lambda w: (w - M)**2, bw))**(1 / len(bw))
return utils.FitObjPair(fitness=1 / (S + 1), objective=max(bw) - min(bw))
