def squareeval(genome: Genome):
table = []
for i in range(100):
table.append([[i], [i * i]])
score = 0
for experience in table:
x, y = experience[0], experience[1]
output = genome.forward(x)
score += abs(output[0] - y[0])
return score
def squarerooteval(genome: Genome):
score = 0.0
table = []
for i in range(100):
table.append([[i], [np.sqrt(i)]])
for experience in table:
x, y = experience[0], experience[1]
output = genome.forward(x)
score += abs(output[0] - y[0])
return score
def xorevaluate(genome: Genome):
genome.cleaner()
table = [([0, 0], [0]), ([0, 1], [1]), ([1, 0], [1]), ([1, 1], [0])]
score = 0.0
episodes = 20
for _ in range(episodes):
shuffle(table)
for experience in table:
x, y = experience[0], experience[1]
output = genome.forward(x)
score += abs(output[0] - y[0])
return score
def lunareval(genome: Genome):
env = gym.make('LunarLander-v2')
number_of_episode = 20
total_reward = 0
for i_episode in range(number_of_episode):
observation = env.reset()
episode_reward = 0
for t in range(100):
# env.render()
forward_values = genome.forward(observation)
softmax_values = softmax(forward_values)
# action_vector = np.array(np.round(softmax_values))
action = np.random.choice(np.arange(4), p=softmax_values)
observation, reward, done, info = env.step(int(action))
if done:
break
episode_reward += reward
total_reward += episode_reward
env.close()
print(total_reward / number_of_episode)
return total_reward / number_of_episode
def experience1():
# Experience 1
genome = Genome(input_size=2, output_size=1)
genome.nodes += [Node(), Node(), Node()]
genome.connections += [
Connection(0, 3, 0.0),
Connection(0, 4, 0.0),
Connection(0, 5, 0.0),
Connection(1, 3, 0.0),
Connection(1, 4, 0.0),
Connection(1, 5, 0.0),
Connection(3, 2, 0.0),
Connection(4, 2, 0.0),
Connection(5, 2, 0.0)
]
fitter = Fitter(genome=genome, evaluate=xorevaluate)
genome = fitter.fit(episode=5000, timebreak=300, scorebreak=0.1)
genome.cleaner()
table = [([0, 0], [0]), ([0, 1], [1]), ([1, 0], [1]), ([1, 1], [0])]
score = 0
for _ in range(20):
shuffle(table)
for experience in table:
x, y = experience[0], experience[1]
output = genome.forward(x)
score += abs(output[0] - y[0])
print(x, ' : ', output)
print('score : ', score)
print(genome.print_graph())
criterions = ['scores', 'familysizes', 'episodetimes']
for crit in criterions:
plt.plot(fitter.history[crit], label=crit)
mean = [
np.mean(fitter.history[crit])
for _ in range(len(fitter.history[crit]))
]
plt.plot(mean, label=crit + '_mean')
plt.xlabel('episode')
plt.ylabel(crit)
plt.show()