from rltoolbox.algorithm.cmac import *
from rltoolbox.environment.continuous import BallBeam
from rltoolbox.approximator import CMACApproximator
from rltoolbox.misc import compare_learning_curves, plot_learning_stats
if __name__ == "__main__":
algorithm = CMACSARSA
environment = BallBeam(max_steps=10000)
environment.approximate_with(CMACApproximator, n_layers=4)
n_episodes = 10
n_repeats = 5
book_parameters = {
'alpha': 1,
'gamma': 0.995,
'epsilon': 0.1,
'lambd': 0.0
}
learning_curves = np.zeros((n_repeats, n_episodes))
for i in range(n_repeats):
alg_instance = algorithm(environment, **book_parameters)
alg_instance.learn(n_episodes, render=False)
learning_curves[i, :] = alg_instance.steps_per_episode
plot_learning_stats(learning_curves,
'BallBeam CMACSARSA learning',
savefig=False)
#!/usr/bin/env python
import numpy as np
from rltoolbox.algorithm.fuzzy import *
from rltoolbox.environment.continuous import BallBeam
from rltoolbox.approximator import FuzzyApproximator
from rltoolbox.misc import compare_learning_curves, plot_learning_stats
if __name__ == "__main__":
algorithm = FR
environment = BallBeam(max_steps=10000)
environment.approximate_with(FuzzyApproximator)
n_episodes = 5
n_repeats = 5
book_parameters = {'alpha': 0.1, 'beta': 0.01, 'epsilon': 0.1, 'lambd': 0.5}
learning_curves = np.zeros((n_repeats, n_episodes))
for i in range(n_repeats):
alg_instance = algorithm(environment, **book_parameters)
alg_instance.learn(n_episodes, render=False)
learning_curves[i, :] = alg_instance.steps_per_episode
plot_learning_stats(learning_curves, 'BallBeam FR learning')
#!/usr/bin/env python
import numpy as np
from rltoolbox.algorithm.fuzzy import *
from rltoolbox.environment.continuous import BallBeam
from rltoolbox.approximator import FuzzyApproximator
from rltoolbox.misc import compare_learning_curves, plot_learning_stats
if __name__ == "__main__":
algorithm = FQ
environment = BallBeam(max_steps=10000)
environment.approximate_with(FuzzyApproximator)
n_episodes = 5
n_repeats = 5
book_parameters = {
'alpha': 0.1,
'epsilon': 0.1,
'gamma': 0.995,
'lambd': 0.5
}
learning_curves = np.zeros((n_repeats, n_episodes))
for i in range(n_repeats):
alg_instance = algorithm(environment, **book_parameters)
alg_instance.learn(n_episodes, render=False)
learning_curves[i, :] = alg_instance.steps_per_episode
plot_learning_stats(learning_curves, 'BallBeam FQ learning', savefig=False)
#!/usr/bin/env python
from rltoolbox.algorithm.classic import Q
from rltoolbox.environment.continuous import BallBeam
from rltoolbox.approximator import TableApproximator
from rltoolbox.misc import plot_learning_stats
environment = BallBeam(max_steps=10000)
environment.approximate_with(TableApproximator)
algorithm_instance = Q(environment,
alpha=0.01,
lambd=0.5,
epsilon=0.1,
gamma=0.995)
algorithm_instance.learn(n_episodes=20, render=False)
plot_learning_stats(algorithm_instance.steps_per_episode,
title='BallBeam Q(lambda)-learning')