def __init__(self, netsize, Nsensors=1, Nmotors=1): # Create ising model
self.size = netsize # Network size
self.Ssize = Nsensors # Number of sensors
self.Msize = Nmotors # Number of sensors
self.h = np.zeros(netsize)
self.J = np.zeros((netsize, netsize))
self.max_weights = 2
self.randomize_state()
self.env = MountainCarEnv()
self.env.min_position = -np.pi / 2
self.env.max_position = np.pi / 6
self.env.goal_position = np.pi / 6
self.env.max_speed = 0.045
self.observation = self.env.reset()
self.Beta = 1.0
self.defaultT = max(100, netsize * 20)
self.Ssize1 = 0
self.maxspeed = self.env.max_speed
self.Update(-1)
rewards_smoothed = pd.Series(stats.episode_rewards).rolling(
smoothing_window, min_periods=smoothing_window).mean()
plt.plot(rewards_smoothed)
plt.xlabel("Episode")
plt.ylabel("Episode Reward (Smoothed)")
plt.title("Episode Reward over Time (Smoothed over window size {})".format(
smoothing_window))
fig2.savefig('reward.png')
if noshow:
plt.close(fig2)
else:
plt.show(fig2)
if __name__ == "__main__":
env = MountainCarEnv()
approx = NeuralNetwork()
target = TargetNetwork()
sess = tf.Session()
sess.run(tf.global_variables_initializer())
# Choose one.
#stats = q_learning(sess, env, approx, 3000, 1000)
stats = q_learning(sess,
env,
approx,
1000,
1000,
use_experience_replay=True,
batch_size=128,
target=target)
settings = termios.tcgetattr(sys.stdin)
def getKey():
tty.setraw(sys.stdin.fileno())
rlist, _, _ = select.select([sys.stdin], [], [])
if rlist:
key = sys.stdin.read(1)
else:
key = ''
termios.tcsetattr(sys.stdin, termios.TCSADRAIN, settings)
return key
env = MountainCarEnv() #gym.make('MountainCar-v0')
env.reset()
while True:
env.render()
key = getKey()
if key == 'x':
exit()
elif key == 'r': # repeat this game
ob = env.reset()
print 'Reset:', ob
continue
# Possible actions are: MoveLeft, MoveRight, MoveAhead, MoveBack, LookUp, LookDown, RotateRight, RotateLeft
if key not in moveBindings:
# Plot the episode reward over time
fig2 = plt.figure(figsize=(10,5))
rewards_smoothed = pd.Series(stats.episode_rewards).rolling(smoothing_window, min_periods=smoothing_window).mean()
plt.plot(rewards_smoothed)
plt.xlabel("Episode")
plt.ylabel("Episode Reward (Smoothed)")
plt.title("Episode Reward over Time (Smoothed over window size {})".format(smoothing_window))
fig2.savefig('reward.png')
if noshow:
plt.close(fig2)
else:
plt.show(fig2)
if __name__ == "__main__":
env = MountainCarEnv() #gym.make("MountainCar-v0")
state_dim = env.observation_space.shape[0]
action_dim = env.action_space.n
dqn = DQN(state_dim, action_dim, gamma=0.99)
episodes = 1000
time_steps = 200
epsilon = 0.2
stats = dqn.train(episodes, time_steps, epsilon)
plot_episode_stats(stats)
for _ in range(5):
s = env.reset()
for _ in range(200):
