def go2():
env = CartPoleEnv()
episode_step_counter = 0
for i_episode in range(10000):
action = env.reset()
step_counter = 0
while True:
env.render()
# 随机选择一个action
# 获取环境给予的奖励
observation_, reward, done, info = env.step(action)
x, x_dot, theta, theta_dot = observation_
r1 = (env.x_threshold - abs(x)) / env.x_threshold - 0.8
r2 = (env.theta_threshold_radians -
abs(theta)) / env.theta_threshold_radians - 0.5
reward = r1 + r2
print(reward)
step_counter = step_counter + 1
if done:
episode_step_counter += step_counter
# print("第{}回合，坚持了{}步".format(i_episode, step_counter))
print("平均步数:{}".format(episode_step_counter / (i_episode + 1)))
break
env.close()
def save():
env = CartPoleEnv()
total_steps = 0
memory = []
memory_counter = 0
for i_episode in range(100):
observation = env.reset()
while True:
env.render()
action = env.action_space.sample()
observation_, reward, done, info = env.step(action)
x, x_dot, theta, theta_dot = observation_
r1 = (env.x_threshold - abs(x)) / env.x_threshold - 0.8
r2 = (env.theta_threshold_radians -
abs(theta)) / env.theta_threshold_radians - 0.5
reward = r1 + r2
transition = np.hstack((observation, [action,
reward], observation_))
memory.append(transition)
if done:
break
observation = observation_
total_steps += 1
memory = np.array(memory)
np.save("memory.npy", memory)
env.close()
def go():
env = CartPoleEnv()
total_steps = 0
memory = []
model = create_model()
epsilon = 0.9
memory_counter = 1000
for i_episode in range(1000):
observation = env.reset()
ep_r = 0
while True:
env.render()
if np.random.uniform() < epsilon:
actions_value = model.predict(np.array([observation]))
action = np.argmax(actions_value)
else:
action = env.action_space.sample()
observation_, reward, done, info = env.step(action)
x, x_dot, theta, theta_dot = observation_
r1 = (env.x_threshold - abs(x)) / env.x_threshold - 0.8
r2 = (env.theta_threshold_radians -
abs(theta)) / env.theta_threshold_radians - 0.5
reward = r1 + r2
transition = np.hstack((observation, [action,
reward], observation_))
memory.append(transition)
if len(memory) > memory_counter:
xx, yy = get_data(np.array(memory), model)
print(xx.shape)
model.fit(xx, yy, epochs=10)
epsilon = epsilon + 0.00001
memory = []
# memory_counter = memory_counter + 5
ep_r = ep_r + reward
if done:
# print(ep_r)
break
observation = observation_
total_steps += 1
model.save("logs/cp.h5")
model.summary()
env.close()
def get():
env = CartPoleEnv()
for i_episode in range(10000):
observation = env.reset()
action = chose_action(model=model)
while True:
observation_, reward, done, info = env.step(action)
x, x_dot, theta, theta_dot = observation_
r1 = (env.x_threshold - abs(x)) / env.x_threshold - 0.8
r2 = (env.theta_threshold_radians - abs(theta)) / env.theta_threshold_radians - 0.5
reward = r1 + r2
ransition = np.hstack((observation, [action, reward], observation_))
print()
def test_discrete_vectorized_original_equality(self):
venv = DiscreteVectorizedCartPoleEnv()
state, action = self.state_action
action = (action > 0).astype(int)
dim1, dim2 = self.dims
venv.state = state
vobs, vreward, vdone, _ = venv.step(action)
env = CartPoleEnv()
for i in range(dim1):
for j in range(dim2):
env.reset()
env.state = state[i, j]
obs, reward, done, _ = env.step(int(action[i, j, 0]))
np.testing.assert_allclose(obs, vobs[i, j])
np.testing.assert_allclose(reward, vreward[i, j])
np.testing.assert_allclose(done, vdone[i, j])
class CartPoleDictEnvWrapper(gym.Env):
def __init__(self, max_angle=12, max_num_steps=1000):
self.env = CartPoleEnv()
# self.env.theta_threshold_radians = max_angle * 2 * math.pi / 360
self.observation_space = self.env.observation_space
self.action_space = self.env.action_space
self.step_counter = 0
self.max_num_steps = max_num_steps
def step(self, action):
if isinstance(action, numpy.ndarray):
action = action[0]
assert isinstance(action, numpy.int64)
obs, _, done, _ = self.env.step(action)
self.step_counter += 1
if self.step_counter % self.max_num_steps == 0:
done = True
if done:
reward = -10.0
obs = self.env.reset()
else:
reward = 0.0
return {"observation": obs, "reward": reward, "done": int(done)}
def reset(self):
obs = self.env.reset()
return {"observation": obs, "reward": 0.0, "done": int(False)}
def render(self, mode="human"):
return self.env.render(mode)
def close(self):
self.env.close()
def seed(self, seed=None):
return self.env.seed(seed)
r1 = (env.x_threshold - abs(x)) / env.x_threshold - 0.8
r2 = (env.theta_threshold_radians - abs(theta)) / env.theta_threshold_radians - 0.5
reward = r1 + r2
ransition = np.hstack((observation, [action, reward], observation_))
print()
if __name__ == '__main__':
env = CartPoleEnv()
for i_episode in range(20):
observation = env.reset()
for t in range(100):
env.render()
action = env.action_space.sample()
observation_, reward, done, info = env.step(action)
x, x_dot, theta, theta_dot = observation_
r1 = (env.x_threshold - abs(x)) / env.x_threshold - 0.8
r2 = (env.theta_threshold_radians - abs(theta)) / env.theta_threshold_radians - 0.5
reward = r1 + r2
print(reward)
ransition = np.hstack((observation, [action, reward], observation_))
print(ransition)
if done:
print("Episode finished after {} timesteps".format(t + 1))
break
env.close()
# Create checkpoint callback