class agent():
def __init__(self, env, act_dim, state_dim, memory_capacity, epsilon,
update_target):
self.env = env
self.algo = Algorithm(0.0001, 0.99, act_dim, state_dim,
memory_capacity, epsilon, 64)
self.memory_capacity = memory_capacity
self.update_target = update_target
def learn(self, epoch):
reward_list = []
plt.ion()
fig, ax = plt.subplots()
for i in range(epoch):
state = self.env.reset()
ep_reward = 0
while True:
self.env.render()
action = self.algo.choose_action(state)
next_state, reward, done, _ = self.env.step(action)
ep_reward += reward
self.algo.store_transition(state, action, reward, next_state,
done)
if (self.algo.memory_counter >= self.memory_capacity):
self.algo.learn()
if epoch % self.update_target == 0:
self.algo.sync_target()
if done:
print("episode: {} , the episode reward is {}".format(
i, round(ep_reward, 3)))
if done:
break
state = next_state
reward_list.append(ep_reward)
ax.set_xlim(0, epoch)
ax.plot(reward_list, 'g-', label='total_loss')
plt.pause(0.001)
self.env.close()
