پیاده سازی cart pole با qlearning (46176)))
سه سال پیش منتشر شده
تعداد بازدید: 258
کد پروژه: 365119
شرح پروژه
#کد_پروژه: 46176
موضوع: پروژه پیاده سازی cart pole با qlearning/
شهبازی هستم در رابطه با پروژه دانشجویی پیاده سازی cart pole با qlearning
https://medium.com/swlh/using-q-learning-for-openais-cartpole-v1-4a216ef237df
for episode in range(EPISODES + 1): //go through the episodes
t0 = time.time() //set the initial time
discrete_state = get_discrete_state(env.reset()) //get the discrete start for the restarted environment
done = False
episode_reward = 0 //reward starts as 0 for each episode
if episode % 2000 == 0:
print("Episode: " + str(episode))
while not done:
if np.random.random() > epsilon:
action = np.argmax(q_table[discrete_state]) //take cordinated action
else:
action = np.random.randint(0, env.action_space.n) //do a random ation
new_state, reward, done, _ = env.step(action) //step action to get new states, reward, and the "done" status.
episode_reward += reward //add the reward
new_discrete_state = get_discrete_state(new_state)
if episode % 2000 == 0: //render
env.render()
if not done: //update q-table
max_future_q = np.max(q_table[new_discrete_state])
current_q = q_table[discrete_state + (action,)]
new_q = (1 - LEARNING_RATE) * current_q + LEARNING_RATE * (reward + DISCOUNT * max_future_q)
q_table[discrete_state + (action,)] = new_q
discrete_state = new_discrete_state
if epsilon > 0.05: //epsilon modification
if episode_reward > prior_reward and episode > 10000:
epsilon = math.pow(epsilon_decay_value, episode - 10000)
if episode % 500 == 0:
print("Epsilon: " + str(epsilon))
t1 = time.time() //episode has finished
episode_total = t1 - t0 //episode total time
total = total + episode_total
total_reward += episode_reward //episode total reward
prior_reward = episode_reward
if episode % 1000 == 0: //every 1000 episodes print the average time and the average reward
mean = total / 1000
print("Time Average: " + str(mean))
total = 0
mean_reward = total_reward / 1000
print("Mean Reward: " + str(mean_reward))
total_reward = 0
env.close()
این چیزیه مه تو این سایته زده
مهارت ها و تخصص های مورد نیاز
مهلت برای انجام
2روز
وضعیت مناقصه
بسته
درباره کارفرما
عضویت چهار سال پیش
