from qlearner import QLearner
import pylab as plt
my_learner = QLearner()
my_learner.load_maze('/u/braun/tlab/QLearner/data/reward_4x4.npy',
'/u/braun/tlab/QLearner/data/meta_4x4.txt')
#print ("testing data load\n\n")
#my_learner.display_Q()
#my_learner.display_R()
print("begin training...")
reward = my_learner.train(0.7)
my_learner.display_Q()
my_learner.display_R()
steps = my_learner.test(7) # 7 foods in 4x4 maze
print("steps")
print(steps)
print("")
plt.hist(reward, 50, normed=1, facecolor='g', alpha=0.75)
plt.xlabel('Episodes required to reach 200')
plt.ylabel('Frequency')
plt.title('Histogram')
plt.show()
# Initialise environment and agent
wrapper = CartPoleWrapperDiscrete()
agent = QLearner(wrapper=wrapper, seed=run)
style.use('fivethirtyeight')
fig = plt.figure()
plt.axis([0, args.episodes, 0, 300])
plt.xlabel('Episodes')
plt.ylabel('AVG Reward last 50 episodes')
# For each episode, train the agent on the environment and record the
# reward of each episode
for episode in range(num_episodes):
rewards[episode] = agent.train()
if (episode % 50) == 0 and episode != 0:
avg_last = float(sum(rewards[episode - 50:episode])) / 50
plt.scatter(episode, avg_last)
plt.pause(0.05)
# Check if environment is solved
if wrapper.solved(rewards[:episode]):
end_episode = episode
break
# Record and print performance
runtime_per_run.append(timer() - start)
rewards_per_run['run' + str(run)] = rewards
if end_episode >= 99:
print('average reward of last 100 episodes of run', run, '=',
float(sum(rewards[-100:])) / 100)
