def simulate(env: Env, agent: Agent, monitor: Monitor, episodes: int) -> None:
for episode in range(episodes):
agent.episode_start()
done = False
obs = env.reset()
while not done:
action = agent.select_action(obs)
next_obs, reward, done, _ = env.step(action)
transition = Transition(obs, action, reward, next_obs, done)
agent.store_transition(transition)
monitor.store_transition(env, transition)
obs = next_obs
agent.episode_end()
monitor.episode_end(episode, episodes)
env.close()
monitor.simulation_end()
replace_target=10000)
if load_checkpoint:
agent.load_models()
scores = []
num_games = 10000
score = 0
print("Loading up the agent's memory with random driving")
while agent.mem_cntr < 5000:
done = False
observation = env._reset()
while not done:
action = np.random.choice(list(range(agent.n_actions)))
observation_, reward, done = env.step(agent.action_space[action])
agent.store_transition(observation, action, reward, observation_,
int(done))
observation = observation_
print("Done with random driving. Learning...")
history_file = 'tmp/' + experiment_name + '/ep_{}.pkl'
writer = csv.writer(open('tmp/' + experiment_name + 'scores.csv', 'w'),
delimiter=',')
writer.writerow(['episode', 'score', 'epsilon'])
for i in range(num_games):
done = False
if i % 10 == 0 and i > 0:
avg_score = np.mean(scores[max(0, i - 10):(i + 1)])
print('episode: ', i, 'score: ', score,
' average score %.3f' % avg_score,