render_demo = False
imax = 5000
period_print = 100
eval_periods = 100
print("\nStarting")
total_reward = 0
period_rewards = deque(maxlen=2 * period_print)
best_avg_reward = float('-inf')
procStartTime = time.time()
for i in range(1, imax + 1):
doEpisode(experiment)
processLastReward(task, agent) ## store final reward for learner
agent.learn()
reward = task.getCumulativeReward()
total_reward += reward
period_rewards.append(reward)
avg_reward = np.mean(period_rewards)
if avg_reward > best_avg_reward:
best_avg_reward = avg_reward
if i % period_print == 0:
print(
"Episode ended: %i/%i period reward: %f total reward: %d best avg reward: %f rate: %f"
% (i, imax, avg_reward, total_reward, best_avg_reward,
total_reward / i))
render_demo = False
render_steps = False
imax = 7000
period_print = 100
eval_periods = 100
print("\nStarting")
total_reward = 0
period_reward = 0
procStartTime = time.time()
for i in range(1, imax + 1):
doEpisode(experiment, render_steps)
reward = task.getCumulativeReward()
total_reward += reward
period_reward += reward
processLastReward(task, agent) ## store final reward for learner
agent.learn()
if i % period_print == 0:
epsil = explorer.epsilon
print(
"Episode ended: %i/%i period reward: %f total reward: %d rate: %f epsilon: %f"
% (i, imax, period_reward / period_print, total_reward,
total_reward / i, epsil))
period_reward = 0
def doEpisode(self, demonstrate=False):
doEpisode(self.exp, demonstrate)