stepcount = 0
state = env.reset()
done = False
current_rewards = 0
current_time_learned = []
while not done:
action = agent.choose_action(state, i_episode, train=True)
next_state, reward, done, info = env.step(action)
if reward == 1: total_cleared += 1
agent.store_transition(state, action, reward, next_state, done)
state = next_state
time_start = datetime.now()
agent.learn(BATCH_SIZE, i_episode)
time_difference = datetime.now() - time_start
current_time_learned.append(time_difference.total_seconds())
stepcount += 1
current_rewards += reward
if stepcount % 10000 == 0:
print("At step", stepcount)
if USE_TB:
writer.add_scalar("reward", current_rewards, i_episode)
writer.add_scalar("steps done", stepcount, i_episode)
writer.add_scalar("epsilon", agent.get_eps(i_episode), i_episode)
writer.add_scalar("Total cleared", total_cleared, i_episode)
rewards.append(current_rewards)
for i_episode in range(NUM_EPISODES):
stepcount = 0
state = env.reset()
done = False
current_rewards = 0
current_time_learned = []
while not done:
action = agent.choose_action(state, i_episode, train=True)
next_state, reward, done, info = env.step(action)
agent.store_transition(state, action, reward, next_state, done)
time_start = datetime.now()
agent.learn(BATCH_SIZE, i_episode, use_DDQN=False)
state = next_state
time_difference = datetime.now() - time_start
current_time_learned.append(time_difference.total_seconds())
stepcount += 1
current_rewards += reward
if USE_TB:
writer.add_scalar("reward", current_rewards, i_episode)
writer.add_scalar("steps done", stepcount, i_episode)
writer.add_scalar("epsilon", agent.get_eps(i_episode), i_episode)
rewards.append(current_rewards)
time_learned.append(sum(current_time_learned) / len(current_time_learned))
if (i_episode % 50 == 0 and i_episode != 0):