]
pick_cells = [PickUpCell(2, 0), PickUpCell(0, 2)]
agent = Agent(4, 0)
# while ai has not reached a terminal state
session_reward = 0
while True:
current_state = get_state(agent, drop_cells, pick_cells)
if step < 500:
action = pRandom()
else:
action = pExploit(current_state, q_table)
step += 1
reward, new_state = perform_action(agent, action, drop_cells,
pick_cells)
q_table[current_state][action] = calculate_new_q(
LEARNING_RATE, DISCOUNT, reward, action, current_state, new_state,
q_table)
session_reward += reward
# visualization removed
# all drop off locations are filled
if len(list(
filter(lambda cell: cell.has_space() == False,
drop_cells))) == len(drop_cells) or step == HM_STEPS:
if step != HM_STEPS:
terminated += 1
break
all_rewards.append(session_reward)
def experiment_2(LEARNING_RATE, DISCOUNT, HM_STEPS):
q_table = initialize_q_table()
step = 0
all_rewards = []
while step < HM_STEPS:
# initial environment state
drop_cells = [
DropOffCell(0, 0),
DropOffCell(0, 4),
DropOffCell(2, 2),
DropOffCell(4, 4)
]
pick_cells = [PickUpCell(2, 4), PickUpCell(3, 1)]
agent = Agent(4, 0)
# while ai has not reached a terminal state
session_reward = 0
while True:
current_state = get_state(agent, drop_cells, pick_cells)
if step < 500:
action = pRandom()
else:
action = pExploit(current_state, q_table)
step += 1
reward, new_state = perform_action(agent, action, drop_cells,
pick_cells)
next_action = pExploit(new_state, q_table)
q_table[current_state][action] = sarsa_calculate_new_q(
LEARNING_RATE, DISCOUNT, reward, action, next_action,
current_state, new_state, q_table)
session_reward += reward
# visualization
env = create_display_environment(drop_cells, pick_cells, agent)
img = Image.fromarray(env, "RGB")
cv2.imshow("", np.array(img))
if (reward == PICK_UP_REWARD or reward == DROP_OFF_REWARD):
if cv2.waitKey(WAIT_TIME) & 0xFF == ord("q"):
break
else:
if cv2.waitKey(WAIT_TIME) & 0xFF == ord("q"):
break
# all drop off locations are filled
if len(
list(
filter(lambda cell: cell.has_space() == False,
drop_cells))) == len(
drop_cells) or step == HM_STEPS:
break
all_rewards.append(session_reward)
plt.plot([i for i in range(len(all_rewards))], all_rewards)
plt.ylabel("Reward Collected")
plt.xlabel("Session")
plt.show()
print(f"max reward reached: {max(all_rewards)}")
with open(f"qtable-experiment-2.pickle", "wb") as f:
pickle.dump(q_table, f)