def play():
current_position_index = random.randint(0, len(s) - 1)
current_position = s[current_position_index][:]
global Q
global total_time_steps
total_rewards = 0
while current_position != [10, 10]:
total_time_steps += 1
current_position_index = state_get_index(current_position[:])
optimal_action = action_get_index(current_position_index, Q)
next_action = optimal_action
next_position = probability_getter(current_position[:], next_action,
p1, p2)
next_position_index = state_get_index(next_position[:])
if next_position == [10, 10]:
total_rewards = 500
else:
total_rewards = -1
Q[state_get_index(current_position)][next_action] = Q[state_get_index(
current_position)][:][next_action] + ALPHA * (
total_rewards + GAMMA *
(Q[state_get_index(next_position)][:][action_get_index(
next_position_index, Q)] -
Q[state_get_index(current_position)][:][next_action]))
current_position = next_position
if current_position == [10, 10]:
break
def play(): global total_time_steps current_position_index = random.randint(0,len(s)-1) current_position = s[current_position_index][:] optimal_action = action_get_index(current_position_index, Q) total_rewards = 0 while current_position != [10, 10]: total_time_steps +=1 randomizer = random.random() if randomizer <= EPSILON: next_action = random.randint(0, 3) while next_action is optimal_action: next_action = random.randint(0, 3) else: next_action = optimal_action next_position = probability_getter(current_position[:], next_action, p1, p2) next_position_index = state_get_index(next_position[:]) if next_position == [10, 10]: total_rewards = 500 else: total_rewards = -1 Q[state_get_index(current_position)][next_action] = Q[state_get_index(current_position)][:][next_action] + ALPHA*(total_rewards + GAMMA*(Q[state_get_index(next_position)][:][action_get_index(next_position_index, Q)]-Q[state_get_index(current_position)][:][next_action])) current_position = next_position current_position_index = next_position_index optimal_action = action_get_index(current_position_index, Q)
def play(): current_position_index = random.randint(0,len(s)-1) current_position = s[current_position_index][:] optimal_action = action_get_index(current_position_index, Q) global total_time_steps total_rewards = 0 while current_position != [10, 10]: total_time_steps += 1 randomizer = random.random() if randomizer <= EPSILON: next_action = random.randint(0, 3) while next_action is optimal_action: next_action = random.randint(0, 3) else: next_action = optimal_action next_position = probability_getter(current_position[:], next_action, p1, p2) next_position_index = state_get_index(next_position[:]) if next_position == [10, 10]: total_rewards = 500 else: total_rewards = -1 V = 0 test_action = action_get_index(next_position_index, Q[:]) for x in range(0, len(a)): if test_action != Q[next_position_index][:][x]: V += (EPSILON) * Q[next_position_index][:][x] else: V += (1-EPSILON) * action_get_index(next_position_index, Q[:]) Q[state_get_index(current_position)][next_action] = Q[state_get_index(current_position)][:][next_action] + ALPHA*(total_rewards + ((GAMMA*V)-Q[state_get_index(current_position)][:][next_action])) current_position = next_position current_position_index = next_position_index optimal_action = action_get_index(current_position_index, Q)
def play(current_position, action_according_to_policy, set_of_states,
set_of_actions):
global total_time_steps
set_of_states = []
set_of_actions = []
set_of_states.append(current_position[:])
set_of_actions.append(action_according_to_policy)
while current_position != [10, 10]:
total_time_steps += 1
current_position = probability_getter(current_position,
action_according_to_policy, p1,
p2)
current_position_index = state_get_index(current_position[:])
set_of_states.append(current_position[:])
randomize_action = random.random()
if randomize_action <= (1 - EPSILON + (EPSILON / 4)):
next_action = POL[current_position_index]
else:
next_action = random.randint(0, 3)
while next_action is POL[current_position_index]:
next_action = random.randint(0, 3)
POL[current_position_index] = next_action
action_according_to_policy = POL[current_position_index]
set_of_actions.append(action_according_to_policy)
update_values(set_of_states, set_of_actions)