def test_grid_value_iteration():
pygame.init()
w = 6
h = 5
rewards = ((24, 1),
(2, -1),
(11, -1),
(27, -1))
terminal = [2, 11, 24]
new_pos = {"top": 2, "bot": 3, "left": 0, "right": 1}
S, A, T, P = create_grid_world(w, h, rewards, terminal)
start_time = time()
V, Pi = value_iteration(S, A, P, T)
print("--- %s seconds ---" % (time() - start_time))
win = pygame.display.set_mode((w * 100, h * 100))
for i in range(w * h):
if i % w == 0 and i != 0:
print("")
print(round(V[i], 7), end=" ")
print("")
for i in range(w * h):
if i % w == 0 and i != 0:
print("")
print(Pi[i], end=" ")
st = reset_grid(w, h)
while not is_terminal(st, T):
display_grid(win, w, h)
event_loop()
display_reward_grid(win, rewards, w, h)
display_mouse_grid(win, st, w, h)
sleep(1)
positions = {"top": st - w,
"bot": st + w,
"left": st - 1,
"right": st + 1
}
positions_bis = {key: Pi[st][new_pos[key]] for key, value in positions.items() if 0 <= value < w * h and Pi[st][new_pos[key]] > 0}
action = max(positions_bis, key=positions_bis.get)
a = new_pos[action]
st, r, term = step(st, a, T, S, P)
display_grid(win, w, h)
display_reward_grid(win, rewards, w, h)
display_mouse_grid(win, st, w, h)
def test_grid_q_learning():
pygame.init()
w = 6
h = 5
rewards = ((24, 1),
(2, -1),
(11, -1),
(27, -1))
terminal = [2, 11, 24]
new_pos = {"top": 2, "bot": 3, "left": 0, "right": 1}
S, A, T, P = create_grid_world(w, h, rewards, terminal)
start_time = time()
Q, Pi = tabular_q_learning_control(T, S, P, len(S), len(A), reset_line, is_terminal, step,
episodes_count=10000, max_steps_per_episode=100)
print("--- %s seconds ---" % (time() - start_time))
win = pygame.display.set_mode((w * 100, h * 100))
for i in range(w * h):
if i % w == 0 and i != 0:
print("")
print(Pi[i], end=" ")
st = reset_grid(w, h)
while not is_terminal(st, T):
display_grid(win, w, h)
event_loop()
display_reward_grid(win, rewards, w, h)
display_mouse_grid(win, st, w, h)
sleep(1)
positions = {"top": st - w,
"bot": st + w,
"left": st - 1,
"right": st + 1
}
a = np.argmax(Q[st])
st, r, term = step(st, a, T, S, P)
display_grid(win, w, h)
display_reward_grid(win, rewards, w, h)
display_mouse_grid(win, st, w, h)
def test_line_iterative_policy_evaluation():
pygame.init()
num_states = 15
rewards = ((0, -1),
(14, 1))
terminal = [0, 14]
S, A, T, P = create_line_world(num_states, rewards, terminal)
Pi = tabular_uniform_random_policy(S.shape[0], A.shape[0])
start_time = time()
V = iterative_policy_evaluation(S, A, P, T, Pi)
print("--- %s seconds ---" % (time() - start_time))
print(V)
win = pygame.display.set_mode((num_states * 100, 100))
st = reset_line(num_states)
while not is_terminal(st, T):
display_line(win, num_states)
event_loop()
display_reward_line(win, rewards, num_states)
display_mouse_line(win, st, num_states)
sleep(1)
if V[st + 1] > V[st - 1] or V[st + 1] == 0:
a = 1
elif V[st + 1] < V[st - 1] or V[st - 1] == 0:
a = 0
st, r, term = step(st, a, T, S, P)
display_line(win, num_states)
display_reward_line(win, rewards,num_states)
display_mouse_line(win, st, num_states)
sleep(1)
def test_line_sarsa():
pygame.init()
num_states = 15
rewards = ((0, -1),
(14, 1))
terminal = [0, 14]
S, A, T, P = create_line_world(num_states, rewards, terminal)
start_time = time()
Q, Pi = tabular_sarsa_control(T, S, P, len(S), len(A), reset_line, is_terminal, step,
episodes_count=10000, max_steps_per_episode=100)
print("--- %s seconds ---" % (time() - start_time))
for i in range(num_states):
print(Q[i], end=" ")
win = pygame.display.set_mode((num_states * 100, 100))
st = reset_line(num_states)
while not is_terminal(st, T):
display_line(win, num_states)
event_loop()
display_reward_line(win, rewards, num_states)
display_mouse_line(win, st, num_states)
sleep(1)
a = np.argmax(Q[st])
st, r, term = step(st, a, T, S, P)
display_line(win, num_states)
display_reward_line(win, rewards,num_states)
display_mouse_line(win, st, num_states)
sleep(1)