def run_me(parameter):
done = False
reward_sum = 0.
s = env.reset()
env.render('rgb_array')
a = np.array([0.0, 0.0, 0.0])
counter = 0
for _ in range(100):
obs = state_to_1_batch_tensor(s)
_, _, _, z = V(obs)
a = torch.from_numpy(a).float()
a = torch.reshape(a, (1, 1, 3))
z = torch.reshape(z, (1, 1, 32))
_, _, _, (h, c) = M(z, a)
z = z.detach().numpy()
h = h.detach().numpy()
z = z.reshape((1, 32))
h = h.reshape((1, 256))
# print(type(z))
# print(type(h))
# print(len(h))
# # print(h.size())
# print(z.shape)
# print(h.shape)
# z = z.squeez()
a = get_a(z, h, parameter)
# print(a)
s, reward, done, _ = env.step(a)
# env.render()
reward_sum += reward
counter += 1
print(counter)
print(counter)
return reward_sum
def multiple_runs(v, on):
env = CarRacing()
z_set = []
action_set = []
for run in range(MAX_RUNS):
zs = []
actions = []
state = env.reset()
env.render() # must have!
# done = False
counter = 0
for game_time in range(MAX_GAME_TIME):
# env.render()
action = generate_action()
obs = state_to_1_batch_tensor(state)
_, _, _, z = v(obs)
z = z.detach().numpy()
z = z.reshape(32)
# print(z.shape)
# if game_time == 5:
# plt.imshow(state)
# plt.show()
# state = _process_frame(state)
# plt.imshow(state)
# plt.show()
zs.append(z)
actions.append(action)
state, r, done, _ = env.step(action)
# print(r)
print(
'RUN:{},GT:{},DATA:{}'.format(
run, game_time, len(actions)
)
)
# if counter == REST_NUM:
#
# position = np.random.randint(len(env.track))
# env.car = Car(env.world, *env.track[position][1:4])
# counter = 0
# counter += 1
zs = np.array(zs, dtype=np.float16)
# print(zs.shape)
actions = np.array(actions, dtype=np.float16)
# print(actions.shape)
# np.save(dst + '/' + save_name, frame_and_action)
# np.savez_compressed(dst + '/' + save_name, action=actions, z=zs)
z_set.append(zs)
action_set.append(actions)
z_set = np.array(z_set)
# print(z_set.shape)
action_set = np.array(action_set)
# print(action_set.shape)
save_name = name_this + '_{}.npz'.format(on)
np.savez_compressed(dst + '/' + save_name, action=action_set, z=z_set)
if k == key.LEFT and a[0] == -1.0: a[0] = 0
if k == key.RIGHT and a[0] == +1.0: a[0] = 0
if k == key.UP: a[1] = 0
if k == key.DOWN: a[2] = 0
env = CarRacing()
env.render()
env.viewer.window.on_key_press = key_press
env.viewer.window.on_key_release = key_release
while True:
env.reset()
total_reward = 0.0
steps = 0
restart = False
env.render()
while True:
s, r, done, info = env.step(a)
total_reward += r
if steps % 200 == 0 or done:
# print("\naction " + str(["{:+0.2f}".format(x) for x in a]))
# print("step {} total_reward {:+0.2f}".format(steps, total_reward))
pass
steps += 1
env.render()
obs = state_to_1_batch_tensor(s)
rec, _, _, _ = V(obs)
img = rec.detach().numpy()
img = one_batch_tensor_to_img(img)
show_state(img)
if done or restart: break
env.monitor.close()
