bar = np.ones(
(5, 3), dtype=np.uint8) * np.array([255, 255, 0], dtype=np.uint8)
# For benchmark
st = time.time()
env = gym.make('mouseCl-v0')
o = env.reset()
if args.load:
player = Player(env.observation_space, env.action_space, args.load,
args.log_name, cur_loop * total_steps, cur_r, buf_full,
load_buffer, buf_count)
else:
player = Player(env.observation_space, env.action_space)
if not args.vm:
env.render()
for step in trange(total_steps, ncols=50):
action = player.act(o)
o, r, d, i = env.step(action)
player.step(o, r, d, i)
if d:
o = env.reset()
if not args.vm:
env.render()
next_save = player.save_model()
if not args.load:
save_dir = player.save_dir
else:
save_dir, _ = os.path.split(args.load)
next_dir = os.path.join(save_dir, str(next_save))
score = player.evaluate(gym.make('mouseCl-v0'), vid_type)
print('eval_score:{0}'.format(score))
log_name= args.log_name
)
else :
player = Player(
observation_space= env.observation_space,
action_space= env.action_space,
model_f= model_f,
tqdm= my_tqdm,
)
if args.render :
env.render()
if args.profile:
# Warm up
for step in range(hp.Learn_start+20):
action = player.act(bef_o)
aft_o,r,d,i = env.step(action)
player.step(bef_o,action,r,d,i)
if d :
bef_o = env.reset()
else:
bef_o = aft_o
if args.render :
env.render()
with Profile(f'log/{args.log_name}'):
for step in range(5):
action = player.act(bef_o)
aft_o,r,d,i = env.step(action)
player.step(bef_o,action,r,d,i)
if d :
# action = 2
# elif step%5 == 1 :
# action = 1
# elif step%5 == 2 :
# action = 2
# elif step%5 == 3 :
# action = 1
# elif step%5 == 4 :
# action = 0
# o, r, d, i = test_env.step(action)
# player.step(action, r,d,i)
# if d :
# o = test_env.reset()
if args.profile:
for step in trange(hp.Learn_start + 50, ncols=100):
action = player.act(o, training=True)
o, r, d, i = test_env.step(action)
player.step(action, r, d, i)
if d:
o = test_env.reset()
with tf.profiler.experimental.Profile('log/profile'):
for step in trange(5, ncols=100):
action = player.act(o, training=True)
o, r, d, i = test_env.step(action)
player.step(action, r, d, i)
if d:
o = test_env.reset()
else:
for step in trange(10000, ncols=100):
action = player.act(o, training=True)