def test():
env = envs.make(args.env,
'atari',
render=bool(args.render),
record=bool(args.record),
directory=args.log_dir)
learning_prop = json.load(
open(os.path.join(args.log_dir, '../learning_prop.json'), 'r'))
act_params = {
'scope':
"seed_%d" % learning_prop['seed'] + "/" + learning_prop['scope'],
'eps': args.test_eps
}
act = deepq.load(os.path.join(args.log_dir, args.log_fname), act_params)
episode_rew = 0
t = 0
while True:
obs, done = env.reset(), False
while (not done):
if args.render:
env.render()
time.sleep(0.05)
obs, rew, done, info = env.step(act(obs[None])[0])
# Reset only the enviornment but not the recorder
if args.record and done:
obs, done = env.env.reset(), False
episode_rew += rew
t += 1
if info['ale.lives'] == 0:
print("Episode reward %.2f after %d steps" % (episode_rew, t))
episode_rew = 0
t = 0
def test():
env = gym.make(args.env)
act = deepq.load(os.path.join(args.log_dir, args.log_fname))
if args.record:
env = Monitor(env, directory=args.log_dir)
while True:
obs, done = env.reset(), False
episode_rew = 0
while not done:
env.render(mode='test')
obs, rew, done, _ = env.step(act(obs[None])[0])
episode_rew += rew
print("Episode reward", episode_rew)
def test():
env = gym.make(args.env,
render=bool(args.render),
record=bool(args.record))
act = deepq.load(os.path.join(args.log_dir, args.log_fname))
while True:
obs, done = env.reset(), False
episode_rew = 0
while not done:
if args.render:
env.render()
obs, rew, done, _ = env.step(act(obs[None])[0])
episode_rew += rew
print("Episode reward", episode_rew)
def test():
from baselines0.deepq.utils import BatchInput
env = make_atari(args.env)
env = deepq.wrap_atari_dqn(env)
observation_space_shape = env.observation_space.shape
def make_obs_ph(name):
return BatchInput(observation_space_shape, name=name)
model = deepq.models.cnn_to_mlp(
convs=[(32, 8, 4), (64, 4, 2), (64, 3, 1)],
hiddens=[args.num_units] * args.num_layers,
dueling=bool(args.dueling),
)
act_params = {
'make_obs_ph': make_obs_ph,
'q_func': model,
'scope': args.scope
}
act = deepq.load(os.path.join(args.log_dir, args.log_fname), act_params)
if args.record:
env = Monitor(env, directory=args.log_dir)
episode_rew = 0
t = 0
while True:
obs, done = env.reset(), False
while not done:
if not (args.record):
env.render()
#time.sleep(0.01)
obs, rew, done, info = env.step(act(obs[None])[0])
episode_rew += rew
t += 1
if info['ale.lives'] == 0:
print("Episode reward %.2f after %d steps" % (episode_rew, t))
episode_rew = 0
t = 0
def test():
env = envs.make(args.env,
'classic_control',
render=bool(args.render),
record=bool(args.record),
directory=args.log_dir)
learning_prop = json.load(
open(os.path.join(args.log_dir, '../learning_prop.json'), 'r'))
act_params = {
'scope':
"seed_%d" % learning_prop['seed'] + "/" + learning_prop['scope'],
'eps': args.test_eps
}
act = deepq.load(os.path.join(args.log_dir, args.log_fname), act_params)
while True:
obs, done = env.reset(), False
episode_rew = 0
while not done:
if args.render:
env.render()
obs, rew, done, _ = env.step(act(obs[None])[0])
episode_rew += rew
print("Episode reward", episode_rew)