def test(self):
"""Test."""
def policy_fn():
return Policy(TicTacToeNet())
a0 = partial(AlphaZero, game=TicTacToe(), policy=policy_fn())
train('test', a0, maxt=100000, eval=True, eval_period=1000)
def test_feed_forward_ppo(self):
"""Test feed forward ppo."""
def env_fn(nenv):
return make_atari_env('Pong', nenv, frame_stack=4)
def policy_fn(env):
return Policy(NatureDQN(env.observation_space,
env.action_space))
ppo = partial(PPO, env_fn=env_fn, policy_fn=policy_fn)
train('test', ppo, maxt=1000, eval=True, eval_period=1000)
shutil.rmtree('test')
def test_sac(self):
"""Test."""
ddpg = partial(DDPG,
env_fn=env_fn,
policy_fn=policy_fn,
qf_fn=qf_fn,
learning_starts=300,
eval_num_episodes=1,
buffer_size=500)
train('logs', ddpg, maxt=1000, eval=False, eval_period=1000)
alg = ddpg('logs')
assert alg.load() == 1000
shutil.rmtree('logs')
def test_sac(self):
"""Test."""
sac = partial(SAC,
env_fn=env_fn,
policy_fn=policy_fn,
qf_fn=qf_fn,
learning_starts=300,
eval_num_episodes=1,
buffer_size=500,
target_update_period=100)
train('logs', sac, maxt=1000, eval=False, eval_period=1000)
alg = sac('logs')
alg.load()
shutil.rmtree('logs')
def test_ddpg(self):
"""Test."""
td3 = partial(TD3,
env_fn=env_fn,
policy_fn=policy_fn,
qf_fn=qf_fn,
learning_starts=300,
eval_num_episodes=1,
buffer_size=500,
policy_update_period=2)
train('logs', td3, maxt=1000, eval=False, eval_period=1000)
alg = td3('logs')
assert alg.load() == 1000
shutil.rmtree('logs')
def test_feed_forward_ppo2(self):
"""Test feed forward ppo2."""
def env_fn(nenv):
return make_env('LunarLanderContinuous-v2', nenv)
def policy_fn(env):
return Policy(PolicyNet(env.observation_space,
env.action_space))
def vf_fn(env):
return ValueFunction(VFNet(env.observation_space,
env.action_space))
ppo = partial(ResidualPPO2, env_fn=env_fn, policy_fn=policy_fn,
value_fn=vf_fn)
train('test', ppo, maxt=1000, eval=True, eval_period=1000)
shutil.rmtree('test')
def test_feed_forward_ppo2(self):
"""Test feed forward ppo2."""
def env_fn(nenv):
return make_atari_env('Pong', nenv, frame_stack=4)
def policy_fn(env):
return Policy(NatureDQN(env.observation_space,
env.action_space))
def vf_fn(env):
return ValueFunction(NatureDQNVF(env.observation_space,
env.action_space))
ppo = partial(PPO2RND, env_fn=env_fn, policy_fn=policy_fn,
value_fn=vf_fn, rnd_net=RNDNet)
train('test', ppo, maxt=100, eval=True, eval_period=100)
shutil.rmtree('test')
def test_feed_forward_ppo(self):
"""Test feed forward ppo."""
def env_fn(nenv):
return make_env(env_id="LunarLanderRandomConstrained-v2",
nenv=nenv)
def policy_fn(env):
return Policy(
FeedForwardActorCriticBase(env.observation_space,
env.action_space))
ppo = partial(ConstrainedResidualPPO,
env_fn=env_fn,
nenv=32,
policy_fn=policy_fn,
base_actor_cls=RandomActor,
policy_training_start=500)
train('test', ppo, maxt=1000, eval=True, eval_period=1000)
alg = ppo('test')
alg.load()
shutil.rmtree('test')
def test_ql(self):
"""Test."""
def env_fn(nenv):
return make_atari_env('Pong', nenv, frame_stack=1)
def qf_fn(env):
return QFunction(
NatureDQN(env.observation_space, env.action_space))
ql = partial(DQN,
env_fn=env_fn,
qf_fn=qf_fn,
learning_starts=100,
buffer_size=200,
update_period=4,
frame_stack=4,
exploration_timesteps=500,
target_update_period=100)
train('logs', ql, maxt=1000, eval=True, eval_period=1000)
alg = ql('logs')
alg.load()
assert np.allclose(alg.eps_schedule.value(alg.t), 0.1)
shutil.rmtree('logs')
import argparse
import dl
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Train Agent.')
parser.add_argument('--expdir', type=str, help='expdir', required=True)
parser.add_argument('--gin_config',
type=str,
help='gin config',
required=True)
parser.add_argument('-b',
'--gin_bindings',
nargs='+',
help='gin bindings to overwrite config')
args = parser.parse_args()
dl.load_config(args.gin_config, args.gin_bindings)
dl.train(args.expdir)
import argparse
import dl
import yaml
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Train Agent.')
parser.add_argument('config', type=str, help='config')
args = parser.parse_args()
with open(args.config, 'r') as f:
config = yaml.load(f)
gin_bindings = []
for k, v in config['gin_bindings'].items():
if isinstance(v, str) and v[0] != '@':
gin_bindings.append(f'{k}="{v}"')
else:
gin_bindings.append(f"{k}={v}")
dl.load_config(config['base_config'], gin_bindings)
dl.train(config['logdir'])
"""Main script for training models."""
import dl
import argparse
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Train Script.')
parser.add_argument('logdir', type=str, help='logdir')
parser.add_argument('config', type=str, help='gin config')
parser.add_argument('-b',
'--bindings',
nargs='+',
help='gin bindings to overwrite config')
args = parser.parse_args()
dl.load_config(args.config, args.bindings)
dl.train(args.logdir)
