gaspersic/baselines
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

make tests use single-threaded session for determinism of KfacOptimizer (#298)

Browse Source 
* make tests use single-threaded session for determinism of KfacOptimizer

* updated comment in kfac.py

* remove unused sess_config
This commit is contained in:
pzhokhov
2019-03-22 16:28:29 -07:00
committed by Jacob Hilton
parent cc88c8e4c0
commit a08af5d07d
3 changed files with 10 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
baselines/acktr/acktr.py
View File

@@ -90,7 +90,7 @@ class Model(object):
self.initial_state = step_model.initial_state self.initial_state = step_model.initial_state
tf.global_variables_initializer().run(session=sess) tf.global_variables_initializer().run(session=sess)
def learn(network, env, seed, total_timesteps=int(40e6), gamma=0.99, log_interval=1, nprocs=32, nsteps=20, def learn(network, env, seed, total_timesteps=int(40e6), gamma=0.99, log_interval=100, nprocs=32, nsteps=20,
ent_coef=0.01, vf_coef=0.5, vf_fisher_coef=1.0, lr=0.25, max_grad_norm=0.5, ent_coef=0.01, vf_coef=0.5, vf_fisher_coef=1.0, lr=0.25, max_grad_norm=0.5,
kfac_clip=0.001, save_interval=None, lrschedule='linear', load_path=None, is_async=True, **network_kwargs): kfac_clip=0.001, save_interval=None, lrschedule='linear', load_path=None, is_async=True, **network_kwargs):
set_global_seeds(seed) set_global_seeds(seed)

2
baselines/acktr/kfac.py
View File

@@ -11,7 +11,7 @@ KFAC_DEBUG = False
class KfacOptimizer(): class KfacOptimizer():
# note that KfacOptimizer will be truly synchronous (and thus deterministic) only if a single-threaded session is used
def __init__(self, learning_rate=0.01, momentum=0.9, clip_kl=0.01, kfac_update=2, stats_accum_iter=60, full_stats_init=False, cold_iter=100, cold_lr=None, is_async=False, async_stats=False, epsilon=1e-2, stats_decay=0.95, blockdiag_bias=False, channel_fac=False, factored_damping=False, approxT2=False, use_float64=False, weight_decay_dict={},max_grad_norm=0.5): def __init__(self, learning_rate=0.01, momentum=0.9, clip_kl=0.01, kfac_update=2, stats_accum_iter=60, full_stats_init=False, cold_iter=100, cold_lr=None, is_async=False, async_stats=False, epsilon=1e-2, stats_decay=0.95, blockdiag_bias=False, channel_fac=False, factored_damping=False, approxT2=False, use_float64=False, weight_decay_dict={},max_grad_norm=0.5):
self.max_grad_norm = max_grad_norm self.max_grad_norm = max_grad_norm
self._lr = learning_rate self._lr = learning_rate

10
baselines/common/tests/util.py
View File

@@ -5,6 +5,12 @@ from baselines.common.vec_env.dummy_vec_env import DummyVecEnv
N_TRIALS = 10000 N_TRIALS = 10000
N_EPISODES = 100 N_EPISODES = 100
_sess_config = tf.ConfigProto(
allow_soft_placement=True,
intra_op_parallelism_threads=1,
inter_op_parallelism_threads=1
)
def simple_test(env_fn, learn_fn, min_reward_fraction, n_trials=N_TRIALS): def simple_test(env_fn, learn_fn, min_reward_fraction, n_trials=N_TRIALS):
def seeded_env_fn(): def seeded_env_fn():
env = env_fn() env = env_fn()
@@ -13,7 +19,7 @@ def simple_test(env_fn, learn_fn, min_reward_fraction, n_trials=N_TRIALS):
np.random.seed(0) np.random.seed(0)
env = DummyVecEnv([seeded_env_fn]) env = DummyVecEnv([seeded_env_fn])
with tf.Graph().as_default(), tf.Session(config=tf.ConfigProto(allow_soft_placement=True)).as_default(): with tf.Graph().as_default(), tf.Session(config=_sess_config).as_default():
tf.set_random_seed(0) tf.set_random_seed(0)
model = learn_fn(env) model = learn_fn(env)
sum_rew = 0 sum_rew = 0
@@ -34,7 +40,7 @@ def simple_test(env_fn, learn_fn, min_reward_fraction, n_trials=N_TRIALS):
def reward_per_episode_test(env_fn, learn_fn, min_avg_reward, n_trials=N_EPISODES): def reward_per_episode_test(env_fn, learn_fn, min_avg_reward, n_trials=N_EPISODES):
env = DummyVecEnv([env_fn]) env = DummyVecEnv([env_fn])
with tf.Graph().as_default(), tf.Session(config=tf.ConfigProto(allow_soft_placement=True)).as_default(): with tf.Graph().as_default(), tf.Session(config=_sess_config).as_default():
model = learn_fn(env) model = learn_fn(env)
N_TRIALS = 100 N_TRIALS = 100
observations, actions, rewards = rollout(env, model, N_TRIALS) observations, actions, rewards = rollout(env, model, N_TRIALS)