wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-07-28 11:25:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

[rllib] [tune] Custom preprocessors and models, various fixes (#1372)

Browse Source
This commit is contained in:
Eric Liang
2017-12-28 13:19:04 -08:00
committed by Richard Liaw
parent 3d224c4edf
commit 22c7c87e14
28 changed files with 296 additions and 329 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/ray/rllib/es/es.py
+9 -9
View File
@@ -63,7 +63,7 @@ class SharedNoiseTable(object):
@ray.remote
class Worker(object):
def __init__(self, config, policy_params, env_creator, noise,
def __init__(self, registry, config, policy_params, env_creator, noise,
min_task_runtime=0.2):
self.min_task_runtime = min_task_runtime
self.config = config
@@ -71,13 +71,12 @@ class Worker(object):
self.noise = SharedNoiseTable(noise)
self.env = env_creator()
self.preprocessor = ModelCatalog.get_preprocessor(self.env)
self.preprocessor = ModelCatalog.get_preprocessor(registry, self.env)
self.sess = utils.make_session(single_threaded=True)
self.policy = policies.GenericPolicy(self.sess, self.env.action_space,
self.preprocessor,
config["observation_filter"],
**policy_params)
self.policy = policies.GenericPolicy(
registry, self.sess, self.env.action_space, self.preprocessor,
config["observation_filter"], **policy_params)
def rollout(self, timestep_limit, add_noise=True):
rollout_rewards, rollout_length = policies.rollout(
@@ -143,11 +142,11 @@ class ESAgent(Agent):
}
env = self.env_creator()
preprocessor = ModelCatalog.get_preprocessor(env)
preprocessor = ModelCatalog.get_preprocessor(self.registry, env)
self.sess = utils.make_session(single_threaded=False)
self.policy = policies.GenericPolicy(
self.sess, env.action_space, preprocessor,
self.registry, self.sess, env.action_space, preprocessor,
self.config["observation_filter"], **policy_params)
self.optimizer = optimizers.Adam(self.policy, self.config["stepsize"])
@@ -160,7 +159,8 @@ class ESAgent(Agent):
print("Creating actors.")
self.workers = [
Worker.remote(
self.config, policy_params, self.env_creator, noise_id)
self.registry, self.config, policy_params, self.env_creator,
noise_id)
for _ in range(self.config["num_workers"])]
self.episodes_so_far = 0
python/ray/rllib/es/policies.py
+2 -2
View File
@@ -39,7 +39,7 @@ def rollout(policy, env, timestep_limit=None, add_noise=False):
class GenericPolicy(object):
def __init__(self, sess, action_space, preprocessor,
def __init__(self, registry, sess, action_space, preprocessor,
observation_filter, action_noise_std):
self.sess = sess
self.action_space = action_space
@@ -53,7 +53,7 @@ class GenericPolicy(object):
# Policy network.
dist_class, dist_dim = ModelCatalog.get_action_dist(
self.action_space, dist_type="deterministic")
model = ModelCatalog.get_model(self.inputs, dist_dim)
model = ModelCatalog.get_model(registry, self.inputs, dist_dim)
dist = dist_class(model.outputs)
self.sampler = dist.sample()