wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-08-05 13:21:03 +08:00
Code Issues Packages Projects Releases Wiki Activity

[rllib] Native support for Dict and Tuple spaces; fix Tuple action spaces; add prev a, r to LSTM (#3051)

Browse Source
This commit is contained in:
Eric Liang
2018-10-20 15:21:22 -07:00
committed by GitHub
parent 9a2b5333ef
commit 59901a88a0
39 changed files with 921 additions and 185 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/ray/rllib/test/test_supported_spaces.py
+30 -23
View File
@@ -2,7 +2,7 @@ import unittest
import traceback
import gym
from gym.spaces import Box, Discrete, Tuple
from gym.spaces import Box, Discrete, Tuple, Dict
from gym.envs.registration import EnvSpec
import numpy as np
import sys
@@ -14,33 +14,28 @@ from ray.tune.registry import register_env
ACTION_SPACES_TO_TEST = {
"discrete": Discrete(5),
"vector": Box(0.0, 1.0, (5, ), dtype=np.float32),
"simple_tuple": Tuple([
Box(0.0, 1.0, (5, ), dtype=np.float32),
Box(0.0, 1.0, (5, ), dtype=np.float32)
]),
"mixed_tuple": Tuple(
"vector": Box(-1.0, 1.0, (5, ), dtype=np.float32),
"tuple": Tuple(
[Discrete(2),
Discrete(3),
Box(0.0, 1.0, (5, ), dtype=np.float32)]),
Box(-1.0, 1.0, (5, ), dtype=np.float32)]),
}
OBSERVATION_SPACES_TO_TEST = {
"discrete": Discrete(5),
"vector": Box(0.0, 1.0, (5, ), dtype=np.float32),
"image": Box(0.0, 1.0, (84, 84, 1), dtype=np.float32),
"atari": Box(0.0, 1.0, (210, 160, 3), dtype=np.float32),
"atari_ram": Box(0.0, 1.0, (128, ), dtype=np.float32),
"simple_tuple": Tuple([
Box(0.0, 1.0, (5, ), dtype=np.float32),
Box(0.0, 1.0, (5, ), dtype=np.float32)
]),
"mixed_tuple": Tuple(
[Discrete(10), Box(0.0, 1.0, (5, ), dtype=np.float32)]),
"vector": Box(-1.0, 1.0, (5, ), dtype=np.float32),
"image": Box(-1.0, 1.0, (84, 84, 1), dtype=np.float32),
"atari": Box(-1.0, 1.0, (210, 160, 3), dtype=np.float32),
"tuple": Tuple([Discrete(10),
Box(-1.0, 1.0, (5, ), dtype=np.float32)]),
"dict": Dict({
"task": Discrete(10),
"position": Box(-1.0, 1.0, (5, ), dtype=np.float32),
}),
}
def make_stub_env(action_space, obs_space):
def make_stub_env(action_space, obs_space, check_action_bounds):
class StubEnv(gym.Env):
def __init__(self):
self.action_space = action_space
@@ -52,16 +47,23 @@ def make_stub_env(action_space, obs_space):
return sample
def step(self, action):
if check_action_bounds and not self.action_space.contains(action):
raise ValueError("Illegal action for {}: {}".format(
self.action_space, action))
if (isinstance(self.action_space, Tuple)
and len(action) != len(self.action_space.spaces)):
raise ValueError("Illegal action for {}: {}".format(
self.action_space, action))
return self.observation_space.sample(), 1, True, {}
return StubEnv
def check_support(alg, config, stats):
def check_support(alg, config, stats, check_bounds=False):
for a_name, action_space in ACTION_SPACES_TO_TEST.items():
for o_name, obs_space in OBSERVATION_SPACES_TO_TEST.items():
print("=== Testing", alg, action_space, obs_space, "===")
stub_env = make_stub_env(action_space, obs_space)
stub_env = make_stub_env(action_space, obs_space, check_bounds)
register_env("stub_env", lambda c: stub_env())
stat = "ok"
a = None
@@ -105,8 +107,13 @@ class ModelSupportedSpaces(unittest.TestCase):
"num_sgd_iter": 1,
"train_batch_size": 10,
"sample_batch_size": 10,
"sgd_minibatch_size": 1
}, stats)
"sgd_minibatch_size": 1,
"model": {
"squash_to_range": True
},
},
stats,
check_bounds=True)
check_support(
"ES", {
"num_workers": 1,