Merge branch '26-sequencingreplanning-feature-for-episode-based-environments'
Conflicts: fancy_gym/dmc/README.MD fancy_gym/dmc/manipulation/__init__.py fancy_gym/dmc/manipulation/reach_site/__init__.py fancy_gym/dmc/manipulation/reach_site/mp_wrapper.py fancy_gym/dmc/suite/__init__.py fancy_gym/dmc/suite/ball_in_cup/__init__.py fancy_gym/dmc/suite/ball_in_cup/mp_wrapper.py fancy_gym/dmc/suite/cartpole/__init__.py fancy_gym/dmc/suite/cartpole/mp_wrapper.py fancy_gym/dmc/suite/reacher/__init__.py fancy_gym/dmc/suite/reacher/mp_wrapper.py fancy_gym/envs/classic_control/README.MD fancy_gym/envs/classic_control/__init__.py fancy_gym/envs/classic_control/base_reacher/base_reacher.py fancy_gym/envs/classic_control/base_reacher/base_reacher_direct.py fancy_gym/envs/classic_control/base_reacher/base_reacher_torque.py fancy_gym/envs/classic_control/hole_reacher/__init__.py fancy_gym/envs/classic_control/hole_reacher/hole_reacher.py fancy_gym/envs/classic_control/hole_reacher/hr_dist_vel_acc_reward.py fancy_gym/envs/classic_control/hole_reacher/hr_simple_reward.py fancy_gym/envs/classic_control/hole_reacher/mp_wrapper.py fancy_gym/envs/classic_control/simple_reacher/__init__.py fancy_gym/envs/classic_control/simple_reacher/simple_reacher.py fancy_gym/envs/classic_control/utils.py fancy_gym/envs/classic_control/viapoint_reacher/__init__.py fancy_gym/envs/classic_control/viapoint_reacher/mp_wrapper.py fancy_gym/envs/classic_control/viapoint_reacher/viapoint_reacher.py fancy_gym/envs/mujoco/README.MD fancy_gym/envs/mujoco/beerpong/assets/beerpong.xml fancy_gym/envs/mujoco/beerpong/assets/beerpong_wo_cup.xml fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/base_link_convex.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/base_link_fine.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/bhand_finger_dist_link_convex.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/bhand_finger_dist_link_fine.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/bhand_finger_med_link_convex.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/bhand_finger_med_link_fine.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/bhand_finger_prox_link_convex_decomposition_p1.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/bhand_finger_prox_link_convex_decomposition_p2.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/bhand_finger_prox_link_convex_decomposition_p3.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/bhand_finger_prox_link_fine.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/bhand_palm_fine.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/bhand_palm_link_convex_decomposition_p1.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/bhand_palm_link_convex_decomposition_p2.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/bhand_palm_link_convex_decomposition_p3.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/bhand_palm_link_convex_decomposition_p4.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split1.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split10.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split11.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split12.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split13.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split14.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split15.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split16.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split17.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split18.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split2.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split3.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split4.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split5.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split6.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split7.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split8.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/cup_split9.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/elbow_link_convex.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/elbow_link_fine.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/forearm_link_convex_decomposition_p1.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/forearm_link_convex_decomposition_p2.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/forearm_link_fine.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/shoulder_link_convex_decomposition_p1.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/shoulder_link_convex_decomposition_p2.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/shoulder_link_convex_decomposition_p3.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/shoulder_link_fine.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/shoulder_pitch_link_convex.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/shoulder_pitch_link_fine.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/upper_arm_link_convex_decomposition_p1.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/upper_arm_link_convex_decomposition_p2.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/upper_arm_link_fine.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/wrist_palm_link_convex.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/wrist_palm_link_fine.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/wrist_pitch_link_convex_decomposition_p1.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/wrist_pitch_link_convex_decomposition_p2.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/wrist_pitch_link_convex_decomposition_p3.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/wrist_pitch_link_fine.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/wrist_yaw_link_convex_decomposition_p1.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/wrist_yaw_link_convex_decomposition_p2.stl fancy_gym/envs/mujoco/beerpong/assets/meshes/wam/wrist_yaw_link_fine.stl fancy_gym/envs/mujoco/hopper_throw/__init__.py fancy_gym/envs/mujoco/reacher/assets/reacher_5links.xml fancy_gym/envs/mujoco/reacher/assets/reacher_7links.xml fancy_gym/examples/examples_dmc.py fancy_gym/examples/examples_general.py fancy_gym/examples/examples_metaworld.py fancy_gym/meta/README.MD fancy_gym/meta/__init__.py fancy_gym/meta/goal_change_mp_wrapper.py fancy_gym/meta/goal_endeffector_change_mp_wrapper.py fancy_gym/meta/goal_object_change_mp_wrapper.py fancy_gym/meta/object_change_mp_wrapper.py fancy_gym/open_ai/README.MD fancy_gym/open_ai/deprecated_needs_gym_robotics/robotics/__init__.py fancy_gym/open_ai/deprecated_needs_gym_robotics/robotics/fetch/__init__.py fancy_gym/open_ai/deprecated_needs_gym_robotics/robotics/fetch/mp_wrapper.py fancy_gym/open_ai/mujoco/__init__.py fancy_gym/open_ai/mujoco/reacher_v2/__init__.py fancy_gym/utils/__init__.py test/test_dmc.py
This commit is contained in:
Fabian
@@ -0,0 +1,130 @@
|
||||
import unittest
|
||||
|
||||
import gym
|
||||
import numpy as np
|
||||
from dm_control import suite, manipulation
|
||||
|
||||
import fancy_gym
|
||||
from fancy_gym import make
|
||||
|
||||
SUITE_IDS = [f'dmc:{env}-{task}' for env, task in suite.ALL_TASKS if env != "lqr"]
|
||||
MANIPULATION_IDS = [f'dmc:manipulation-{task}' for task in manipulation.ALL if task.endswith('_features')]
|
||||
SEED = 1
|
||||
|
||||
|
||||
class TestDMCEnvironments(unittest.TestCase):
|
||||
|
||||
def _run_env(self, env_id, iterations=None, seed=SEED, render=False):
|
||||
"""
|
||||
Example for running a DMC based env in the step based setting.
|
||||
The env_id has to be specified as `dmc:domain_name-task_name` or
|
||||
for manipulation tasks as `manipulation-environment_name`
|
||||
|
||||
Args:
|
||||
env_id: Either `dmc:domain_name-task_name` or `dmc:manipulation-environment_name`
|
||||
iterations: Number of rollout steps to run
|
||||
seed: random seeding
|
||||
render: Render the episode
|
||||
|
||||
Returns: observations, rewards, dones, actions
|
||||
|
||||
"""
|
||||
env: gym.Env = make(env_id, seed=seed)
|
||||
rewards = []
|
||||
observations = []
|
||||
actions = []
|
||||
dones = []
|
||||
obs = env.reset()
|
||||
self._verify_observations(obs, env.observation_space, "reset()")
|
||||
|
||||
iterations = iterations or (env.spec.max_episode_steps or 1)
|
||||
|
||||
# number of samples(multiple environment steps)
|
||||
for i in range(iterations):
|
||||
observations.append(obs)
|
||||
|
||||
ac = env.action_space.sample()
|
||||
actions.append(ac)
|
||||
# ac = np.random.uniform(env.action_space.low, env.action_space.high, env.action_space.shape)
|
||||
obs, reward, done, info = env.step(ac)
|
||||
|
||||
self._verify_observations(obs, env.observation_space, "step()")
|
||||
self._verify_reward(reward)
|
||||
self._verify_done(done)
|
||||
|
||||
rewards.append(reward)
|
||||
dones.append(done)
|
||||
|
||||
if render:
|
||||
env.render("human")
|
||||
|
||||
if done:
|
||||
break
|
||||
|
||||
assert done, "Done flag is not True after end of episode."
|
||||
observations.append(obs)
|
||||
env.close()
|
||||
del env
|
||||
return np.array(observations), np.array(rewards), np.array(dones), np.array(actions)
|
||||
|
||||
def _run_env_determinism(self, ids):
|
||||
seed = 0
|
||||
for env_id in ids:
|
||||
with self.subTest(msg=env_id):
|
||||
traj1 = self._run_env(env_id, seed=seed)
|
||||
traj2 = self._run_env(env_id, seed=seed)
|
||||
for i, time_step in enumerate(zip(*traj1, *traj2)):
|
||||
obs1, rwd1, done1, ac1, obs2, rwd2, done2, ac2 = time_step
|
||||
self.assertTrue(np.array_equal(obs1, obs2), f"Observations [{i}] delta {obs1 - obs2} is not zero.")
|
||||
self.assertTrue(np.array_equal(ac1, ac2), f"Actions [{i}] delta {ac1 - ac2} is not zero.")
|
||||
self.assertEqual(done1, done2, f"Dones [{i}] {done1} and {done2} do not match.")
|
||||
self.assertEqual(rwd1, rwd2, f"Rewards [{i}] {rwd1} and {rwd2} do not match.")
|
||||
|
||||
def _verify_observations(self, obs, observation_space, obs_type="reset()"):
|
||||
self.assertTrue(observation_space.contains(obs),
|
||||
f"Observation {obs} received from {obs_type} "
|
||||
f"not contained in observation space {observation_space}.")
|
||||
|
||||
def _verify_reward(self, reward):
|
||||
self.assertIsInstance(reward, (float, int), f"Returned type {type(reward)} as reward, expected float or int.")
|
||||
|
||||
def _verify_done(self, done):
|
||||
self.assertIsInstance(done, bool, f"Returned {done} as done flag, expected bool.")
|
||||
|
||||
def test_suite_functionality(self):
|
||||
"""Tests that suite step environments run without errors using random actions."""
|
||||
for env_id in SUITE_IDS:
|
||||
with self.subTest(msg=env_id):
|
||||
self._run_env(env_id)
|
||||
|
||||
def test_suite_determinism(self):
|
||||
"""Tests that for step environments identical seeds produce identical trajectories."""
|
||||
self._run_env_determinism(SUITE_IDS)
|
||||
|
||||
def test_manipulation_functionality(self):
|
||||
"""Tests that manipulation step environments run without errors using random actions."""
|
||||
for env_id in MANIPULATION_IDS:
|
||||
with self.subTest(msg=env_id):
|
||||
self._run_env(env_id)
|
||||
|
||||
def test_manipulation_determinism(self):
|
||||
"""Tests that for step environments identical seeds produce identical trajectories."""
|
||||
self._run_env_determinism(MANIPULATION_IDS)
|
||||
|
||||
def test_bb_functionality(self):
|
||||
"""Tests that black box environments run without errors using random actions."""
|
||||
for traj_gen, env_ids in fancy_gym.ALL_DMC_MOVEMENT_PRIMITIVE_ENVIRONMENTS.items():
|
||||
with self.subTest(msg=traj_gen):
|
||||
for id in env_ids:
|
||||
with self.subTest(msg=id):
|
||||
self._run_env(id)
|
||||
|
||||
def test_bb_determinism(self):
|
||||
"""Tests that for black box environment identical seeds produce identical trajectories."""
|
||||
for traj_gen, env_ids in fancy_gym.ALL_DMC_MOVEMENT_PRIMITIVE_ENVIRONMENTS.items():
|
||||
with self.subTest(msg=traj_gen):
|
||||
self._run_env_determinism(env_ids)
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
unittest.main()
|
||||
@@ -1,49 +0,0 @@
|
||||
from typing import Tuple
|
||||
|
||||
import fancy_gym
|
||||
import pytest
|
||||
from dm_control import suite, manipulation
|
||||
|
||||
from test.utils import run_env_determinism, run_env
|
||||
|
||||
SUITE_IDS = [f'dmc:{env}-{task}' for env, task in suite.ALL_TASKS if env != "lqr"]
|
||||
MANIPULATION_IDS = [f'dmc:manipulation-{task}' for task in manipulation.ALL if task.endswith('_features')]
|
||||
SEED = 1
|
||||
|
||||
|
||||
@pytest.mark.parametrize('env_id', SUITE_IDS)
|
||||
def test_step_suite_functionality(env_id: str):
|
||||
"""Tests that suite step environments run without errors using random actions."""
|
||||
run_env(env_id)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('env_id', SUITE_IDS)
|
||||
def test_step_suite_determinism(env_id: str):
|
||||
"""Tests that for step environments identical seeds produce identical trajectories."""
|
||||
seed = 0
|
||||
run_env_determinism(env_id, seed)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('env_id', MANIPULATION_IDS)
|
||||
def test_step_manipulation_functionality(env_id: str):
|
||||
"""Tests that manipulation step environments run without errors using random actions."""
|
||||
run_env(env_id)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('env_id', MANIPULATION_IDS)
|
||||
def test_step_manipulation_determinism(env_id: str):
|
||||
"""Tests that for step environments identical seeds produce identical trajectories."""
|
||||
seed = 0
|
||||
run_env_determinism(env_id, seed)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('env_id', [fancy_gym.ALL_DMC_MOVEMENT_PRIMITIVE_ENVIRONMENTS.values()])
|
||||
def test_bb_dmc_functionality(env_id: str):
|
||||
"""Tests that black box environments run without errors using random actions."""
|
||||
run_env(env_id)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('env_id', [fancy_gym.ALL_DMC_MOVEMENT_PRIMITIVE_ENVIRONMENTS.values()])
|
||||
def test_bb_dmc_determinism(env_id: str):
|
||||
"""Tests that for black box environment identical seeds produce identical trajectories."""
|
||||
run_env_determinism(env_id)
|
||||
@@ -3,14 +3,15 @@ import unittest
|
||||
import gym
|
||||
import numpy as np
|
||||
|
||||
from alr_envs import make
|
||||
from metaworld.envs import ALL_V2_ENVIRONMENTS_GOAL_OBSERVABLE
|
||||
import fancy_gym # noqa
|
||||
from fancy_gym.utils.make_env_helpers import make
|
||||
|
||||
ALL_ENVS = [env.split("-goal-observable")[0] for env, _ in ALL_V2_ENVIRONMENTS_GOAL_OBSERVABLE.items()]
|
||||
CUSTOM_IDS = [spec.id for spec in gym.envs.registry.all() if
|
||||
"fancy_gym" in spec.entry_point and 'make_bb_env_helper' not in spec.entry_point]
|
||||
SEED = 1
|
||||
|
||||
|
||||
class TestStepMetaWorlEnvironments(unittest.TestCase):
|
||||
class TestCustomEnvironments(unittest.TestCase):
|
||||
|
||||
def _run_env(self, env_id, iterations=None, seed=SEED, render=False):
|
||||
"""
|
||||
@@ -21,26 +22,21 @@ class TestStepMetaWorlEnvironments(unittest.TestCase):
|
||||
Args:
|
||||
env_id: Either `domain_name-task_name` or `manipulation-environment_name`
|
||||
iterations: Number of rollout steps to run
|
||||
seed= random seeding
|
||||
seed: random seeding
|
||||
render: Render the episode
|
||||
|
||||
Returns:
|
||||
Returns: observations, rewards, dones, actions
|
||||
|
||||
"""
|
||||
env: gym.Env = make(env_id, seed=seed)
|
||||
rewards = []
|
||||
observations = []
|
||||
actions = []
|
||||
observations = []
|
||||
dones = []
|
||||
obs = env.reset()
|
||||
self._verify_observations(obs, env.observation_space, "reset()")
|
||||
|
||||
length = env.max_path_length
|
||||
if iterations is None:
|
||||
if length is None:
|
||||
iterations = 1
|
||||
else:
|
||||
iterations = length
|
||||
iterations = iterations or (env.spec.max_episode_steps or 1)
|
||||
|
||||
# number of samples(multiple environment steps)
|
||||
for i in range(iterations):
|
||||
@@ -48,7 +44,6 @@ class TestStepMetaWorlEnvironments(unittest.TestCase):
|
||||
|
||||
ac = env.action_space.sample()
|
||||
actions.append(ac)
|
||||
# ac = np.random.uniform(env.action_space.low, env.action_space.high, env.action_space.shape)
|
||||
obs, reward, done, info = env.step(ac)
|
||||
|
||||
self._verify_observations(obs, env.observation_space, "step()")
|
||||
@@ -62,36 +57,17 @@ class TestStepMetaWorlEnvironments(unittest.TestCase):
|
||||
env.render("human")
|
||||
|
||||
if done:
|
||||
obs = env.reset()
|
||||
break
|
||||
|
||||
assert done, "Done flag is not True after max episode length."
|
||||
assert done, "Done flag is not True after end of episode."
|
||||
observations.append(obs)
|
||||
env.close()
|
||||
del env
|
||||
return np.array(observations), np.array(rewards), np.array(dones), np.array(actions)
|
||||
|
||||
def _verify_observations(self, obs, observation_space, obs_type="reset()"):
|
||||
self.assertTrue(observation_space.contains(obs),
|
||||
f"Observation {obs} received from {obs_type} "
|
||||
f"not contained in observation space {observation_space}.")
|
||||
|
||||
def _verify_reward(self, reward):
|
||||
self.assertIsInstance(reward, float, f"Returned {reward} as reward, expected float.")
|
||||
|
||||
def _verify_done(self, done):
|
||||
self.assertIsInstance(done, bool, f"Returned {done} as done flag, expected bool.")
|
||||
|
||||
def test_metaworld_functionality(self):
|
||||
"""Tests that environments runs without errors using random actions."""
|
||||
for env_id in ALL_ENVS:
|
||||
with self.subTest(msg=env_id):
|
||||
self._run_env(env_id)
|
||||
|
||||
def test_metaworld_determinism(self):
|
||||
"""Tests that identical seeds produce identical trajectories."""
|
||||
def _run_env_determinism(self, ids):
|
||||
seed = 0
|
||||
# Iterate over two trajectories, which should have the same state and action sequence
|
||||
for env_id in ALL_ENVS:
|
||||
for env_id in ids:
|
||||
with self.subTest(msg=env_id):
|
||||
traj1 = self._run_env(env_id, seed=seed)
|
||||
traj2 = self._run_env(env_id, seed=seed)
|
||||
@@ -99,9 +75,44 @@ class TestStepMetaWorlEnvironments(unittest.TestCase):
|
||||
obs1, rwd1, done1, ac1, obs2, rwd2, done2, ac2 = time_step
|
||||
self.assertTrue(np.array_equal(ac1, ac2), f"Actions [{i}] delta {ac1 - ac2} is not zero.")
|
||||
self.assertTrue(np.array_equal(obs1, obs2), f"Observations [{i}] delta {obs1 - obs2} is not zero.")
|
||||
self.assertAlmostEqual(rwd1, rwd2, f"Rewards [{i}] {rwd1} and {rwd2} do not match.")
|
||||
self.assertEqual(rwd1, rwd2, f"Rewards [{i}] {rwd1} and {rwd2} do not match.")
|
||||
self.assertEqual(done1, done2, f"Dones [{i}] {done1} and {done2} do not match.")
|
||||
|
||||
def _verify_observations(self, obs, observation_space, obs_type="reset()"):
|
||||
self.assertTrue(observation_space.contains(obs),
|
||||
f"Observation {obs} received from {obs_type} "
|
||||
f"not contained in observation space {observation_space}.")
|
||||
|
||||
def _verify_reward(self, reward):
|
||||
self.assertIsInstance(reward, (float, int), f"Returned type {type(reward)} as reward, expected float or int.")
|
||||
|
||||
def _verify_done(self, done):
|
||||
self.assertIsInstance(done, bool, f"Returned {done} as done flag, expected bool.")
|
||||
|
||||
def test_step_functionality(self):
|
||||
"""Tests that step environments run without errors using random actions."""
|
||||
for env_id in CUSTOM_IDS:
|
||||
with self.subTest(msg=env_id):
|
||||
self._run_env(env_id)
|
||||
|
||||
def test_step_determinism(self):
|
||||
"""Tests that for step environments identical seeds produce identical trajectories."""
|
||||
self._run_env_determinism(CUSTOM_IDS)
|
||||
|
||||
def test_bb_functionality(self):
|
||||
"""Tests that black box environments run without errors using random actions."""
|
||||
for traj_gen, env_ids in fancy_gym.ALL_FANCY_MOVEMENT_PRIMITIVE_ENVIRONMENTS.items():
|
||||
with self.subTest(msg=traj_gen):
|
||||
for id in env_ids:
|
||||
with self.subTest(msg=id):
|
||||
self._run_env(id)
|
||||
|
||||
def test_bb_determinism(self):
|
||||
"""Tests that for black box environment identical seeds produce identical trajectories."""
|
||||
for traj_gen, env_ids in fancy_gym.ALL_FANCY_MOVEMENT_PRIMITIVE_ENVIRONMENTS.items():
|
||||
with self.subTest(msg=traj_gen):
|
||||
self._run_env_determinism(env_ids)
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
unittest.main()
|
||||
@@ -0,0 +1,118 @@
|
||||
import unittest
|
||||
|
||||
import gym
|
||||
import numpy as np
|
||||
|
||||
import fancy_gym
|
||||
from fancy_gym import make
|
||||
|
||||
GYM_IDS = [spec.id for spec in gym.envs.registry.all() if
|
||||
"fancy_gym" not in spec.entry_point and 'make_bb_env_helper' not in spec.entry_point]
|
||||
SEED = 1
|
||||
|
||||
|
||||
class TestGymEnvironments(unittest.TestCase):
|
||||
|
||||
def _run_env(self, env_id, iterations=None, seed=SEED, render=False):
|
||||
"""
|
||||
Example for running a openai gym env in the step based setting.
|
||||
The env_id has to be specified as `env_id-vX`.
|
||||
|
||||
Args:
|
||||
env_id: env id in the form `env_id-vX`
|
||||
iterations: Number of rollout steps to run
|
||||
seed: random seeding
|
||||
render: Render the episode
|
||||
|
||||
Returns:
|
||||
|
||||
"""
|
||||
env: gym.Env = make(env_id, seed=seed)
|
||||
rewards = []
|
||||
observations = []
|
||||
actions = []
|
||||
dones = []
|
||||
obs = env.reset()
|
||||
self._verify_observations(obs, env.observation_space, "reset()")
|
||||
|
||||
iterations = iterations or (env.spec.max_episode_steps or 1)
|
||||
|
||||
# number of samples(multiple environment steps)
|
||||
for i in range(iterations):
|
||||
observations.append(obs)
|
||||
|
||||
ac = env.action_space.sample()
|
||||
actions.append(ac)
|
||||
# ac = np.random.uniform(env.action_space.low, env.action_space.high, env.action_space.shape)
|
||||
obs, reward, done, info = env.step(ac)
|
||||
|
||||
self._verify_observations(obs, env.observation_space, "step()")
|
||||
self._verify_reward(reward)
|
||||
self._verify_done(done)
|
||||
|
||||
rewards.append(reward)
|
||||
dones.append(done)
|
||||
|
||||
if render:
|
||||
env.render("human")
|
||||
|
||||
if done:
|
||||
break
|
||||
|
||||
assert done or env.spec.max_episode_steps is None, "Done flag is not True after end of episode."
|
||||
observations.append(obs)
|
||||
env.close()
|
||||
del env
|
||||
return np.array(observations), np.array(rewards), np.array(dones), np.array(actions)
|
||||
|
||||
def _run_env_determinism(self, ids):
|
||||
seed = 0
|
||||
for env_id in ids:
|
||||
with self.subTest(msg=env_id):
|
||||
traj1 = self._run_env(env_id, seed=seed)
|
||||
traj2 = self._run_env(env_id, seed=seed)
|
||||
for i, time_step in enumerate(zip(*traj1, *traj2)):
|
||||
obs1, rwd1, done1, ac1, obs2, rwd2, done2, ac2 = time_step
|
||||
self.assertTrue(np.array_equal(ac1, ac2), f"Actions [{i}] delta {ac1 - ac2} is not zero.")
|
||||
self.assertTrue(np.array_equal(obs1, obs2), f"Observations [{i}] delta {obs1 - obs2} is not zero.")
|
||||
self.assertEqual(rwd1, rwd2, f"Rewards [{i}] {rwd1} and {rwd2} do not match.")
|
||||
self.assertEqual(done1, done2, f"Dones [{i}] {done1} and {done2} do not match.")
|
||||
|
||||
def _verify_observations(self, obs, observation_space, obs_type="reset()"):
|
||||
self.assertTrue(observation_space.contains(obs),
|
||||
f"Observation {obs} received from {obs_type} "
|
||||
f"not contained in observation space {observation_space}.")
|
||||
|
||||
def _verify_reward(self, reward):
|
||||
self.assertIsInstance(reward, (float, int), f"Returned type {type(reward)} as reward, expected float or int.")
|
||||
|
||||
def _verify_done(self, done):
|
||||
self.assertIsInstance(done, bool, f"Returned {done} as done flag, expected bool.")
|
||||
|
||||
def test_step_functionality(self):
|
||||
"""Tests that step environments run without errors using random actions."""
|
||||
for env_id in GYM_IDS:
|
||||
with self.subTest(msg=env_id):
|
||||
self._run_env(env_id)
|
||||
|
||||
def test_step_determinism(self):
|
||||
"""Tests that for step environments identical seeds produce identical trajectories."""
|
||||
self._run_env_determinism(GYM_IDS)
|
||||
|
||||
def test_bb_functionality(self):
|
||||
"""Tests that black box environments run without errors using random actions."""
|
||||
for traj_gen, env_ids in fancy_gym.ALL_GYM_MOVEMENT_PRIMITIVE_ENVIRONMENTS.items():
|
||||
with self.subTest(msg=traj_gen):
|
||||
for id in env_ids:
|
||||
with self.subTest(msg=id):
|
||||
self._run_env(id)
|
||||
|
||||
def test_bb_determinism(self):
|
||||
"""Tests that for black box environment identical seeds produce identical trajectories."""
|
||||
for traj_gen, env_ids in fancy_gym.ALL_GYM_MOVEMENT_PRIMITIVE_ENVIRONMENTS.items():
|
||||
with self.subTest(msg=traj_gen):
|
||||
self._run_env_determinism(env_ids)
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
unittest.main()
|
||||
@@ -0,0 +1,119 @@
|
||||
import unittest
|
||||
|
||||
import gym
|
||||
import numpy as np
|
||||
from metaworld.envs import ALL_V2_ENVIRONMENTS_GOAL_OBSERVABLE
|
||||
|
||||
import fancy_gym
|
||||
from fancy_gym import make
|
||||
|
||||
METAWORLD_IDS = [f'metaworld:{env.split("-goal-observable")[0]}' for env, _ in
|
||||
ALL_V2_ENVIRONMENTS_GOAL_OBSERVABLE.items()]
|
||||
SEED = 1
|
||||
|
||||
|
||||
class TestMetaWorldEnvironments(unittest.TestCase):
|
||||
|
||||
def _run_env(self, env_id, iterations=None, seed=SEED, render=False):
|
||||
"""
|
||||
Example for running a metaworld based env in the step based setting.
|
||||
The env_id has to be specified as `metaworld:env_id-vX`.
|
||||
|
||||
Args:
|
||||
env_id: env id in the form `metaworld:env_id-vX`
|
||||
iterations: Number of rollout steps to run
|
||||
seed: random seeding
|
||||
render: Render the episode
|
||||
|
||||
Returns:
|
||||
|
||||
"""
|
||||
env: gym.Env = make(env_id, seed=seed)
|
||||
rewards = []
|
||||
observations = []
|
||||
actions = []
|
||||
dones = []
|
||||
obs = env.reset()
|
||||
self._verify_observations(obs, env.observation_space, "reset()")
|
||||
|
||||
iterations = iterations or (env.spec.max_episode_steps or 1)
|
||||
|
||||
# number of samples(multiple environment steps)
|
||||
for i in range(iterations):
|
||||
observations.append(obs)
|
||||
|
||||
ac = env.action_space.sample()
|
||||
actions.append(ac)
|
||||
# ac = np.random.uniform(env.action_space.low, env.action_space.high, env.action_space.shape)
|
||||
obs, reward, done, info = env.step(ac)
|
||||
|
||||
self._verify_observations(obs, env.observation_space, "step()")
|
||||
self._verify_reward(reward)
|
||||
self._verify_done(done)
|
||||
|
||||
rewards.append(reward)
|
||||
dones.append(done)
|
||||
|
||||
if render:
|
||||
env.render("human")
|
||||
|
||||
if done:
|
||||
break
|
||||
|
||||
assert done, "Done flag is not True after end of episode."
|
||||
observations.append(obs)
|
||||
env.close()
|
||||
del env
|
||||
return np.array(observations), np.array(rewards), np.array(dones), np.array(actions)
|
||||
|
||||
def _run_env_determinism(self, ids):
|
||||
seed = 0
|
||||
for env_id in ids:
|
||||
with self.subTest(msg=env_id):
|
||||
traj1 = self._run_env(env_id, seed=seed)
|
||||
traj2 = self._run_env(env_id, seed=seed)
|
||||
for i, time_step in enumerate(zip(*traj1, *traj2)):
|
||||
obs1, rwd1, done1, ac1, obs2, rwd2, done2, ac2 = time_step
|
||||
self.assertTrue(np.array_equal(ac1, ac2), f"Actions [{i}] delta {ac1 - ac2} is not zero.")
|
||||
self.assertTrue(np.array_equal(obs1, obs2), f"Observations [{i}] delta {obs1 - obs2} is not zero.")
|
||||
self.assertEqual(rwd1, rwd2, f"Rewards [{i}] {rwd1} and {rwd2} do not match.")
|
||||
self.assertEqual(done1, done2, f"Dones [{i}] {done1} and {done2} do not match.")
|
||||
|
||||
def _verify_observations(self, obs, observation_space, obs_type="reset()"):
|
||||
self.assertTrue(observation_space.contains(obs),
|
||||
f"Observation {obs} received from {obs_type} "
|
||||
f"not contained in observation space {observation_space}.")
|
||||
|
||||
def _verify_reward(self, reward):
|
||||
self.assertIsInstance(reward, (float, int), f"Returned type {type(reward)} as reward, expected float or int.")
|
||||
|
||||
def _verify_done(self, done):
|
||||
self.assertIsInstance(done, bool, f"Returned {done} as done flag, expected bool.")
|
||||
|
||||
def test_step_functionality(self):
|
||||
"""Tests that step environments run without errors using random actions."""
|
||||
for env_id in METAWORLD_IDS:
|
||||
with self.subTest(msg=env_id):
|
||||
self._run_env(env_id)
|
||||
|
||||
def test_step_determinism(self):
|
||||
"""Tests that for step environments identical seeds produce identical trajectories."""
|
||||
self._run_env_determinism(METAWORLD_IDS)
|
||||
|
||||
def test_bb_functionality(self):
|
||||
"""Tests that black box environments run without errors using random actions."""
|
||||
for traj_gen, env_ids in fancy_gym.ALL_METAWORLD_MOVEMENT_PRIMITIVE_ENVIRONMENTS.items():
|
||||
with self.subTest(msg=traj_gen):
|
||||
for id in env_ids:
|
||||
with self.subTest(msg=id):
|
||||
self._run_env(id)
|
||||
|
||||
def test_bb_determinism(self):
|
||||
"""Tests that for black box environment identical seeds produce identical trajectories."""
|
||||
for traj_gen, env_ids in fancy_gym.ALL_METAWORLD_MOVEMENT_PRIMITIVE_ENVIRONMENTS.items():
|
||||
with self.subTest(msg=traj_gen):
|
||||
self._run_env_determinism(env_ids)
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
unittest.main()
|
||||
Reference in New Issue
Block a user