renameing alr module and updating tests
This commit is contained in:
Fabian
@@ -1,168 +0,0 @@
|
||||
import unittest
|
||||
|
||||
import gym
|
||||
import numpy as np
|
||||
|
||||
import alr_envs # noqa
|
||||
from alr_envs.utils.make_env_helpers import make
|
||||
|
||||
ALL_SPECS = list(spec for spec in gym.envs.registry.all() if "alr_envs" in spec.entry_point)
|
||||
SEED = 1
|
||||
|
||||
|
||||
class TestMPEnvironments(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 `domain_name-task_name` or
|
||||
for manipulation tasks as `manipulation-environment_name`
|
||||
|
||||
Args:
|
||||
env_id: Either `domain_name-task_name` or `manipulation-environment_name`
|
||||
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 = []
|
||||
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)
|
||||
|
||||
actions = env.action_space.sample()
|
||||
# ac = np.random.uniform(env.action_space.low, env.action_space.high, env.action_space.shape)
|
||||
obs, reward, done, info = env.step(actions)
|
||||
|
||||
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_alr_environment_functionality(self):
|
||||
"""Tests that environments runs without errors using random actions for ALR MP envs."""
|
||||
with self.subTest(msg="DMP"):
|
||||
for env_id in alr_envs.ALL_ALR_MOTION_PRIMITIVE_ENVIRONMENTS['DMP']:
|
||||
with self.subTest(msg=env_id):
|
||||
self._run_env(env_id)
|
||||
|
||||
with self.subTest(msg="ProMP"):
|
||||
for env_id in alr_envs.ALL_ALR_MOTION_PRIMITIVE_ENVIRONMENTS['ProMP']:
|
||||
with self.subTest(msg=env_id):
|
||||
self._run_env(env_id)
|
||||
|
||||
def test_openai_environment_functionality(self):
|
||||
"""Tests that environments runs without errors using random actions for OpenAI gym MP envs."""
|
||||
with self.subTest(msg="DMP"):
|
||||
for env_id in alr_envs.ALL_GYM_MOTION_PRIMITIVE_ENVIRONMENTS['DMP']:
|
||||
with self.subTest(msg=env_id):
|
||||
self._run_env(env_id)
|
||||
|
||||
with self.subTest(msg="ProMP"):
|
||||
for env_id in alr_envs.ALL_GYM_MOTION_PRIMITIVE_ENVIRONMENTS['ProMP']:
|
||||
with self.subTest(msg=env_id):
|
||||
self._run_env(env_id)
|
||||
|
||||
def test_dmc_environment_functionality(self):
|
||||
"""Tests that environments runs without errors using random actions for DMC MP envs."""
|
||||
with self.subTest(msg="DMP"):
|
||||
for env_id in alr_envs.ALL_DMC_MOVEMENT_PRIMITIVE_ENVIRONMENTS['DMP']:
|
||||
with self.subTest(msg=env_id):
|
||||
self._run_env(env_id)
|
||||
|
||||
with self.subTest(msg="ProMP"):
|
||||
for env_id in alr_envs.ALL_DMC_MOVEMENT_PRIMITIVE_ENVIRONMENTS['ProMP']:
|
||||
with self.subTest(msg=env_id):
|
||||
self._run_env(env_id)
|
||||
|
||||
def test_metaworld_environment_functionality(self):
|
||||
"""Tests that environments runs without errors using random actions for Metaworld MP envs."""
|
||||
with self.subTest(msg="DMP"):
|
||||
for env_id in alr_envs.ALL_METAWORLD_MOTION_PRIMITIVE_ENVIRONMENTS['DMP']:
|
||||
with self.subTest(msg=env_id):
|
||||
self._run_env(env_id)
|
||||
|
||||
with self.subTest(msg="ProMP"):
|
||||
for env_id in alr_envs.ALL_METAWORLD_MOTION_PRIMITIVE_ENVIRONMENTS['ProMP']:
|
||||
with self.subTest(msg=env_id):
|
||||
self._run_env(env_id)
|
||||
|
||||
def test_alr_environment_determinism(self):
|
||||
"""Tests that identical seeds produce identical trajectories for ALR MP Envs."""
|
||||
with self.subTest(msg="DMP"):
|
||||
self._run_env_determinism(alr_envs.ALL_ALR_MOTION_PRIMITIVE_ENVIRONMENTS["DMP"])
|
||||
with self.subTest(msg="ProMP"):
|
||||
self._run_env_determinism(alr_envs.ALL_ALR_MOTION_PRIMITIVE_ENVIRONMENTS["ProMP"])
|
||||
|
||||
def test_openai_environment_determinism(self):
|
||||
"""Tests that identical seeds produce identical trajectories for OpenAI gym MP Envs."""
|
||||
with self.subTest(msg="DMP"):
|
||||
self._run_env_determinism(alr_envs.ALL_GYM_MOTION_PRIMITIVE_ENVIRONMENTS["DMP"])
|
||||
with self.subTest(msg="ProMP"):
|
||||
self._run_env_determinism(alr_envs.ALL_GYM_MOTION_PRIMITIVE_ENVIRONMENTS["ProMP"])
|
||||
|
||||
def test_dmc_environment_determinism(self):
|
||||
"""Tests that identical seeds produce identical trajectories for DMC MP Envs."""
|
||||
with self.subTest(msg="DMP"):
|
||||
self._run_env_determinism(alr_envs.ALL_DMC_MOVEMENT_PRIMITIVE_ENVIRONMENTS["DMP"])
|
||||
with self.subTest(msg="ProMP"):
|
||||
self._run_env_determinism(alr_envs.ALL_DMC_MOVEMENT_PRIMITIVE_ENVIRONMENTS["ProMP"])
|
||||
|
||||
def test_metaworld_environment_determinism(self):
|
||||
"""Tests that identical seeds produce identical trajectories for Metaworld MP Envs."""
|
||||
with self.subTest(msg="DMP"):
|
||||
self._run_env_determinism(alr_envs.ALL_METAWORLD_MOTION_PRIMITIVE_ENVIRONMENTS["DMP"])
|
||||
with self.subTest(msg="ProMP"):
|
||||
self._run_env_determinism(alr_envs.ALL_METAWORLD_MOTION_PRIMITIVE_ENVIRONMENTS["ProMP"])
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
unittest.main()
|
||||
@@ -0,0 +1,118 @@
|
||||
import unittest
|
||||
|
||||
import gym
|
||||
import numpy as np
|
||||
|
||||
import alr_envs # noqa
|
||||
from alr_envs.utils.make_env_helpers import make
|
||||
|
||||
CUSTOM_IDS = [spec.id for spec in gym.envs.registry.all() if
|
||||
"alr_envs" in spec.entry_point and not 'make_bb_env_helper' in spec.entry_point]
|
||||
SEED = 1
|
||||
|
||||
|
||||
class TestCustomEnvironments(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 `domain_name-task_name` or
|
||||
for manipulation tasks as `manipulation-environment_name`
|
||||
|
||||
Args:
|
||||
env_id: Either `domain_name-task_name` or `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 = []
|
||||
actions = []
|
||||
observations = []
|
||||
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)
|
||||
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 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 alr_envs.ALL_ALR_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 alr_envs.ALL_ALR_MOVEMENT_PRIMITIVE_ENVIRONMENTS.items():
|
||||
with self.subTest(msg=traj_gen):
|
||||
self._run_env_determinism(env_ids)
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
unittest.main()
|
||||
@@ -5,31 +5,31 @@ import numpy as np
|
||||
|
||||
from dm_control import suite, manipulation
|
||||
|
||||
import alr_envs
|
||||
from alr_envs import make
|
||||
|
||||
DMC_ENVS = [f'dmc:{env}-{task}' for env, task in suite.ALL_TASKS if env != "lqr"]
|
||||
MANIPULATION_SPECS = [f'dmc:manipulation-{task}' for task in manipulation.ALL if task.endswith('_features')]
|
||||
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 TestStepDMCEnvironments(unittest.TestCase):
|
||||
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 `domain_name-task_name` or
|
||||
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 `domain_name-task_name` or `manipulation-environment_name`
|
||||
env_id: Either `dmc:domain_name-task_name` or `dmc: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
|
||||
|
||||
"""
|
||||
print(env_id)
|
||||
env: gym.Env = make(env_id, seed=seed)
|
||||
rewards = []
|
||||
observations = []
|
||||
@@ -68,6 +68,19 @@ class TestStepDMCEnvironments(unittest.TestCase):
|
||||
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} "
|
||||
@@ -79,47 +92,39 @@ class TestStepDMCEnvironments(unittest.TestCase):
|
||||
def _verify_done(self, done):
|
||||
self.assertIsInstance(done, bool, f"Returned {done} as done flag, expected bool.")
|
||||
|
||||
def test_dmc_functionality(self):
|
||||
"""Tests that environments runs without errors using random actions."""
|
||||
for env_id in DMC_ENVS:
|
||||
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_dmc_determinism(self):
|
||||
"""Tests that identical seeds produce identical trajectories."""
|
||||
seed = 0
|
||||
# Iterate over two trajectories, which should have the same state and action sequence
|
||||
for env_id in DMC_ENVS:
|
||||
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 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 environments runs without errors using random actions."""
|
||||
for env_id in MANIPULATION_SPECS:
|
||||
"""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 identical seeds produce identical trajectories."""
|
||||
seed = 0
|
||||
# Iterate over two trajectories, which should have the same state and action sequence
|
||||
for env_id in MANIPULATION_SPECS:
|
||||
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.")
|
||||
"""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 alr_envs.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 alr_envs.ALL_DMC_MOVEMENT_PRIMITIVE_ENVIRONMENTS.items():
|
||||
with self.subTest(msg=traj_gen):
|
||||
self._run_env_determinism(env_ids)
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
@@ -3,25 +3,24 @@ import unittest
|
||||
import gym
|
||||
import numpy as np
|
||||
|
||||
import alr_envs
|
||||
from alr_envs import make
|
||||
from metaworld.envs import ALL_V2_ENVIRONMENTS_GOAL_OBSERVABLE
|
||||
|
||||
ALL_ENVS = [f'metaworld:{env.split("-goal-observable")[0]}' for env, _ in ALL_V2_ENVIRONMENTS_GOAL_OBSERVABLE.items()]
|
||||
METAWORLD_IDS = []
|
||||
SEED = 1
|
||||
|
||||
|
||||
class TestStepMetaWorlEnvironments(unittest.TestCase):
|
||||
class TestGymEnvironments(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 `domain_name-task_name` or
|
||||
for manipulation tasks as `manipulation-environment_name`
|
||||
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: Either `domain_name-task_name` or `manipulation-environment_name`
|
||||
env_id: env id in the form `env_id-vX`
|
||||
iterations: Number of rollout steps to run
|
||||
seed= random seeding
|
||||
seed: random seeding
|
||||
render: Render the episode
|
||||
|
||||
Returns:
|
||||
@@ -65,6 +64,19 @@ class TestStepMetaWorlEnvironments(unittest.TestCase):
|
||||
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} "
|
||||
@@ -76,26 +88,29 @@ class TestStepMetaWorlEnvironments(unittest.TestCase):
|
||||
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:
|
||||
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_metaworld_determinism(self):
|
||||
"""Tests that identical seeds produce identical trajectories."""
|
||||
seed = 0
|
||||
# Iterate over two trajectories, which should have the same state and action sequence
|
||||
for env_id in ALL_ENVS:
|
||||
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 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 alr_envs.ALL_GYM_MOTION_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 alr_envs.ALL_GYM_MOTION_PRIMITIVE_ENVIRONMENTS.items():
|
||||
with self.subTest(msg=traj_gen):
|
||||
self._run_env_determinism(env_ids)
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
@@ -0,0 +1,119 @@
|
||||
import unittest
|
||||
|
||||
import gym
|
||||
import numpy as np
|
||||
|
||||
import alr_envs
|
||||
from alr_envs import make
|
||||
from metaworld.envs import ALL_V2_ENVIRONMENTS_GOAL_OBSERVABLE
|
||||
|
||||
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 alr_envs.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 alr_envs.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