def test_prop_default(self):
self._randomizing_method = Method.ABSOLUTE
variations = Variations()
variations.randomize() \
.at_xpath("//body[@name=\'{0}\']".format(self._bodyname)) \
.attribute("{0}".format(self._attribute)) \
.with_method(self._randomizing_method) \
.sampled_from(Distribution.UNIFORM) \
.with_range(self._lower_bound, self._upper_bound) \
.add()
assert variations.get_list()[0].default is None
def test_prop_mean_std(self):
self._randomizing_method = Method.ABSOLUTE
variations = Variations()
variations.randomize() \
.at_xpath("//body[@name=\'{0}\']".format(self._bodyname)) \
.attribute("{0}".format(self._attribute)) \
.with_method(self._randomizing_method) \
.sampled_from(Distribution.GAUSSIAN) \
.with_mean_std(self._mean, self._std_dev) \
.add()
assert variations.get_list()[0].mean_std == (self._mean, self._std_dev)
def test_exception_uniform_dist(self):
self._randomizing_method = Method.ABSOLUTE
variations = Variations()
with self.assertRaises(ValueError) as context:
variations.randomize() \
.at_xpath("//body[@name=\'{0}\']".format(self._bodyname)) \
.attribute("{0}".format(self._attribute)) \
.with_method(self._randomizing_method) \
.sampled_from(Distribution.UNIFORM) \
.with_mean_std(self._lower_bound, self._upper_bound) \
.add()
assert "Need to call with_range when sampled from Uniform" \
in str(context.exception)
def test_dynamics_rand(self):
variations = Variations()
variations.randomize() \
.at_xpath(".//geom[@name='torso']") \
.attribute("density") \
.with_method(Method.COEFFICIENT) \
.sampled_from(Distribution.UNIFORM) \
.with_range(0.5, 1.5) \
.add()
env = randomize(SwimmerEnv(), variations)
for i in range(5):
env.reset()
for j in range(5):
env.step(env.action_space.sample())
def test_xml_attrib_exception(self):
fake_attrib = "fake_attrib"
self._randomizing_method = Method.ABSOLUTE
variations = Variations()
variations.randomize() \
.at_xpath("//body[@name=\'{0}\']".format(self._bodyname)) \
.attribute("{0}".format(fake_attrib)) \
.with_method(self._randomizing_method) \
.sampled_from(Distribution.UNIFORM) \
.with_range(self._lower_bound, self._upper_bound) \
.add()
with self.assertRaises(ValueError) as context:
randomize(self._env, variations)
assert fake_attrib in str(context.exception)
def test_exception_bad_sampling_method(self):
# Use unused Enum value for test
self._randomizing_method = 3
variations = Variations()
variations.randomize() \
.at_xpath("//body[@name=\'{0}\']".format(self._bodyname)) \
.attribute("{0}".format(self._attribute)) \
.with_method(self._randomizing_method) \
.sampled_from(Distribution.UNIFORM) \
.with_range(self._lower_bound, self._upper_bound) \
.add()
randomized_env = randomize(self._env, variations)
with self.assertRaises(ValueError) as context:
randomized_env.reset()
assert "Unknown method" in str(context.exception)
def test_env_step(self):
self._randomizing_method = Method.ABSOLUTE
variations = Variations()
variations.randomize() \
.at_xpath("//body[@name=\'{0}\']".format(self._bodyname)) \
.attribute("{0}".format(self._attribute)) \
.with_method(self._randomizing_method) \
.sampled_from(Distribution.UNIFORM) \
.with_range(self._lower_bound, self._upper_bound) \
.add()
second_var = "torso" # Add a second
variations.randomize() \
.at_xpath("//body[@name=\'{0}\']".format(second_var)) \
.attribute("{0}".format(self._attribute)) \
.with_method(self._randomizing_method) \
.sampled_from(Distribution.UNIFORM) \
.with_range(self._lower_bound, self._upper_bound) \
.add()
randomized_env = randomize(self._env, variations)
# step() shouldn't error out (so it actually steps the wrapped env)
for j in range(5):
assert randomized_env.step(randomized_env.action_space.sample())
def test_absolute_method(self):
self._randomizing_method = Method.ABSOLUTE
variations = Variations()
variations.randomize() \
.at_xpath("//body[@name=\'{0}\']".format(self._bodyname)) \
.attribute("{0}".format(self._attribute)) \
.with_method(self._randomizing_method) \
.sampled_from(Distribution.UNIFORM) \
.with_range(self._lower_bound, self._upper_bound) \
.add()
randomized_env = randomize(self._env, variations)
randomized_vals = []
for i in range(5):
with unittest.mock.patch(
'garage.envs.mujoco.randomization.variation.Variations.'
'get_randomized_xml_model', self.create_randomized_xml):
randomized_env.reset()
body_id = randomized_env.wrapped_env.sim.model._body_name2id[
self._bodyname]
randomized_val = np.array(
randomized_env.wrapped_env.sim.model.body_pos[body_id])
randomized_vals.append(randomized_val)
# check if within range
assert all(self._lower_bound <= randomized_val), (
"Randomised {1} < lower bound {0}".format(
self._lower_bound, randomized_val))
assert all(randomized_val <= self._upper_bound), (
"Randomised {1} > upper bound {0}".format(
self._upper_bound, randomized_val))
# check that you have actual variation
randomized_vals = np.array(randomized_vals)
assert np.std(randomized_vals) > 0, (
"Std Dev of randomized values "
"not > 0. Getting the exact "
"same numbers?\n {0}".format(randomized_vals))
from garage.envs.mujoco import SwimmerEnv
from garage.envs.mujoco.randomization import Distribution
from garage.envs.mujoco.randomization import Method
from garage.envs.mujoco.randomization import randomize
from garage.envs.mujoco.randomization import Variations
variations = Variations()
variations.randomize() \
.at_xpath(".//geom[@name='torso']") \
.attribute("density") \
.with_method(Method.COEFFICIENT) \
.sampled_from(Distribution.UNIFORM) \
.with_range(0.5, 1.5) \
.add()
env = randomize(SwimmerEnv(), variations)
for i in range(1000):
env.reset()
for j in range(1000):
env.step(env.action_space.sample())
def test_prop_get_list(self):
variations = Variations()
assert variations.get_list() == []