def instantiate_adapt_agent(self) -> agents.Agent:
self.agent = agents.Agent(
actor=fakes.Actor(self._spec),
learner=FakeLearner(),
min_observations=1,
observations_per_step=1,
)
return self.agent
def test_environment_loop(self):
# Create the actor/environment and stick them in a loop.
environment = fakes.DiscreteEnvironment(episode_length=10)
actor = fakes.Actor(specs.make_environment_spec(environment))
loop = environment_loop.EnvironmentLoop(environment, actor)
# Run the loop. There should be episode_length+1 update calls per episode.
loop.run(num_episodes=10)
self.assertEqual(actor.num_updates, 100)
def _parameterized_setup(discount_spec: Optional[types.NestedSpec] = None,
reward_spec: Optional[types.NestedSpec] = None):
"""Common setup code that, unlike self.setUp, takes arguments.
Args:
discount_spec: None, or a (nested) specs.BoundedArray.
reward_spec: None, or a (nested) specs.Array.
Returns:
environment, actor, loop
"""
env_kwargs = {'episode_length': EPISODE_LENGTH}
if discount_spec:
env_kwargs['discount_spec'] = discount_spec
if reward_spec:
env_kwargs['reward_spec'] = reward_spec
environment = fakes.DiscreteEnvironment(**env_kwargs)
actor = fakes.Actor(specs.make_environment_spec(environment))
loop = environment_loop.EnvironmentLoop(environment, actor)
return actor, loop
def instantiate_test_actor(self) -> acme.Actor:
self.test_actor = fakes.Actor(self._spec)
return self.test_actor
def setUp(self):
super().setUp()
# Create the actor/environment and stick them in a loop.
environment = fakes.DiscreteEnvironment(episode_length=EPISODE_LENGTH)
self.actor = fakes.Actor(specs.make_environment_spec(environment))
self.loop = environment_loop.EnvironmentLoop(environment, self.actor)