def test_no_error(self, env, task, algo):
policies = algo.train_policies if task.training else [algo.test_policy]
envs = [copy.deepcopy(env) for _ in policies]
simulator = Simulator(task, policies, envs)
algo.begin_epoch()
for score in simulator:
assert score is not None
algo.end_epoch()
algo.begin_epoch()
def test_individual_episode_when_parallel(self, task, policy, env):
def equal(x):
x = list(x)
return all(y == x[0] for y in x)
policies = copy.copy(policy), copy.copy(policy)
envs = copy.copy(env), copy.copy(env)
simulator = Simulator(task, policies, envs)
for _ in range(task.epochs):
simulator()
assert equal(x.task.episode for x in policies)
def test_individual_episode_when_parallel(
self, task, policy, second_policy, env, second_env):
def equal(x):
x = list(x)
return all(y == x[0] for y in x)
policies = [policy, second_policy]
envs = [env, second_env]
simulator = Simulator(task, policies, envs)
for _ in range(task.epochs):
simulator()
assert equal(x.task.episode for x in policies)
def _create_testing(self, algorithm):
policies = [self._prepend_score_step(algorithm.test_policy)]
envs = [self._create_env(self._task.directory)]
return Simulator(self._test_task, policies, envs)
def _create_training(self, algorithm):
policies = algorithm.train_policies
policies = [self._prepend_score_step(x) for x in policies]
envs = [self._create_env() for _ in policies]
return Simulator(self._train_task, policies, envs)