class _TestAgentInterface(unittest.TestCase):
def setUp(self):
self.env = ABC(discrete=self.discrete,
partially_observable=self.partially_observable,
episodic=self.episodic)
def create_agent(self, env):
raise NotImplementedError()
def test_save_load(self):
a = self.create_agent(self.env)
dirname = tempfile.mkdtemp()
a.save(dirname)
self.assertTrue(os.path.exists(dirname))
b = self.create_agent(self.env)
b.load(dirname)
def test_run_episode(self):
agent = self.create_agent(self.env)
done = False
obs = self.env.reset()
t = 0
while t < 10 and not done:
a = agent.act(obs)
obs, r, done, info = self.env.step(a)
t += 1
@testing.attr.slow
def test_train(self):
agent = self.create_agent(self.env)
train_agent(agent=agent,
env=self.env,
steps=2000,
outdir=tempfile.mkdtemp(),
max_episode_len=10)
def make_env(process_idx, test):
size = 2
return ABC(size=size,
discrete=discrete,
episodic=episodic or test,
partially_observable=self.use_lstm,
deterministic=test)
def make_env_and_successful_return(self, test):
env = ABC(
discrete=self.discrete,
episodic=self.episodic or test,
deterministic=test,
)
return env, 1
def make_env_and_successful_return(self, test):
env = ABC(
discrete=self.discrete,
episodic=self.episodic or test,
deterministic=test,
partially_observable=self.recurrent,
)
return env, 1
def make_env_and_successful_return(self, test):
return ABC(discrete=self.discrete, deterministic=test), 1
def make_env_and_successful_return(self, test):
return ABC(discrete=False,
partially_observable=True,
deterministic=test), 1
def make_env(process_idx, test):
return ABC(episodic=self.episodic or test,
partially_observable=self.use_lstm,
deterministic=test)
def setUp(self):
self.env = ABC(discrete=self.discrete,
partially_observable=self.partially_observable,
episodic=self.episodic)
def make_env():
return ABC(discrete=self.discrete, deterministic=test)
def make_env():
return ABC(
discrete=self.discrete,
deterministic=test,
episodic=self.episodic,
)
