def test_no_user_id(self):
env = movie_lens_dynamic.create_gym_environment(self.env_config)
agent = batched_movielens_rnn_agent.MovieLensRNNAgent(
env.observation_space,
env.action_space,
stateful=True,
batch_size=1,
user_id_input=False,
user_embedding_size=0,
max_episode_length=None)
for _ in range(3):
for _ in range(7):
reward = 0
observation = env.reset()
for _ in range(2):
slate = agent.step(reward, observation)
observation, reward, _, _ = env.step(slate)
agent.end_episode(reward, observation, eval_mode=True)
# There are 7 episodes in every batch used to update the model.
agent.set_batch_size(7)
agent.model_update(learning_rate=0.1,
lambda_learning_rate=0.1,
var_learning_rate=0.1)
agent.empty_buffer()
# The agent and environment simulate one episode at a time.
agent.set_batch_size(1)
def test_interaction(self):
env = movie_lens_dynamic.create_gym_environment(self.env_config)
agent = batched_movielens_rnn_agent.MovieLensRNNAgent(
env.observation_space, env.action_space, max_episode_length=None)
for _ in range(3):
for _ in range(2):
reward = 0
observation = env.reset()
for _ in range(2):
slate = agent.step(reward, observation)
observation, reward, _, _ = env.step(slate)
agent.end_episode(reward, observation, eval_mode=True)
agent.model_update(learning_rate=0.1,
lambda_learning_rate=0.1,
var_learning_rate=0.1)
agent.empty_buffer()
def test_batch_interaction(self):
envs = [
movie_lens_dynamic.create_gym_environment(self.env_config)
for _ in range(5)
]
agent = batched_movielens_rnn_agent.MovieLensRNNAgent(
envs[0].observation_space,
envs[0].action_space,
max_episode_length=None)
for _ in range(3):
rewards = [0 for _ in envs]
observations = [env.reset() for env in envs]
for _ in range(2):
slates = agent.step(rewards, observations)
observations, rewards, _, _ = zip(
*[env.step(slate) for env, slate in zip(envs, slates)])
agent.end_episode(rewards, observations, eval_mode=True)
agent.model_update(learning_rate=0.1,
lambda_learning_rate=0.1,
var_learning_rate=0.1)
agent.empty_buffer()