def iev_user_model_creator(env_ctx):
return iev.IEvUserModel(
env_ctx["slate_size"],
choice_model_ctor=choice_model.MultinomialProportionalChoiceModel,
response_model_ctor=iev.IEvResponse,
user_state_ctor=iev.IEvUserState,
seed=env_ctx["seed"])
def setUp(self):
super(InterestEvolutionTest, self).setUp()
self._user_model = interest_evolution.IEvUserModel(
slate_size=1,
choice_model_ctor=choice_model.MultinomialLogitChoiceModel,
seed=0,
no_click_mass=0)
self._user_model.choice_model = FakeChoiceModel(scores=[2],
score_no_click=0,
select_index=0)
def test_slate_indices_and_length(self):
# Initialize agent
slate_size = 2
num_candidates = 100
action_space = spaces.MultiDiscrete(num_candidates * np.ones((slate_size,)))
user_model = iev.IEvUserModel(
slate_size,
choice_model_ctor=choice_model.MultinomialLogitChoiceModel,
response_model_ctor=iev.IEvResponse)
agent = random_agent.RandomAgent(action_space, random_seed=0)
# Create a set of documents
document_sampler = iev.IEvVideoSampler()
ievenv = environment.Environment(user_model, document_sampler,
num_candidates, slate_size)
# Test that slate indices in correct range and length is correct
observation, documents = ievenv.reset()
slate = agent.step(1, dict(user=observation, doc=documents))
self.assertLen(slate, slate_size)
self.assertAllInSet(slate, range(num_candidates))
def test_step(self):
# Create a simple user
slate_size = 2
user_model = iev.IEvUserModel(
slate_size,
choice_model_ctor=choice_model.MultinomialLogitChoiceModel,
response_model_ctor=iev.IEvResponse)
# Create a candidate_set with 5 items
num_candidates = 5
document_sampler = iev.IEvVideoSampler()
ievsim = environment.Environment(user_model, document_sampler,
num_candidates, slate_size)
# Create agent
action_space = spaces.MultiDiscrete(num_candidates * np.ones((slate_size,)))
agent = random_agent.RandomAgent(action_space, random_seed=0)
# This agent doesn't use the previous user response
observation, documents = ievsim.reset()
slate = agent.step(1, dict(user=observation, doc=documents))
self.assertAllEqual(slate, [2, 0])
if __name__ == '__main__':
from recsim.environments import interest_evolution
slate_size = 3
num_candidates = 10
env_config = {
'num_candidates': 5,
'slate_size': slate_size,
'resample_documents': True,
'seed': 0,
}
tmp_base_dir = 'C:/Users/Moham/PycharmProjects/ExpertDrivenRec/'
user_model = interest_evolution.IEvUserModel(
slate_size,
choice_model_ctor=choice_model.MultinomialProportionalChoiceModel,
response_model_ctor=interest_evolution.IEvResponse,
user_state_ctor=interest_evolution.IEvUserState,
seed=0)
expertEnv = environment.Environment(ExpertModel(slate_size),
EEVideoSampler(),
num_candidates,
slate_size,
resample_documents=True)
steps = 10
traj, states = generate_trajectories(expertEnv, 3, steps, num_candidates,
slate_size)