def init_prediction(self, session, trick_prediction, featurizer=None):
self.session = session
self.trick_prediction = trick_prediction
if featurizer is None:
self.featurizer = Featurizers.Featurizer()
else:
self.featurizer = featurizer
def __init__(self,
estimator=None,
policy=None,
featurizer=None,
trick_prediction=None,
session=None,
path=None):
if featurizer is None:
featurizer = Featurizers.Featurizer()
if estimator is None:
assert session is not None
estimator = PolicyGradient(session,
featurizer.get_state_size(),
path=path)
else:
assert isinstance(estimator, PolicyGradient)
if policy is None:
policy = Policies.MaxPolicy(estimator)
super(PGAgent, self).__init__(estimator=estimator,
policy=policy,
featurizer=featurizer,
trick_prediction=trick_prediction,
session=session,
path=path)
def __init__(self,
estimator=None,
policy=None,
featurizer=None,
trick_prediction=None,
session=None,
path=None):
if featurizer is None:
featurizer = Featurizers.Featurizer()
if estimator is None:
assert session is not None
estimator = ValueEstimators.DuelingDQNEstimator(
session, input_shape=featurizer.get_state_size(), path=path)
super(DuelingAgent, self).__init__(estimator=estimator,
policy=policy,
featurizer=featurizer,
trick_prediction=trick_prediction,
session=session,
path=path)
def init_player(self,
session,
path,
estimator=None,
policy=None,
featurizer=None):
self.session = session
self.path = path
if featurizer is None:
self.featurizer = Featurizers.Featurizer()
else:
self.featurizer = featurizer
if estimator is None:
assert session is not None
self.estimator = ValueEstimators.DQNEstimator(
session,
input_shape=self.featurizer.get_state_size(),
path=path)
else:
self.estimator = estimator
if policy is None:
self.policy = Policies.EGreedyPolicy(self.estimator, epsilon=0.1)
else:
self.policy = policy