def default_replay():
return replays.Buffer(size=int(1e6),
batch_iterations=50,
batch_size=100,
discount_factor=0.98,
steps_before_batches=5000,
steps_between_batches=50)
def __init__(
self, model=None, replay=None, actor_updater=None, critic_updater=None
):
self.model = model or default_model()
self.replay = replay or replays.Buffer(return_steps=5)
self.actor_updater = actor_updater or \
updaters.MaximumAPosterioriPolicyOptimization()
self.critic_updater = critic_updater or updaters.ExpectedSARSA()
def __init__(
self, model=None, replay=None, exploration=None, actor_updater=None,
critic_updater=None
):
self.model = model or default_model()
self.replay = replay or replays.Buffer()
self.exploration = exploration or explorations.NormalActionNoise()
self.actor_updater = actor_updater or \
updaters.DeterministicPolicyGradient()
self.critic_updater = critic_updater or \
updaters.DeterministicQLearning()
def __init__(
self, model=None, replay=None, exploration=None, actor_updater=None,
critic_updater=None
):
model = model or default_model()
replay = replay or replays.Buffer(num_steps=5)
actor_updater = actor_updater or \
updaters.DistributionalDeterministicPolicyGradient()
critic_updater = critic_updater or \
updaters.DistributionalDeterministicQLearning()
super().__init__(
model, replay, exploration, actor_updater, critic_updater)