def default_model():
return models.ActorCritic(
actor=models.Actor(encoder=models.ObservationEncoder(),
torso=models.MLP((64, 64), 'tanh'),
head=models.DetachedScaleGaussianPolicyHead()),
critic=models.Critic(encoder=models.ObservationEncoder(),
torso=models.MLP((64, 64), 'tanh'),
head=models.ValueHead()),
observation_normalizer=normalizers.MeanStd())
def default_model():
return models.ActorCriticWithTargets(
actor=models.Actor(encoder=models.ObservationEncoder(),
torso=models.MLP((256, 256), 'relu'),
head=models.DeterministicPolicyHead()),
critic=models.Critic(encoder=models.ObservationActionEncoder(),
torso=models.MLP((256, 256), 'relu'),
head=models.ValueHead()),
observation_normalizer=normalizers.MeanStd())
def default_model():
return models.ActorCriticWithTargets(
actor=models.Actor(encoder=models.ObservationEncoder(),
torso=models.MLP((256, 256), 'relu'),
head=models.DeterministicPolicyHead()),
critic=models.Critic(
encoder=models.ObservationActionEncoder(),
torso=models.MLP((256, 256), 'relu'),
# These values are for the control suite with 0.99 discount.
head=models.DistributionalValueHead(-150., 150., 51)),
observation_normalizer=normalizers.MeanStd())
def default_model():
return models.ActorTwinCriticWithTargets(
actor=models.Actor(
encoder=models.ObservationEncoder(),
torso=models.MLP((256, 256), 'relu'),
head=models.GaussianPolicyHead(
loc_activation=None,
distribution=models.SquashedMultivariateNormalDiag)),
critic=models.Critic(encoder=models.ObservationActionEncoder(),
torso=models.MLP((256, 256), 'relu'),
head=models.ValueHead()),
observation_normalizer=normalizers.MeanStd(),
target_coeff=0.005)