def _create_ppo_algorithm(env):
return create_ac_algorithm(env=env,
actor_fc_layers=(),
value_fc_layers=(),
use_rnns=True,
learning_rate=1e-3,
algorithm_class=PPOAlgorithm)
def create_ppo_algorithm(env, debug_summaries=False):
"""Create a simple PPOAlgorithm.
Args:
env (TFEnvironment): A TFEnvironment
debug_summaries (bool): True if debug summaries should be created.
"""
algorithm = create_ac_algorithm(env, debug_summaries=debug_summaries)
return PPOAlgorithm(algorithm)
def create_ppo_algorithm(env, debug_summaries=False):
"""Create a simple PPOAlgorithm.
Args:
env (TFEnvironment): A TFEnvironment
debug_summaries (bool): True if debug summaries should be created.
Returns:
PPOAlgorithm
"""
return create_ac_algorithm(env,
algorithm_class=PPOAlgorithm,
loss_class=PPOLoss,
debug_summaries=debug_summaries)
def _create_ac_algorithm(env):
return create_ac_algorithm(env=env,
actor_fc_layers=(8, ),
value_fc_layers=(8, ))
def _create_ac_algorithm(env):
return OffPolicyAdapter(
create_ac_algorithm(env=env,
actor_fc_layers=(8, ),
value_fc_layers=(8, )))
def _create_ppo_algorithm(env):
return PPOAlgorithm(
create_ac_algorithm(env=env,
actor_fc_layers=(16, 16),
value_fc_layers=(16, 16),
learning_rate=1e-3))