allenact.algorithms.onpolicy_sync.losses.a2cacktr
#
Implementation of A2C and ACKTR losses.
A2CACKTR
#
class A2CACKTR(AbstractActorCriticLoss)
Class implementing A2C and ACKTR losses.
Attributes
acktr
:True
if should use ACKTR loss (currently not supported), otherwise uses A2C loss.value_loss_coef
: Weight of value loss.entropy_coef
: Weight of entropy (encouraging) loss.entropy_method_name
: Name of Distr's entropy method name. Default isentropy
, but we might useconditional_entropy
forSequentialDistr
.
A2CACKTR.__init__
#
| __init__(value_loss_coef, entropy_coef, acktr=False, entropy_method_name: str = "entropy", *args, **kwargs, *, ,)
Initializer.
See class documentation for parameter definitions.
A2C
#
class A2C(A2CACKTR)
A2C Loss.
ACKTR
#
class ACKTR(A2CACKTR)
ACKTR Loss.
This code is not supported as it currently lacks an implementation for recurrent models.