def act(self, *, states, horizons, internals, auxiliaries, independent,
return_internals):
if independent: # TODO: or temp constant 0.0
embedding = self.network.apply(x=states,
horizons=horizons,
internals=internals,
independent=independent,
return_internals=return_internals)
if return_internals:
embedding, internals = embedding
def function(name, distribution):
conditions = auxiliaries.get(name, default=TensorDict())
parameters = distribution.parametrize(x=embedding,
conditions=conditions)
return distribution.mode(parameters=parameters)
actions = self.distributions.fmap(function=function,
cls=TensorDict,
with_names=True)
if return_internals:
return actions, internals
else:
return actions
else:
return Stochastic.act(self=self,
states=states,
horizons=horizons,
internals=internals,
auxiliaries=auxiliaries,
independent=independent,
return_internals=return_internals)
def tf_act(self, states, internals, auxiliaries, return_internals):
return Stochastic.tf_act(self=self,
states=states,
internals=internals,
auxiliaries=auxiliaries,
return_internals=return_internals)