def forward(self, input):
mus, sigmas, logpi, rewards, not_terminals, next_hiddens = self.data_parallel(
input.action.float_features, input.state.float_features)
return rlt.MemoryNetworkOutput(
mus=mus,
sigmas=sigmas,
logpi=logpi,
reward=rewards,
terminal=not_terminals,
next_lstm_hidden=next_hiddens[0],
next_lstm_cell=next_hiddens[1],
)
def forward(self, input):
(
mus,
sigmas,
logpi,
rewards,
not_terminals,
all_steps_hidden,
last_step_hidden_and_cell,
) = self.data_parallel(input.action.float_features,
input.state.float_features)
return rlt.MemoryNetworkOutput(
mus=mus,
sigmas=sigmas,
logpi=logpi,
reward=rewards,
terminal=not_terminals,
next_lstm_hidden=last_step_hidden_and_cell[0],
next_lstm_cell=last_step_hidden_and_cell[1],
all_steps_lstm_hidden=all_steps_hidden,
)
