def predict_option_termination(self, state, current_option):
termination = self.terminations(state)[:, current_option].sigmoid()
option_termination = Bernoulli(termination).sample()
Q = self.get_Q(state)
next_option = Q.argmax(dim=-1)
return bool(option_termination.item()), next_option.item()
def predict_option_termination(self, state, current_option):
state = torch.tensor(state).float().to(device)
state = self.qnetwork_local.state(state)
termination = self.qnetwork_local.terminations(state).softmax(dim = -1)
termination = termination[current_option]
#termination = self.qnetwork_local.terminations(state)[current_option].sigmoid()
option_termination = Bernoulli(termination).sample()
Q = self.get_Q(state)
next_option = Q.argmax(dim=-1)
return bool(option_termination.item()), next_option.item()