hucc/agents/sac.py [270:293]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
                ):
                    tp.data.lerp_(p.data, 1.0 - self._polyak)

        # These are the stats for the last update
        self.tbw_add_scalar('Loss/Policy', pi_loss.item())
        self.tbw_add_scalar('Loss/QValue', q_loss.item())
        self.tbw_add_scalar('Health/Entropy', -log_prob.mean())
        if self._optim_alpha:
            self.tbw_add_scalar('Health/Alpha', self._log_alpha.exp().item())
        if self._n_updates % 100 == 1:
            self.tbw.add_scalars(
                'Health/GradNorms',
                {
                    k: v.grad.norm().item()
                    for k, v in self._model.named_parameters()
                    if v.grad is not None
                },
                self.n_samples,
            )

        avg_cr = th.cat(self._cur_rewards).mean().item()
        log.info(
            f'Sample {self._n_samples}, up {self._n_updates*self._num_updates}, avg cur reward {avg_cr:+0.3f}, pi loss {pi_loss.item():+.03f}, q loss {q_loss.item():+.03f}, entropy {-log_prob.mean().item():+.03f}, alpha {self._log_alpha.exp().item():.03f}'
        )
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



hucc/agents/sachrl.py [435:458]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
                ):
                    tp.data.lerp_(p.data, 1.0 - self._polyak)

        # These are the stats for the last update
        self.tbw_add_scalar('Loss/Policy', pi_loss.item())
        self.tbw_add_scalar('Loss/QValue', q_loss.item())
        self.tbw_add_scalar('Health/Entropy', -log_prob.mean())
        if self._optim_alpha:
            self.tbw_add_scalar('Health/Alpha', self._log_alpha.exp().item())
        if self._n_updates % 100 == 1:
            self.tbw.add_scalars(
                'Health/GradNorms',
                {
                    k: v.grad.norm().item()
                    for k, v in self._model.named_parameters()
                    if v.grad is not None
                },
                self.n_samples,
            )

        avg_cr = th.cat(self._cur_rewards).mean().item()
        log.info(
            f'Sample {self._n_samples}, up {self._n_updates*self._num_updates}, avg cur reward {avg_cr:+0.3f}, pi loss {pi_loss.item():+.03f}, q loss {q_loss.item():+.03f}, entropy {-log_prob.mean().item():+.03f}, alpha {self._log_alpha.exp().item():.03f}'
        )
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -