0d25397aeff73831e41aab8a788837633c75f2a6,slm_lab/agent/algorithm/sil.py,PPOSIL,calc_sil_policy_val_loss,#PPOSIL#Any#,286

Before Change

        sil_val_loss = (max(R - v_pred, 0)^2) / 2
        This is called on a randomly-sample batch from experience replay
        """
        returns = math_util.calc_returns(batch, self.gamma)
        v_preds = self.calc_v(batch["states"], evaluate=False)
        clipped_advs = torch.clamp(returns - v_preds, min=0.0)
        log_probs = policy_util.calc_log_probs(self, self.net, self.body, batch)

After Change

        sil_val_loss = (max(R - v_pred, 0)^2) / 2
        This is called on a randomly-sample batch from experience replay
        """
        returns = batch["rets"]
        v_preds = self.calc_v(batch["states"], evaluate=False)
        clipped_advs = torch.clamp(returns - v_preds, min=0.0)
        log_probs = policy_util.calc_log_probs(self, self.net, self.body, batch)

Instances