034ee51fb01fea37fb75e196a04cd96de102da28,torch_ac/algos/ppo.py,PPOAlgo,update_parameters,#PPOAlgo#,22

Before Change

                preprocessed_obs = self.preprocess_obss(b.obs, use_gpu=torch.cuda.is_available())
                dist = self.acmodel.get_dist(preprocessed_obs)
                value = self.acmodel.get_value(preprocessed_obs)

                entropy = dist.entropy()

                ratio = torch.exp(dist.log_prob(b.action) - b.old_log_prob)

After Change

                preprocessed_obs = self.preprocess_obss(b.obs, device=self.device)
                if self.is_recurrent:
                    dist, value, _ = self.acmodel(preprocessed_obs, b.state * b.mask)
                else:
                    dist, value = self.acmodel(preprocessed_obs)

Instances