ff9c1dac887643e464f5f829c7d8b920b0aa8140,rllib/utils/exploration/gaussian_noise.py,GaussianNoise,_get_torch_exploration_action,#GaussianNoise#Any#Any#Any#,139

Before Change

            else:
                det_actions = action_dist.deterministic_sample()
                scale = self.scale_schedule(self.last_timestep)
                gaussian_sample = scale * torch.normal(
                    mean=torch.zeros(det_actions.size()), std=self.stddev)
                action = torch.clamp(det_actions + gaussian_sample,
                                     self.action_space.low.item(0),
                                     self.action_space.high.item(0))
        // No exploration -> Return deterministic actions.

After Change

            else:
                det_actions = action_dist.deterministic_sample()
                scale = self.scale_schedule(self.last_timestep)
                gaussian_sample = scale * torch.normal(
                    mean=torch.zeros(det_actions.size()), std=self.stddev).to(
                    self.device)
                action = torch.min(
                    torch.max(
                        det_actions + gaussian_sample,
                        torch.tensor(
                            self.action_space.low,
                            dtype=torch.float32,
                            device=self.device)),
                    torch.tensor(
                        self.action_space.high,
                        dtype=torch.float32,

Instances