19ad2d60a7022bb5125855c029f27d86aaa46d64,rl_coach/filters/reward/reward_normalization_filter.py,RewardNormalizationFilter,filter,#RewardNormalizationFilter#Any#Any#,65

Before Change

        if update_internal_state:
            self.running_rewards_stats.push(reward)

        reward = (reward - self.running_rewards_stats.mean) / \
                      (self.running_rewards_stats.std + 1e-15)
        reward = np.clip(reward, self.clip_min, self.clip_max)

        return reward

    def get_filtered_reward_space(self, input_reward_space: RewardSpace) -> RewardSpace:
        return input_reward_space

After Change

                reward = np.array([[reward]])
            self.running_rewards_stats.push(reward)

        return self.running_rewards_stats.normalize(reward).squeeze()

    def get_filtered_reward_space(self, input_reward_space: RewardSpace) -> RewardSpace:
        self.running_rewards_stats.set_params(shape=(1,), clip_values=(self.clip_min, self.clip_max))
        return input_reward_space

Instances