Off-policy reinforcement learning algorithms maintain a replay buffer to utilize samples obtained from earlier policies. The sampling strategy that prioritizes certain data in a buffer to train the value function or the policy, has been shown to ...