Random downsampling of the experience replay unit
dfridovi opened this issue · comments
Try randomly downsampling past experience before collecting each set of new rollouts. This way random samples of past experience are more likely to be new.
A homebrewed C++ library for reinforcement learning.
dfridovi opened this issue · comments
Try randomly downsampling past experience before collecting each set of new rollouts. This way random samples of past experience are more likely to be new.