search

facebookresearch / rlmeta

RLMeta is a light-weight flexible framework for Distributed Reinforcement Learning Research.
MIT License
284 stars 28 forks source link

Fix bugs in ApeX DQN #60

Closed xiaomengy closed 2 years ago

xiaomengy commented 2 years ago

This PR fixed several issues in ApeX-DQN

  1. Add global step_counter so control target_net sync frequency.
  2. Fix a bug in SegmentTree when the size is exactly 2^n