search

PaddlePaddle / PARL

A high-performance distributed training framework for Reinforcement Learning
https://parl.readthedocs.io/
Apache License 2.0
3.22k stars 817 forks source link

CompatWrapper impact #1030

Open ShuaibinLi opened 1 year ago

ShuaibinLi commented 1 year ago

Using CompatWrapper causes convergence of algorithm to become slower, e.g. TD3 in Humanoid.