search

heronsystems / adeptRL

Reinforcement learning framework to accelerate research
GNU General Public License v3.0
203 stars 29 forks source link

acer #11

Open jtatusko opened 5 years ago

jtatusko commented 5 years ago

actor critic with experience replay https://arxiv.org/abs/1611.01224

SethKitchen commented 5 years ago

Is this Ape-X? That's the other cutting edge right now right?

SethKitchen commented 5 years ago

https://www.google.com/url?sa=t&source=web&rct=j&url=https://openreview.net/pdf%3Fid%3DH1Dy---0Z&ved=2ahUKEwiNkYvJsYveAhUM4YMKHbC6BZMQFjAAegQIBhAB&usg=AOvVaw3f2N945YY0ny8LB4kMpqQL

benbellheron commented 5 years ago

I will be making local, towered (separate replays per machine), and impala style (where workers submit experience) versions of Acer. Ape-X seems like a slightly modified version of the impala style updates where workers calculate the priority before sending so it shouldn't be that hard for me to add on once acer is done.