search

heronsystems / adeptRL

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

sc2 3d action_space #44

Open jtatusko opened 5 years ago

jtatusko commented 5 years ago

would be nice to replicate some of the work in the papers where convLSTM, RMC, RMA are used. currently action space is all 1d outputs. 84x84 outputs are treated as separate heads, 84, 84