search

PKU-RL / DGN

DGN Code
334 stars 89 forks source link

DGN with continuous action spaces #25

Open risufaj opened 3 years ago

risufaj commented 3 years ago

Hi,

I am working on a project where I plan on applying DGN. However, the ideal setting of the experiments entails a continuous action space. First of all, can DGN be used as is with continuous actions spaces? I know that DDPG or MADDPG support cont. action spaces due to their actor-critic architectures. However I can't seem to figure out if DGN is so different from such method as not to support continuous action spaces.

Any words of advice will be greatly appreciated.

iromani commented 2 years ago

Hi risufaj, have you solved your problem? My initial suggestion would be to simply discretize the action space.