isl-org
What is continuous_controls for?
When I read the paper, they say that it works at discrete action space.
Is it also possible at continuous action space???
There is just discrete control. Continuous is not implemented. I started at one point, but never had time to finish.
Thank you for your fast reply.
Have you ever thought of applying DFP to actor-critic algorithms???
Now, I am considering it to expand it to enable it to work at continuous action space. If it is possible, I guess it would work with DDPG algorithm.
During the research, I have used gradients of f over actions to update/train actor network. However, I am not sure if it is working.... ( from my implementation, it is not working.... )
Could you give me some advice??
Wonchul Kim
2017-09-05 16:32 GMT+09:00 dosovits [email protected]:
There is just discrete control. Continuous is not implemented. I started at one point, but never had time to finish.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/IntelVCL/DirectFuturePrediction/issues/6#issuecomment-327095275, or mute the thread https://github.com/notifications/unsubscribe-auth/AT7jweJYEBWAIRSb4h54OPfMKuU55nwTks5sfPkJgaJpZM4PMagr .
We also quickly tried putting DFP in actor-critic, and it didn't work amazingly well. We only looked very briefly at this, though, and don't have a very good understanding of what might be going on.