support for DDPG? #1
Description
I'm trying to replicate the results (cliff-cheetah) from the paper but it seems that DDPG isn't supported yet. In the README it says support for continuous control (which I assume is the DDPG) should have arrived in May 2018 but perhaps the repo hasn't been updated yet? If there are any updates you're willing to share that would be much appreciated.
I've had success running the default code (small-gridworld).
Despite the NotImplementedError from the DDPGAgent I ran the cliff-cheetah experiment both with learn_safely and learn_dangerously and the code runs but in both cases the DDPG agent fails to learn (total reward stuck around 7). If it's difficult to share the updated code outright, any advice on fixing this issue would be much appreciated.
(edit) perhaps this issue has to do with the outdated version of coach?
Below is the result for the cliff-cheetah env with learn_safely.
