Kuang-Huei Lee

Do you mean you run into an error? If so, can you provide error logs for more context?

#216 - for continuous actions

Yes, I think this makes sense. Please make a pull request if you are interested.

Can you provide the results you get on both datasets? Thanks.

It seems like this is the only mismatch between meta data and actual file path. I am investigating why this happened and will update the dataset once I conclude. Thanks.

Hi, many people were able to reproduce similar results, either in their published works, feedback in this repo, or privately informed me. I myself also cloned the code from this...

What NormalProjectionNetwork does is squashing actions with tanh and it shouldn't go out of the bounds if `scale_distribution=False`. I am not very sure why this can happen. Would you like...

PPO does not respect action boundaries: https://github.com/openai/baselines/issues/121. Environment is expected to clip action values. DDPG/D4PG clips action values in its policy. SAC nicely handles this with a tanh squashed action...

So TF-Agents ddpg does clipping in policy: https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ddpg/ddpg_agent.py#L166 If you are using ddpg, you should be good. If you are using TF-Agents PPO, you should use the ActionClipWrapper that @oars...