Discrete Actions in MADDPG and A3C Algorithm Support

[SunSkyLH]

Hello, I see that the MADDPG code does not seem to support discrete actions. Have you considered supporting it?
In addition, have you considered adding the A3C algorithm?

[wenzhangliu]

We have received feedback from several users about MADDPG with discrete actions, but since MADDPG is inherently designed for continuous action spaces, it currently does not support discrete actions in XuanCe.

As for the A3C algorithm, we have internally discussed the possibility of implementing it. However, our current focus is on enhancing the performance of existing algorithms and developing cutting-edge methods in multi-agent reinforcement learning. In the future, we will consider supporting A3C based on user demand and the project’s development priorities.

[SunSkyLH]

Thank you for your response!