Are there any demos that can illustrate the process of using MATD3 / MADDPG to process discrete actions?

[niceban]

Are there any demos that can illustrate the process of using MATD3 / MADDPG to process discrete actions?

[Yonv1943]

QMix and VDN are MARL algorithm on discrete action space. These algorithms process discrete actions in MARL env.

MATD3 and MADDPG are MARL algorithm on continuous action space. It is possible to convert continuous actions into discrete actions.

• Code gumbel_softmax_sampling jupyter notebook
• See The Gumbel-Softmax Trick for Inference of Discrete Variables - FEBRUARY 1, 2017 - GONZALO MENA
• Fun gumbel_max_demo