How severe does this issue affect your experience of using Ray?
- High: It blocks me to complete my task.
I am interested in utilizing CQL for a discrete action space. However, I noticed that RLlib’s CQL implementation only supports continuous actions. Is there any plan to extend support to discrete actions?
I have been using RLlib for several years, primarily with PPO, but now I need to use CQL. While TorchRL supports CQL for discrete actions, migrating to TorchRL is not feasible for me as my entire training pipeline is based on RLlib.
It would be immensely beneficial if the RLlib team could incorporate this functionality.