I don't understand the question then. You want to run an actor disconnected from the learner, but you want it to learn? Edit: Or... You mean running the whole

I don't understand the question then. You want to run an actor disconnected from the learner, but you want it to learn? Edit: Or... You mean running the whole distributed learning algorithm on one machine. I don't see why not, but I doubt it would outperform something intended to run on a single machine.