run_selfplay (https://arxiv.org/src/1911.08265v1/anc/pseudocode.py)

pseudocode.py

# Each self-play job is independent of all others; it takes the latest network
# snapshot, produces a game and makes it available to the training job by
# writing it to a shared replay buffer.
def run_selfplay(config: MuZeroConfig, storage: SharedStorage,
                 replay_buffer: ReplayBuffer):
  while True:
    network = storage.latest_network()
    game = play_game(config, network)
    replay_buffer.save_game(game)