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

pseudocode.py

  class ReplayBuffer(object):
    def __init__(self, config: MuZeroConfig):
      self.window_size = config.window_size
      self.batch_size = config.batch_size
      self.buffer = []
    
    def sample_batch(self, num_unroll_steps: int, td_steps: int):
      games = [self.sample_game() for _ in range(self.batch_size)]
      game_pos = [(g, self.sample_position(g)) for g in games]
      return [(g.make_image(i), g.history[i:i + num_unroll_steps],
               g.make_target(i, num_unroll_steps, td_steps, g.to_play()))
              for (g, i) in game_pos]
    
    ...