in level_replay/level_sampler.py [0:0]
def __init__(
self, seeds, obs_space, action_space, num_actors=1,
strategy='random', replay_schedule='fixed', score_transform='power',
temperature=1.0, eps=0.05,
rho=0.2, nu=0.5, alpha=1.0,
staleness_coef=0, staleness_transform='power', staleness_temperature=1.0):
self.obs_space = obs_space
self.action_space = action_space
self.strategy = strategy
self.replay_schedule = replay_schedule
self.score_transform = score_transform
self.temperature = temperature
self.eps = eps
self.rho = rho
self.nu = nu
self.alpha = alpha
self.staleness_coef = staleness_coef
self.staleness_transform = staleness_transform
self.staleness_temperature = staleness_temperature
# Track seeds and scores as in np arrays backed by shared memory
self._init_seed_index(seeds)
self.unseen_seed_weights = np.array([1.]*len(seeds))
self.seed_scores = np.array([0.]*len(seeds), dtype=np.float)
self.partial_seed_scores = np.zeros((num_actors, len(seeds)), dtype=np.float)
self.partial_seed_steps = np.zeros((num_actors, len(seeds)), dtype=np.int64)
self.seed_staleness = np.array([0.]*len(seeds), dtype=np.float)
self.next_seed_index = 0 # Only used for sequential strategy