(self, demo, n_envs, *, move_threshold=0.1, nudge=100, window=None)
| 183 | finishes to march max_starting_point backward.""" |
| 184 | |
| 185 | def __init__(self, demo, n_envs, *, move_threshold=0.1, nudge=100, window=None): |
| 186 | self.n = len(demo["actions"]) |
| 187 | self.n_envs = n_envs |
| 188 | self.move_threshold = move_threshold |
| 189 | self.nudge = nudge |
| 190 | # window = the span of staggered starting points (atari-reset nrstartsteps). |
| 191 | # The move target is move_threshold*window of cumulative success mass. |
| 192 | self.window = window or max(n_envs, 32) |
| 193 | self.max_starting_point = self.n - 1 |
| 194 | self.max_max = self.n - 1 |
| 195 | # latest success-rate per starting-point index |
| 196 | self.success = np.zeros(self.n + 1, dtype=np.float64) |
| 197 | |
| 198 | def assign(self, envs): |
| 199 | """Stagger envs across a window below max_starting_point.""" |
nothing calls this directly
no outgoing calls
no test coverage detected