(self, x, min, max, set_size: int)
| 347 | return self._phi(x, self.reward_support.min, self.reward_support.max, self.reward_support.size) |
| 348 | |
| 349 | def _phi(self, x, min, max, set_size: int): |
| 350 | delta = self.value_support.delta |
| 351 | |
| 352 | x.clamp_(min, max) |
| 353 | x_low = x.floor() |
| 354 | x_high = x.ceil() |
| 355 | p_high = x - x_low |
| 356 | p_low = 1 - p_high |
| 357 | |
| 358 | target = torch.zeros(x.shape[0], x.shape[1], set_size).to(x.device) |
| 359 | x_high_idx, x_low_idx = x_high - min / delta, x_low - min / delta |
| 360 | target.scatter_(2, x_high_idx.long().unsqueeze(-1), p_high.unsqueeze(-1)) |
| 361 | target.scatter_(2, x_low_idx.long().unsqueeze(-1), p_low.unsqueeze(-1)) |
| 362 | return target |
| 363 | |
| 364 | def get_hparams(self): |
| 365 | # get all the hyper-parameters |
no outgoing calls
no test coverage detected