(self, save_name, save_path)
| 266 | 'eval/precision': precision, 'eval/recall': recall, 'eval/F1': F1, 'eval/AUC': AUC} |
| 267 | |
| 268 | def save_model(self, save_name, save_path): |
| 269 | if self.it < 1000000: |
| 270 | return |
| 271 | save_filename = os.path.join(save_path, save_name) |
| 272 | # copy EMA parameters to ema_model for saving with model as temp |
| 273 | self.model.eval() |
| 274 | self.ema.apply_shadow() |
| 275 | ema_model = self.model.state_dict() |
| 276 | self.ema.restore() |
| 277 | self.model.train() |
| 278 | |
| 279 | torch.save({'model': self.model.state_dict(), |
| 280 | 'optimizer': self.optimizer.state_dict(), |
| 281 | 'scheduler': self.scheduler.state_dict(), |
| 282 | 'it': self.it + 1, |
| 283 | 'ema_model': ema_model}, |
| 284 | save_filename) |
| 285 | |
| 286 | self.print_fn(f"model saved: {save_filename}") |
| 287 | |
| 288 | def load_model(self, load_path): |
| 289 | checkpoint = torch.load(load_path) |
no test coverage detected