fix dynamic learning rate

Author: itsmystyle

src/base_predictor.py

@@ -110,7 +110,7 @@ def load(self, path):
  self.optimizer.load_state_dict(torch.load(path)['optimizer'])
  self.epoch = torch.load(path)['epoch']
 
- for g in self.optimizer:
+ for g in self.optimizer.param_groups:
  if g['lr'] != self.learning_rate:
  g['lr'] = self.learning_rate