IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an email to s dot adaszewski at gmail dot com. User accounts are meant only to report issues and/or generate pull requests. This is a purpose-specific Git hosting for ADARED projects. Thank you for your understanding!
Du kannst nicht mehr als 25 Themen auswählen Themen müssen entweder mit einem Buchstaben oder einer Ziffer beginnen. Sie können Bindestriche („-“) enthalten und bis zu 35 Zeichen lang sein.

45 Zeilen
1.7KB

  1. import torch
  2. from icosagon.trainprep import PreparedData
  3. from icosagon.declayer import Predictions
  4. class CrossEntropyLoss(torch.nn.Module):
  5. def __init__(self, data: PreparedData, partition_type: str = 'train',
  6. reduction: str = 'sum', **kwargs) -> None:
  7. super().__init__(**kwargs)
  8. if not isinstance(data, PreparedData):
  9. raise TypeError('data must be an instance of PreparedData')
  10. if partition_type not in ['train', 'val', 'test']:
  11. raise ValueError('partition_type must be set to train, val or test')
  12. if reduction not in ['sum', 'mean']:
  13. raise ValueError('reduction must be set to sum or mean')
  14. self.data = data
  15. self.partition_type = partition_type
  16. self.reduction = reduction
  17. def forward(self, pred: Predictions) -> torch.Tensor:
  18. input = []
  19. target = []
  20. for fam in pred.relation_families:
  21. for rel in fam.relation_types:
  22. for edge_type in ['edges_pos', 'edges_back_pos']:
  23. x = getattr(getattr(rel, edge_type), self.partition_type)
  24. assert len(x.shape) == 1
  25. input.append(x)
  26. target.append(torch.ones_like(x))
  27. for edge_type in ['edges_neg', 'edges_back_neg']:
  28. x = getattr(getattr(rel, edge_type), self.partition_type)
  29. assert len(x.shape) == 1
  30. input.append(x)
  31. target.append(torch.zeros_like(x))
  32. input = torch.cat(input, dim=0)
  33. target = torch.cat(target, dim=0)
  34. res = torch.nn.functional.binary_cross_entropy(input, target,
  35. reduction=self.reduction)
  36. return res