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!
Vous ne pouvez pas sélectionner plus de 25 sujets Les noms de sujets doivent commencer par une lettre ou un nombre, peuvent contenir des tirets ('-') et peuvent comporter jusqu'à 35 caractères.

75 lignes
2.8KB

  1. from .layer import Layer
  2. import torch
  3. from ..convolve import DropoutGraphConvActivation
  4. from ..data import Data
  5. from typing import List, \
  6. Union, \
  7. Callable
  8. from collections import defaultdict
  9. class DecagonLayer(Layer):
  10. def __init__(self,
  11. data: Data,
  12. previous_layer: Layer,
  13. output_dim: Union[int, List[int]],
  14. keep_prob: float = 1.,
  15. rel_activation: Callable[[torch.Tensor], torch.Tensor] = lambda x: x,
  16. layer_activation: Callable[[torch.Tensor], torch.Tensor] = torch.nn.functional.relu,
  17. **kwargs):
  18. if not isinstance(output_dim, list):
  19. output_dim = [ output_dim ] * len(data.node_types)
  20. super().__init__(output_dim, is_sparse=False, **kwargs)
  21. self.data = data
  22. self.previous_layer = previous_layer
  23. self.input_dim = previous_layer.output_dim
  24. self.keep_prob = keep_prob
  25. self.rel_activation = rel_activation
  26. self.layer_activation = layer_activation
  27. self.next_layer_repr = None
  28. self.build()
  29. def build(self):
  30. self.next_layer_repr = defaultdict(list)
  31. for (nt_row, nt_col), relation_types in self.data.relation_types.items():
  32. row_convs = []
  33. col_convs = []
  34. for rel in relation_types:
  35. conv = DropoutGraphConvActivation(self.input_dim[nt_col],
  36. self.output_dim[nt_row], rel.adjacency_matrix,
  37. self.keep_prob, self.rel_activation)
  38. row_convs.append(conv)
  39. if nt_row == nt_col:
  40. continue
  41. conv = DropoutGraphConvActivation(self.input_dim[nt_row],
  42. self.output_dim[nt_col], rel.adjacency_matrix.transpose(0, 1),
  43. self.keep_prob, self.rel_activation)
  44. col_convs.append(conv)
  45. self.next_layer_repr[nt_row].append((row_convs, nt_col))
  46. if nt_row == nt_col:
  47. continue
  48. self.next_layer_repr[nt_col].append((col_convs, nt_row))
  49. def __call__(self):
  50. prev_layer_repr = self.previous_layer()
  51. next_layer_repr = [ [] for _ in range(len(self.data.node_types)) ]
  52. print('next_layer_repr:', next_layer_repr)
  53. for i in range(len(self.data.node_types)):
  54. for convs, neighbor_type in self.next_layer_repr[i]:
  55. convs = [ conv(prev_layer_repr[neighbor_type]) \
  56. for conv in convs ]
  57. convs = sum(convs)
  58. convs = torch.nn.functional.normalize(convs, p=2, dim=1)
  59. next_layer_repr[i].append(convs)
  60. next_layer_repr[i] = sum(next_layer_repr[i])
  61. next_layer_repr[i] = self.layer_activation(next_layer_repr[i])
  62. print('next_layer_repr:', next_layer_repr)
  63. return next_layer_repr