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.

92 Zeilen
4.0KB

  1. from icosagon.trainprep import PreparedData, \
  2. PreparedRelationFamily, \
  3. PreparedRelationType, \
  4. _empty_edge_list_tvt
  5. class BatchedData(PreparedData):
  6. def __init__(self, *args, **kwargs):
  7. super().__init__(*args, **kwargs)
  8. def batched_data_skeleton(data: PreparedData) -> BatchedData:
  9. if not isinstance(data, PreparedData):
  10. raise TypeError('data must be an instance of PreparedData')
  11. fam_skels = []
  12. for fam in data.relation_families:
  13. rel_types_skel = []
  14. for rel in fam.relation_types:
  15. rel_skel = PreparedRelationType(rel.name,
  16. rel.node_type_row, rel.node_type_column,
  17. rel.adjacency_matrix, rel.adjacency_matrix_backward,
  18. _empty_edge_list_tvt(), _empty_edge_list_tvt(),
  19. _empty_edge_list_tvt(), _empty_edge_list_tvt())
  20. rel_types_skel.append(rel_skel)
  21. fam_skels.append(PreparedRelationFamily(fam.data, fam.name,
  22. fam.node_type_row, fam.node_type_column,
  23. fam.is_symmetric, fam.decoder_class,
  24. rel_types_skel))
  25. return BatchedData(data.node_types, fam_skels)
  26. class DataBatcher(object):
  27. def __init__(self, data: PreparedData, batch_size: int) -> None:
  28. self._check_params(data, batch_size)
  29. self.data = data
  30. self.batch_size = batch_size
  31. # def batched_data_iter(self, fam_idx: int, rel_idx: int,
  32. # part_type: str) -> BatchedData:
  33. #
  34. # rel = self.data.relation_families[fam_idx].relation_types[rel_idx]
  35. #
  36. # edges = getattr(rel.edges_pos, part_type)
  37. # for m in range(0, len(edges), self.batch_size):
  38. # batched_data = batched_data_skeleton(self.data)
  39. # setattr(batched_data.relation_families[fam_idx].relation_types[rel_idx].edges_pos,
  40. # part_type, edges[m : m + self.batch_size])
  41. # yield batched_data
  42. #
  43. # edges = getattr(rel.edges_neg, part_type)
  44. # for m in range(0, len(edges), self.batch_size):
  45. # batched_data = batched_data_skeleton(self.data)
  46. # setattr(batched_data.relation_families[fam_idx].relation_types[rel_idx].edges_neg,
  47. # part_type, edges[m : m + self.batch_size])
  48. # yield batched_data
  49. #
  50. # edges = getattr(rel.edges_pos_back, part_type)
  51. # for m in range(0, len(edges), self.batch_size):
  52. # batched_data = batched_data_skeleton(self.data)
  53. # setattr(batched_data.relation_families[i].relation_types[k].edges_pos_back,
  54. # part_type, edges[m : m + self.batch_size])
  55. # yield batched_data
  56. #
  57. # edges = getattr(rel.edges_neg_back, part_type)
  58. # for m in range(0, len(), self.batch_size):
  59. # batched_data = batched_data_skeleton(self.data)
  60. # setattr(batched_data.relation_families[i].relation_types[k].edges_neg_back,
  61. # edges[m : m + self.batch_size])
  62. # yield batched_data
  63. def __iter__(self) -> BatchedData:
  64. for i, fam in enumerate(self.data.relation_families):
  65. for k, rel in enumerate(fam.relation_types):
  66. for edge_type in ['edges_pos', 'edges_neg', 'edges_back_pos', 'edges_back_neg']:
  67. for part_type in ['train', 'val', 'test']:
  68. edges = getattr(getattr(rel, edge_type), part_type)
  69. for m in range(0, len(edges), self.batch_size):
  70. batched_data = batched_data_skeleton(self.data)
  71. setattr(getattr(batched_data.relation_families[i].relation_types[k],
  72. edge_type), part_type, edges[m : m + self.batch_size])
  73. yield batched_data
  74. @staticmethod
  75. def _check_params(data, batch_size):
  76. if not isinstance(data, PreparedData):
  77. raise TypeError('data must be an instance of PreparedData')
  78. if not isinstance(batch_size, int):
  79. raise TypeError('batch_size must be an int')