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.

normalize.py 4.3KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145
  1. #
  2. # Copyright (C) Stanislaw Adaszewski, 2020
  3. # License: GPLv3
  4. #
  5. import numpy as np
  6. import scipy.sparse as sp
  7. import torch
  8. def _check_tensor(adj_mat):
  9. if not isinstance(adj_mat, torch.Tensor):
  10. raise ValueError('adj_mat must be a torch.Tensor')
  11. def _check_sparse(adj_mat):
  12. if not adj_mat.is_sparse:
  13. raise ValueError('adj_mat must be sparse')
  14. def _check_dense(adj_mat):
  15. if adj_mat.is_sparse:
  16. raise ValueError('adj_mat must be dense')
  17. def _check_square(adj_mat):
  18. if len(adj_mat.shape) != 2 or \
  19. adj_mat.shape[0] != adj_mat.shape[1]:
  20. raise ValueError('adj_mat must be a square matrix')
  21. def _check_2d(adj_mat):
  22. if len(adj_mat.shape) != 2:
  23. raise ValueError('adj_mat must be a square matrix')
  24. def _sparse_coo_tensor(indices, values, size):
  25. ctor = { torch.float32: torch.sparse.FloatTensor,
  26. torch.float32: torch.sparse.DoubleTensor,
  27. torch.uint8: torch.sparse.ByteTensor,
  28. torch.long: torch.sparse.LongTensor,
  29. torch.int: torch.sparse.IntTensor,
  30. torch.short: torch.sparse.ShortTensor,
  31. torch.bool: torch.sparse.ByteTensor }[values.dtype]
  32. return ctor(indices, values, size)
  33. def add_eye_sparse(adj_mat: torch.Tensor) -> torch.Tensor:
  34. _check_tensor(adj_mat)
  35. _check_sparse(adj_mat)
  36. _check_square(adj_mat)
  37. adj_mat = adj_mat.coalesce()
  38. indices = adj_mat.indices()
  39. values = adj_mat.values()
  40. eye_indices = torch.arange(adj_mat.shape[0], dtype=indices.dtype,
  41. device=adj_mat.device).view(1, -1)
  42. eye_indices = torch.cat((eye_indices, eye_indices), 0)
  43. eye_values = torch.ones(adj_mat.shape[0], dtype=values.dtype,
  44. device=adj_mat.device)
  45. indices = torch.cat((indices, eye_indices), 1)
  46. values = torch.cat((values, eye_values), 0)
  47. adj_mat = _sparse_coo_tensor(indices, values, adj_mat.shape)
  48. return adj_mat
  49. def norm_adj_mat_one_node_type_sparse(adj_mat: torch.Tensor) -> torch.Tensor:
  50. _check_tensor(adj_mat)
  51. _check_sparse(adj_mat)
  52. _check_square(adj_mat)
  53. adj_mat = add_eye_sparse(adj_mat)
  54. adj_mat = norm_adj_mat_two_node_types_sparse(adj_mat)
  55. return adj_mat
  56. def norm_adj_mat_one_node_type_dense(adj_mat: torch.Tensor) -> torch.Tensor:
  57. _check_tensor(adj_mat)
  58. _check_dense(adj_mat)
  59. _check_square(adj_mat)
  60. adj_mat = adj_mat + torch.eye(adj_mat.shape[0], dtype=adj_mat.dtype,
  61. device=adj_mat.device)
  62. adj_mat = norm_adj_mat_two_node_types_dense(adj_mat)
  63. return adj_mat
  64. def norm_adj_mat_one_node_type(adj_mat: torch.Tensor) -> torch.Tensor:
  65. _check_tensor(adj_mat)
  66. _check_square(adj_mat)
  67. if adj_mat.is_sparse:
  68. return norm_adj_mat_one_node_type_sparse(adj_mat)
  69. else:
  70. return norm_adj_mat_one_node_type_dense(adj_mat)
  71. def norm_adj_mat_two_node_types_sparse(adj_mat: torch.Tensor) -> torch.Tensor:
  72. _check_tensor(adj_mat)
  73. _check_sparse(adj_mat)
  74. _check_2d(adj_mat)
  75. adj_mat = adj_mat.coalesce()
  76. indices = adj_mat.indices()
  77. values = adj_mat.values()
  78. degrees_row = torch.zeros(adj_mat.shape[0], device=adj_mat.device)
  79. degrees_row = degrees_row.index_add(0, indices[0], values.to(degrees_row.dtype))
  80. degrees_col = torch.zeros(adj_mat.shape[1], device=adj_mat.device)
  81. degrees_col = degrees_col.index_add(0, indices[1], values.to(degrees_col.dtype))
  82. values = values.to(degrees_row.dtype) / torch.sqrt(degrees_row[indices[0]] * degrees_col[indices[1]])
  83. adj_mat = _sparse_coo_tensor(indices, values, adj_mat.shape)
  84. return adj_mat
  85. def norm_adj_mat_two_node_types_dense(adj_mat: torch.Tensor) -> torch.Tensor:
  86. _check_tensor(adj_mat)
  87. _check_dense(adj_mat)
  88. _check_2d(adj_mat)
  89. degrees_row = adj_mat.sum(1).view(-1, 1).to(torch.float32)
  90. degrees_col = adj_mat.sum(0).view(1, -1).to(torch.float32)
  91. degrees_row = torch.sqrt(degrees_row)
  92. degrees_col = torch.sqrt(degrees_col)
  93. adj_mat = adj_mat.to(degrees_row.dtype) / degrees_row
  94. adj_mat = adj_mat / degrees_col
  95. return adj_mat
  96. def norm_adj_mat_two_node_types(adj_mat: torch.Tensor) -> torch.Tensor:
  97. _check_tensor(adj_mat)
  98. _check_2d(adj_mat)
  99. if adj_mat.is_sparse:
  100. return norm_adj_mat_two_node_types_sparse(adj_mat)
  101. else:
  102. return norm_adj_mat_two_node_types_dense(adj_mat)