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.

188 Zeilen
7.4KB

  1. from icosagon.batch import PredictionsBatch, \
  2. FlatPredictions, \
  3. flatten_predictions, \
  4. BatchIndices, \
  5. gather_batch_indices
  6. from icosagon.declayer import Predictions, \
  7. RelationPredictions, \
  8. RelationFamilyPredictions
  9. from icosagon.trainprep import prepare_training, \
  10. TrainValTest
  11. from icosagon.data import Data
  12. import torch
  13. import pytest
  14. def test_flat_predictions_01():
  15. pred = FlatPredictions(torch.tensor([0, 1, 0, 1]),
  16. torch.tensor([1, 0, 1, 0]), 'train')
  17. assert torch.all(pred.predictions == torch.tensor([0, 1, 0, 1]))
  18. assert torch.all(pred.truth == torch.tensor([1, 0, 1, 0]))
  19. assert pred.part_type == 'train'
  20. def test_flatten_predictions_01():
  21. rel_pred = RelationPredictions(
  22. TrainValTest(torch.tensor([1, 0, 1, 0, 1], dtype=torch.float32), torch.zeros(0), torch.zeros(0)),
  23. TrainValTest(torch.tensor([1, 0, 1, 0, 1], dtype=torch.float32), torch.zeros(0), torch.zeros(0)),
  24. TrainValTest(torch.zeros(0), torch.zeros(0), torch.zeros(0)),
  25. TrainValTest(torch.zeros(0), torch.zeros(0), torch.zeros(0))
  26. )
  27. fam_pred = RelationFamilyPredictions([ rel_pred ])
  28. pred = Predictions([ fam_pred ])
  29. pred_flat = flatten_predictions(pred, part_type='train')
  30. assert torch.all(pred_flat.predictions == \
  31. torch.tensor([1, 0, 1, 0, 1, 1, 0, 1, 0, 1], dtype=torch.float32))
  32. assert torch.all(pred_flat.truth == \
  33. torch.tensor([1, 1, 1, 1, 1, 0, 0, 0, 0, 0], dtype=torch.float32))
  34. assert pred_flat.part_type == 'train'
  35. def test_flatten_predictions_02():
  36. rel_pred = RelationPredictions(
  37. TrainValTest(torch.tensor([1, 0, 1, 0, 1], dtype=torch.float32), torch.zeros(0), torch.zeros(0)),
  38. TrainValTest(torch.tensor([1, 0, 1, 0, 1], dtype=torch.float32), torch.zeros(0), torch.zeros(0)),
  39. TrainValTest(torch.zeros(0), torch.zeros(0), torch.zeros(0)),
  40. TrainValTest(torch.zeros(0), torch.zeros(0), torch.zeros(0))
  41. )
  42. fam_pred = RelationFamilyPredictions([ rel_pred ])
  43. pred = Predictions([ fam_pred ])
  44. pred_flat = flatten_predictions(pred, part_type='val')
  45. assert len(pred_flat.predictions) == 0
  46. assert len(pred_flat.truth) == 0
  47. assert pred_flat.part_type == 'val'
  48. def test_flatten_predictions_03():
  49. rel_pred = RelationPredictions(
  50. TrainValTest(torch.tensor([1, 0, 1, 0, 1], dtype=torch.float32), torch.zeros(0), torch.zeros(0)),
  51. TrainValTest(torch.tensor([1, 0, 1, 0, 1], dtype=torch.float32), torch.zeros(0), torch.zeros(0)),
  52. TrainValTest(torch.zeros(0), torch.zeros(0), torch.zeros(0)),
  53. TrainValTest(torch.zeros(0), torch.zeros(0), torch.zeros(0))
  54. )
  55. fam_pred = RelationFamilyPredictions([ rel_pred ])
  56. pred = Predictions([ fam_pred ])
  57. pred_flat = flatten_predictions(pred, part_type='test')
  58. assert len(pred_flat.predictions) == 0
  59. assert len(pred_flat.truth) == 0
  60. assert pred_flat.part_type == 'test'
  61. def test_flatten_predictions_04():
  62. rel_pred = RelationPredictions(
  63. TrainValTest(torch.tensor([1, 0, 1, 0, 1], dtype=torch.float32), torch.zeros(0), torch.zeros(0)),
  64. TrainValTest(torch.tensor([1, 0, 1, 0, 1], dtype=torch.float32), torch.zeros(0), torch.zeros(0)),
  65. TrainValTest(torch.zeros(0), torch.zeros(0), torch.zeros(0)),
  66. TrainValTest(torch.zeros(0), torch.zeros(0), torch.zeros(0))
  67. )
  68. fam_pred = RelationFamilyPredictions([ rel_pred ])
  69. pred = Predictions([ fam_pred ])
  70. with pytest.raises(TypeError):
  71. pred_flat = flatten_predictions(1, part_type='test')
  72. with pytest.raises(ValueError):
  73. pred_flat = flatten_predictions(pred, part_type='x')
  74. def test_flatten_predictions_05():
  75. x = torch.rand(5000)
  76. y = torch.cat([ x, x ])
  77. z = torch.cat([ torch.ones(5000), torch.zeros(5000) ])
  78. rel_pred = RelationPredictions(
  79. TrainValTest(x, torch.zeros(0), torch.zeros(0)),
  80. TrainValTest(x, torch.zeros(0), torch.zeros(0)),
  81. TrainValTest(torch.zeros(0), torch.zeros(0), torch.zeros(0)),
  82. TrainValTest(torch.zeros(0), torch.zeros(0), torch.zeros(0))
  83. )
  84. fam_pred = RelationFamilyPredictions([ rel_pred ])
  85. pred = Predictions([ fam_pred ])
  86. for _ in range(10):
  87. pred_flat = flatten_predictions(pred, part_type='train')
  88. assert torch.all(pred_flat.predictions == y)
  89. assert torch.all(pred_flat.truth == z)
  90. assert pred_flat.part_type == 'train'
  91. def test_batch_indices_01():
  92. indices = BatchIndices(torch.tensor([0, 1, 2, 3, 4]), 'train')
  93. assert torch.all(indices.indices == torch.tensor([0, 1, 2, 3, 4]))
  94. assert indices.part_type == 'train'
  95. def test_gather_batch_indices_01():
  96. rel_pred = RelationPredictions(
  97. TrainValTest(torch.tensor([1, 0, 1, 0, 1], dtype=torch.float32), torch.zeros(0), torch.zeros(0)),
  98. TrainValTest(torch.tensor([1, 0, 1, 0, 1], dtype=torch.float32), torch.zeros(0), torch.zeros(0)),
  99. TrainValTest(torch.zeros(0), torch.zeros(0), torch.zeros(0)),
  100. TrainValTest(torch.zeros(0), torch.zeros(0), torch.zeros(0))
  101. )
  102. fam_pred = RelationFamilyPredictions([ rel_pred ])
  103. pred = Predictions([ fam_pred ])
  104. pred_flat = flatten_predictions(pred, part_type='train')
  105. indices = BatchIndices(torch.tensor([0, 2, 4, 5, 7, 9]), 'train')
  106. (input, target) = gather_batch_indices(pred_flat, indices)
  107. assert torch.all(input == \
  108. torch.tensor([1, 1, 1, 1, 1, 1], dtype=torch.float32))
  109. assert torch.all(target == \
  110. torch.tensor([1, 1, 1, 0, 0, 0], dtype=torch.float32))
  111. def test_predictions_batch_01():
  112. d = Data()
  113. d.add_node_type('Dummy', 5)
  114. fam = d.add_relation_family('Dummy-Dummy', 0, 0, False)
  115. fam.add_relation_type('Dummy Rel', torch.tensor([
  116. [0, 1, 0, 0, 0],
  117. [1, 0, 0, 0, 0],
  118. [0, 0, 0, 1, 0],
  119. [0, 0, 0, 0, 1],
  120. [0, 1, 0, 0, 0]
  121. ], dtype=torch.float32))
  122. prep_d = prepare_training(d, TrainValTest(1., 0., 0.))
  123. assert len(prep_d.relation_families) == 1
  124. assert len(prep_d.relation_families[0].relation_types) == 1
  125. assert len(prep_d.relation_families[0].relation_types[0].edges_pos.train) == 5
  126. assert len(prep_d.relation_families[0].relation_types[0].edges_neg.train) == 5
  127. assert len(prep_d.relation_families[0].relation_types[0].edges_pos.val) == 0
  128. assert len(prep_d.relation_families[0].relation_types[0].edges_pos.test) == 0
  129. rel_pred = RelationPredictions(
  130. TrainValTest(torch.tensor([1, 0, 1, 0, 1], dtype=torch.float32), torch.zeros(0), torch.zeros(0)),
  131. TrainValTest(torch.tensor([1, 0, 1, 0, 1], dtype=torch.float32), torch.zeros(0), torch.zeros(0)),
  132. TrainValTest(torch.zeros(0), torch.zeros(0), torch.zeros(0)),
  133. TrainValTest(torch.zeros(0), torch.zeros(0), torch.zeros(0))
  134. )
  135. fam_pred = RelationFamilyPredictions([ rel_pred ])
  136. pred = Predictions([ fam_pred ])
  137. pred_flat = flatten_predictions(pred, part_type='train')
  138. batch = PredictionsBatch(prep_d, part_type='train', batch_size=1)
  139. count = 0
  140. lst = []
  141. for indices in batch:
  142. (input, target) = gather_batch_indices(pred_flat, indices)
  143. assert len(input) == 1
  144. assert len(target) == 1
  145. lst.append((input[0], target[0]))
  146. count += 1
  147. assert lst == [ (1, 1), (0, 1), (1, 1), (0, 1), (1, 1),
  148. (1, 0), (0, 0), (1, 0), (0, 0), (1, 0) ]
  149. assert count == 10