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 kan inte välja fler än 25 ämnen Ämnen måste starta med en bokstav eller siffra, kan innehålla bindestreck ('-') och vara max 35 tecken långa.

test_trainprep.py 11KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268
  1. #
  2. # Copyright (C) Stanislaw Adaszewski, 2020
  3. # License: GPLv3
  4. #
  5. from icosagon.trainprep import TrainValTest, \
  6. train_val_test_split_edges, \
  7. get_edges_and_degrees, \
  8. prepare_adj_mat, \
  9. prepare_relation_type, \
  10. prep_rel_one_node_type, \
  11. prep_rel_two_node_types_sym, \
  12. prep_rel_two_node_types_asym
  13. import torch
  14. import pytest
  15. import numpy as np
  16. from itertools import chain
  17. from icosagon.data import RelationType
  18. import icosagon.trainprep
  19. def test_train_val_test_split_edges_01():
  20. edges = torch.randint(0, 10, (10, 2))
  21. with pytest.raises(ValueError):
  22. _ = train_val_test_split_edges(edges, TrainValTest(.5, .5, .5))
  23. with pytest.raises(ValueError):
  24. _ = train_val_test_split_edges(edges, TrainValTest(.2, .2, .2))
  25. with pytest.raises(ValueError):
  26. _ = train_val_test_split_edges(edges, None)
  27. with pytest.raises(ValueError):
  28. _ = train_val_test_split_edges(edges, (.8, .1, .1))
  29. with pytest.raises(ValueError):
  30. _ = train_val_test_split_edges(np.random.randint(0, 10, (10, 2)), TrainValTest(.8, .1, .1))
  31. with pytest.raises(ValueError):
  32. _ = train_val_test_split_edges(torch.randint(0, 10, (10, 3)), TrainValTest(.8, .1, .1))
  33. with pytest.raises(ValueError):
  34. _ = train_val_test_split_edges(torch.randint(0, 10, (10, 2, 1)), TrainValTest(.8, .1, .1))
  35. with pytest.raises(ValueError):
  36. _ = train_val_test_split_edges(None, TrainValTest(.8, .2, .2))
  37. res = train_val_test_split_edges(edges, TrainValTest(.8, .1, .1))
  38. assert res.train.shape == (8, 2) and res.val.shape == (1, 2) and \
  39. res.test.shape == (1, 2)
  40. res = train_val_test_split_edges(edges, TrainValTest(.8, .0, .2))
  41. assert res.train.shape == (8, 2) and res.val.shape == (0, 2) and \
  42. res.test.shape == (2, 2)
  43. res = train_val_test_split_edges(edges, TrainValTest(.8, .2, .0))
  44. assert res.train.shape == (8, 2) and res.val.shape == (2, 2) and \
  45. res.test.shape == (0, 2)
  46. res = train_val_test_split_edges(edges, TrainValTest(.0, .5, .5))
  47. assert res.train.shape == (0, 2) and res.val.shape == (5, 2) and \
  48. res.test.shape == (5, 2)
  49. res = train_val_test_split_edges(edges, TrainValTest(.0, .0, 1.))
  50. assert res.train.shape == (0, 2) and res.val.shape == (0, 2) and \
  51. res.test.shape == (10, 2)
  52. res = train_val_test_split_edges(edges, TrainValTest(.0, 1., .0))
  53. assert res.train.shape == (0, 2) and res.val.shape == (10, 2) and \
  54. res.test.shape == (0, 2)
  55. def test_train_val_test_split_edges_02():
  56. edges = torch.randint(0, 30, (30, 2))
  57. ratios = TrainValTest(.8, .1, .1)
  58. res = train_val_test_split_edges(edges, ratios)
  59. edges = [ tuple(a) for a in edges ]
  60. res = [ tuple(a) for a in chain(res.train, res.val, res.test) ]
  61. assert all([ a in edges for a in res ])
  62. def test_get_edges_and_degrees_01():
  63. adj_mat_dense = (torch.rand((10, 10)) > .5)
  64. adj_mat_sparse = adj_mat_dense.to_sparse()
  65. edges_dense, degrees_dense = get_edges_and_degrees(adj_mat_dense)
  66. edges_sparse, degrees_sparse = get_edges_and_degrees(adj_mat_sparse)
  67. assert torch.all(degrees_dense == degrees_sparse)
  68. edges_dense = [ tuple(a) for a in edges_dense ]
  69. edges_sparse = [ tuple(a) for a in edges_dense ]
  70. assert len(edges_dense) == len(edges_sparse)
  71. assert all([ a in edges_dense for a in edges_sparse ])
  72. assert all([ a in edges_sparse for a in edges_dense ])
  73. # assert torch.all(edges_dense == edges_sparse)
  74. def test_prepare_adj_mat_01():
  75. adj_mat = (torch.rand((10, 10)) > .5)
  76. adj_mat = adj_mat.to_sparse()
  77. ratios = TrainValTest(.8, .1, .1)
  78. _ = prepare_adj_mat(adj_mat, ratios)
  79. def test_prepare_adj_mat_02():
  80. adj_mat = (torch.rand((10, 10)) > .5)
  81. adj_mat = adj_mat.to_sparse()
  82. ratios = TrainValTest(.8, .1, .1)
  83. (adj_mat_train, edges_pos, edges_neg) = prepare_adj_mat(adj_mat, ratios)
  84. assert isinstance(adj_mat_train, torch.Tensor)
  85. assert adj_mat_train.is_sparse
  86. assert adj_mat_train.shape == adj_mat.shape
  87. assert adj_mat_train.dtype == adj_mat.dtype
  88. assert isinstance(edges_pos, TrainValTest)
  89. assert isinstance(edges_neg, TrainValTest)
  90. for a in ['train', 'val', 'test']:
  91. for b in [edges_pos, edges_neg]:
  92. edges = getattr(b, a)
  93. assert isinstance(edges, torch.Tensor)
  94. assert len(edges.shape) == 2
  95. assert edges.shape[1] == 2
  96. def test_prepare_relation_type_01():
  97. adj_mat = (torch.rand((10, 10)) > .5).to(torch.float32)
  98. r = RelationType('Test', 0, 0, adj_mat, True)
  99. ratios = TrainValTest(.8, .1, .1)
  100. _ = prepare_relation_type(r, ratios, False)
  101. def test_prep_rel_one_node_type_01():
  102. adj_mat = torch.zeros(100)
  103. perm = torch.randperm(100)
  104. adj_mat[perm[:10]] = 1
  105. adj_mat = adj_mat.view(10, 10)
  106. rel = RelationType('Dummy Relation', 0, 0, adj_mat, None)
  107. ratios = TrainValTest(.8, .1, .1)
  108. prep_rel = prep_rel_one_node_type(rel, ratios)
  109. assert prep_rel.name == rel.name
  110. assert prep_rel.node_type_row == rel.node_type_row
  111. assert prep_rel.node_type_column == rel.node_type_column
  112. assert prep_rel.adjacency_matrix.shape == rel.adjacency_matrix.shape
  113. assert prep_rel.adjacency_matrix_backward is None
  114. assert len(prep_rel.edges_pos.train) == 8
  115. assert len(prep_rel.edges_pos.val) == 1
  116. assert len(prep_rel.edges_pos.test) == 1
  117. assert len(prep_rel.edges_neg.train) == 8
  118. assert len(prep_rel.edges_neg.val) == 1
  119. assert len(prep_rel.edges_neg.test) == 1
  120. assert len(prep_rel.edges_back_pos.train) == 0
  121. assert len(prep_rel.edges_back_pos.val) == 0
  122. assert len(prep_rel.edges_back_pos.test) == 0
  123. assert len(prep_rel.edges_back_neg.train) == 0
  124. assert len(prep_rel.edges_back_neg.val) == 0
  125. assert len(prep_rel.edges_back_neg.test) == 0
  126. def test_prep_rel_two_node_types_sym_01():
  127. adj_mat = torch.zeros(200)
  128. perm = torch.randperm(100)
  129. adj_mat[perm[:10]] = 1
  130. adj_mat = adj_mat.view(10, 20)
  131. rel = RelationType('Dummy Relation', 0, 1, adj_mat, None)
  132. ratios = TrainValTest(.8, .1, .1)
  133. prep_rel = prep_rel_two_node_types_sym(rel, ratios)
  134. assert prep_rel.name == rel.name
  135. assert prep_rel.node_type_row == rel.node_type_row
  136. assert prep_rel.node_type_column == rel.node_type_column
  137. assert prep_rel.adjacency_matrix.shape == rel.adjacency_matrix.shape
  138. assert prep_rel.adjacency_matrix_backward.shape == (20, 10)
  139. assert len(prep_rel.edges_pos.train) == 8
  140. assert len(prep_rel.edges_pos.val) == 1
  141. assert len(prep_rel.edges_pos.test) == 1
  142. assert len(prep_rel.edges_neg.train) == 8
  143. assert len(prep_rel.edges_neg.val) == 1
  144. assert len(prep_rel.edges_neg.test) == 1
  145. assert len(prep_rel.edges_back_pos.train) == 0
  146. assert len(prep_rel.edges_back_pos.val) == 0
  147. assert len(prep_rel.edges_back_pos.test) == 0
  148. assert len(prep_rel.edges_back_neg.train) == 0
  149. assert len(prep_rel.edges_back_neg.val) == 0
  150. assert len(prep_rel.edges_back_neg.test) == 0
  151. def test_prep_rel_two_node_types_asym_01():
  152. adj_mat = torch.zeros(200)
  153. perm = torch.randperm(100)
  154. adj_mat[perm[:10]] = 1
  155. adj_mat = adj_mat.view(10, 20)
  156. adj_mat_back = torch.zeros(200)
  157. perm = torch.randperm(100)
  158. adj_mat_back[perm[:10]] = 1
  159. adj_mat_back = adj_mat_back.view(20, 10)
  160. rel = RelationType('Dummy Relation', 0, 1, adj_mat, adj_mat_back)
  161. ratios = TrainValTest(.8, .1, .1)
  162. prep_rel = prep_rel_two_node_types_asym(rel, ratios)
  163. assert prep_rel.name == rel.name
  164. assert prep_rel.node_type_row == rel.node_type_row
  165. assert prep_rel.node_type_column == rel.node_type_column
  166. assert prep_rel.adjacency_matrix.shape == rel.adjacency_matrix.shape
  167. assert prep_rel.adjacency_matrix_backward.shape == rel.adjacency_matrix_backward.shape
  168. assert len(prep_rel.edges_pos.train) == 8
  169. assert len(prep_rel.edges_pos.val) == 1
  170. assert len(prep_rel.edges_pos.test) == 1
  171. assert len(prep_rel.edges_neg.train) == 8
  172. assert len(prep_rel.edges_neg.val) == 1
  173. assert len(prep_rel.edges_neg.test) == 1
  174. assert len(prep_rel.edges_back_pos.train) == 8
  175. assert len(prep_rel.edges_back_pos.val) == 1
  176. assert len(prep_rel.edges_back_pos.test) == 1
  177. assert len(prep_rel.edges_back_neg.train) == 8
  178. assert len(prep_rel.edges_back_neg.val) == 1
  179. assert len(prep_rel.edges_back_neg.test) == 1
  180. def test_prepare_relation_type_02():
  181. with pytest.raises(ValueError):
  182. prepare_relation_type(None, None, True)
  183. adj_mat = torch.rand(10, 10).round()
  184. rel = RelationType('Dummy Relation', 0, 0, adj_mat, None)
  185. with pytest.raises(ValueError):
  186. prepare_relation_type(rel, None, True)
  187. ratios = TrainValTest(.8, .1, .1)
  188. with pytest.raises(ValueError):
  189. prepare_relation_type(None, ratios, True)
  190. _ = prepare_relation_type(rel, ratios, True)
  191. def test_prepare_relation_type_03(monkeypatch):
  192. a = 0
  193. b = 0
  194. c = 0
  195. def fake_prep_rel_one_node_type(*args, **kwargs):
  196. nonlocal a
  197. a += 1
  198. def fake_prep_rel_two_node_types_sym(*args, **kwargs):
  199. nonlocal b
  200. b += 1
  201. def fake_prep_rel_two_node_types_asym(*args, **kwargs):
  202. nonlocal c
  203. c += 1
  204. monkeypatch.setattr(icosagon.trainprep, 'prep_rel_one_node_type',
  205. fake_prep_rel_one_node_type)
  206. monkeypatch.setattr(icosagon.trainprep, 'prep_rel_two_node_types_sym',
  207. fake_prep_rel_two_node_types_sym)
  208. monkeypatch.setattr(icosagon.trainprep, 'prep_rel_two_node_types_asym',
  209. fake_prep_rel_two_node_types_asym)
  210. ratios = TrainValTest(.8, .1, .1)
  211. rel = RelationType('Dummy Relation', 0, 0, None, None)
  212. prepare_relation_type(rel, ratios, False)
  213. assert a == 1
  214. rel = RelationType('Dummy Relation', 0, 0, None, None)
  215. prepare_relation_type(rel, ratios, True)
  216. assert a == 2
  217. rel = RelationType('Dummy Relation', 0, 1, None, None)
  218. prepare_relation_type(rel, ratios, True)
  219. assert b == 1
  220. rel = RelationType('Dummy Relation', 0, 1, None, None)
  221. prepare_relation_type(rel, ratios, False)
  222. assert c == 1
  223. assert a == 2 and b == 1 and c == 1
  224. # def prepare_relation(r, ratios):
  225. # adj_mat = r.adjacency_matrix
  226. # adj_mat_train, edges_pos, edges_neg = prepare_adj_mat(adj_mat)
  227. #
  228. # if r.node_type_row == r.node_type_column:
  229. # adj_mat_train = norm_adj_mat_one_node_type(adj_mat_train)
  230. # else:
  231. # adj_mat_train = norm_adj_mat_two_node_types(adj_mat_train)
  232. #
  233. # return PreparedRelation(r.name, r.node_type_row, r.node_type_column,
  234. # adj_mat_train, edges_pos, edges_neg)