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.

134 lignes
3.8KB

  1. from icosagon.data import Data
  2. from icosagon.trainprep import prepare_training, \
  3. TrainValTest
  4. from icosagon.model import Model
  5. from icosagon.trainloop import TrainLoop
  6. import torch
  7. import pytest
  8. import pdb
  9. import time
  10. def test_train_loop_01():
  11. d = Data()
  12. d.add_node_type('Dummy', 10)
  13. fam = d.add_relation_family('Dummy-Dummy', 0, 0, False)
  14. fam.add_relation_type('Dummy Rel', torch.rand(10, 10).round())
  15. prep_d = prepare_training(d, TrainValTest(.8, .1, .1))
  16. m = Model(prep_d)
  17. loop = TrainLoop(m)
  18. assert loop.model == m
  19. assert loop.lr == 0.001
  20. assert loop.loss == torch.nn.functional.binary_cross_entropy_with_logits
  21. assert loop.batch_size == 100
  22. def test_train_loop_02():
  23. d = Data()
  24. d.add_node_type('Dummy', 10)
  25. fam = d.add_relation_family('Dummy-Dummy', 0, 0, False)
  26. fam.add_relation_type('Dummy Rel', torch.rand(10, 10).round())
  27. prep_d = prepare_training(d, TrainValTest(.8, .1, .1))
  28. m = Model(prep_d)
  29. for prm in m.parameters():
  30. print(prm.shape, prm.is_leaf, prm.requires_grad)
  31. loop = TrainLoop(m)
  32. loop.run_epoch()
  33. for prm in m.parameters():
  34. print(prm.shape, prm.is_leaf, prm.requires_grad)
  35. def test_train_loop_03():
  36. # pdb.set_trace()
  37. if torch.cuda.device_count() == 0:
  38. pytest.skip('CUDA required for this test')
  39. adj_mat = torch.rand(10, 10).round()
  40. dev = torch.device('cuda:0')
  41. adj_mat = adj_mat.to(dev)
  42. d = Data()
  43. d.add_node_type('Dummy', 10)
  44. fam = d.add_relation_family('Dummy-Dummy', 0, 0, False)
  45. fam.add_relation_type('Dummy Rel', adj_mat)
  46. prep_d = prepare_training(d, TrainValTest(.8, .1, .1))
  47. # pdb.set_trace()
  48. m = Model(prep_d)
  49. m = m.to(dev)
  50. print(list(m.parameters()))
  51. for prm in m.parameters():
  52. assert prm.device == dev
  53. loop = TrainLoop(m)
  54. loop.run_epoch()
  55. def test_timing_01():
  56. adj_mat = (torch.rand(2000, 2000) < .001).to(torch.float32).to_sparse()
  57. rep = torch.eye(2000).requires_grad_(True)
  58. t = time.time()
  59. for _ in range(1300):
  60. _ = torch.sparse.mm(adj_mat, rep)
  61. print('Elapsed:', time.time() - t)
  62. def test_timing_02():
  63. adj_mat = (torch.rand(2000, 2000) < .001).to(torch.float32)
  64. adj_mat_batch = [adj_mat.view(1, 2000, 2000)] * 1300
  65. adj_mat_batch = torch.cat(adj_mat_batch)
  66. rep = torch.eye(2000).requires_grad_(True)
  67. t = time.time()
  68. res = torch.matmul(adj_mat_batch, rep)
  69. print('Elapsed:', time.time() - t)
  70. print('res.shape:', res.shape)
  71. def test_timing_03():
  72. adj_mat = (torch.rand(2000, 2000) < .001).to(torch.float32)
  73. adj_mat_batch = [adj_mat.view(1, 2000, 2000).to_sparse()] * 1300
  74. adj_mat_batch = torch.cat(adj_mat_batch)
  75. rep = torch.eye(2000).requires_grad_(True)
  76. rep_batch = [rep.view(1, 2000, 2000)] * 1300
  77. rep_batch = torch.cat(rep_batch)
  78. t = time.time()
  79. with pytest.raises(RuntimeError):
  80. _ = torch.bmm(adj_mat_batch, rep)
  81. print('Elapsed:', time.time() - t)
  82. def test_timing_04():
  83. adj_mat = (torch.rand(2000, 2000) < .0001).to(torch.float32).to_sparse()
  84. rep = torch.eye(2000).requires_grad_(True)
  85. t = time.time()
  86. for _ in range(1300):
  87. _ = torch.sparse.mm(adj_mat, rep)
  88. print('Elapsed:', time.time() - t)
  89. def test_timing_05():
  90. if torch.cuda.device_count() == 0:
  91. pytest.skip('Test requires CUDA')
  92. dev = torch.device('cuda:0')
  93. adj_mat = (torch.rand(2000, 2000) < .001).to(torch.float32).to_sparse().to(dev)
  94. rep = torch.eye(2000).requires_grad_(True).to(dev)
  95. t = time.time()
  96. for _ in range(1300):
  97. _ = torch.sparse.mm(adj_mat, rep)
  98. torch.cuda.synchronize()
  99. print('Elapsed:', time.time() - t)