Register
Login
Resources
Docs Blog Datasets Glossary Case Studies Tutorials & Webinars
Product
Data Engine LLMs Platform Enterprise
Pricing Explore
Connect to our Discord channel
Resources  
Docs Blog Datasets Glossary Case Studies Tutorials & Webinars
Explore Pricing
Product  
Data Engine LLMs Platform Enterprise
Connect to Dagshub Discord channel
Register
Login

Deci-AI
/
super-gradients
connected to https://github.com/Deci-AI/super-gradients.git

Watch 4
Issues 62 Pull Requests 44 Discussions 0

#657 Segmentation Readme

Merged
Ghost merged 1 commits into Deci-AI:master from deci-ai:feature/SG-594-segmentation_readme
Conversation 0 Files changed 3 Commits 1
File Compare
super-gradients
/
tests
/
unit_tests
/
optimizer_params_override_test.py
View left file View right file
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67 
  1. import unittest
    2. 

  2. 

  3. from super_gradients.training.utils.utils import get_param
    4. 

  4. 

  5. from super_gradients import Trainer
    6. 

  6. from super_gradients.training.dataloaders.dataloaders import classification_test_dataloader
    7. 

  7. from super_gradients.training.metrics import Accuracy, Top5
    8. 

  8. from super_gradients.training.models import ResNet18
    9. 

  9. 

  10. 

  11. class TrainOptimizerParamsOverride(unittest.TestCase):
    12. 

  12.     def test_optimizer_params_partial_override(self):
    13. 

  13.         trainer = Trainer("test_optimizer_params_partial_override")
    14. 

  14.         net = ResNet18(num_classes=5, arch_params={})
    15. 

  15.         train_params = {
    16. 

  16.             "max_epochs": 1,
    17. 

  17.             "lr_updates": [1],
    18. 

  18.             "lr_decay_factor": 0.1,
    19. 

  19.             "lr_mode": "step",
    20. 

  20.             "lr_warmup_epochs": 0,
    21. 

  21.             "initial_lr": 0.1,
    22. 

  22.             "loss": "cross_entropy",
    23. 

  23.             "optimizer": "SGD",
    24. 

  24.             "criterion_params": {},
    25. 

  25.             "optimizer_params": {"momentum": 0.9},
    26. 

  26.             "zero_weight_decay_on_bias_and_bn": True,
    27. 

  27.             "train_metrics_list": [Accuracy(), Top5()],
    28. 

  28.             "valid_metrics_list": [Accuracy(), Top5()],
    29. 

  29.             "metric_to_watch": "Accuracy",
    30. 

  30.             "greater_metric_to_watch_is_better": True,
    31. 

  31.         }
    32. 

  32.         trainer.train(
    33. 

  33.             model=net,
    34. 

  34.             training_params=train_params,
    35. 

  35.             train_loader=classification_test_dataloader(batch_size=10),
    36. 

  36.             valid_loader=classification_test_dataloader(batch_size=10),
    37. 

  37.         )
    38. 

  38.         self.assertTrue(get_param(trainer.training_params.optimizer_params, "weight_decay"), 1e-4)
    39. 

  39.         self.assertTrue(get_param(trainer.training_params.optimizer_params, "momentum"), 0.9)
    40. 

  40. 

  41.     def test_optimizer_params_full_override(self):
    42. 

  42.         trainer = Trainer("test_optimizer_params_full_override")
    43. 

  43.         net = ResNet18(num_classes=5, arch_params={})
    44. 

  44.         train_params = {
    45. 

  45.             "max_epochs": 1,
    46. 

  46.             "lr_updates": [1],
    47. 

  47.             "lr_decay_factor": 0.1,
    48. 

  48.             "lr_mode": "step",
    49. 

  49.             "lr_warmup_epochs": 0,
    50. 

  50.             "initial_lr": 0.1,
    51. 

  51.             "loss": "cross_entropy",
    52. 

  52.             "optimizer": "SGD",
    53. 

  53.             "criterion_params": {},
    54. 

  54.             "zero_weight_decay_on_bias_and_bn": True,
    55. 

  55.             "train_metrics_list": [Accuracy(), Top5()],
    56. 

  56.             "valid_metrics_list": [Accuracy(), Top5()],
    57. 

  57.             "metric_to_watch": "Accuracy",
    58. 

  58.             "greater_metric_to_watch_is_better": True,
    59. 

  59.         }
    60. 

  60.         trainer.train(
    61. 

  61.             model=net,
    62. 

  62.             training_params=train_params,
    63. 

  63.             train_loader=classification_test_dataloader(batch_size=10),
    64. 

  64.             valid_loader=classification_test_dataloader(batch_size=10),
    65. 

  65.         )
    66. 

  66.         self.assertTrue(get_param(trainer.training_params.optimizer_params, "weight_decay"), 1e-4)
    67. 

  67.         self.assertTrue(get_param(trainer.training_params.optimizer_params, "momentum"), 0.9)
    68.
Discard
Prev
Next
Tip!

Press p or to see the previous file or, n or to see the next file