1. 1
2. 2
3. 3
4. 4
5. 5
6. 6
7. 7
8. 8
9. 9
10. 10
11. 11
12. 12
13. 13
14. 14
15. 15
16. 16
17. 17
18. 18
19. 19
20. 20
21. 21
22. 22
23. 23
24. 24
25. 25
26. 26
27. 27
28. 28
29. 29
30. 30
31. 31
32. 32
33. 33
34. 34
35. 35
36. 36
37. 37
38. 38
39. 39
40. 40
41. 41
42. 42
43. 43
44. 44
45. 45
46. 46
47. 47
48. 48
49. 49
50. 50
51. 51
52. 52
53. 53
54. 54
55. 55
56. 56
57. 57
58. 58
59. 59
60. 60
61. 61
62. 62
63. 63
64. 64
65. 65
66. 66
67. 67
68. 68
69. 69
70. 70
71. 71
72. 72
73. 73
74. 74
75. 75
76. 76
77. 77
78. 78
79. 79
80. 80
81. 81
82. 82

from argparse import ArgumentParser

import torch
from torch import nn
from torch.utils.data import DataLoader

from dataset import Dataset
from model import Model


parser = ArgumentParser()
parser.add_argument('-d', '--device', help='device to train with', default='cpu')
parser.add_argument('-i', '--input', help='training data file', default='data')
parser.add_argument('-e', '--epochs', help='number of epochs to train for', default=100)
parser.add_argument('-s', '--seq-size', help='sequence size', default=32)
parser.add_argument('-b', '--batch-size', help='size of each training batch', default=256)
parser.add_argument('-m', '--embedding-dim', help='size of the embedding', default=512)
parser.add_argument('-l', '--lstm-size', help='size of the LSTM hidden state', default=512)
parser.add_argument('-a', '--layers', help='number of LSTM layers', default=3)
parser.add_argument('-r', '--dropout', help='how much dropout to apply', default=0.2)
parser.add_argument('-n', '--max-norm', help='maximum norm for gradient clipping', default=5)
args = parser.parse_args()


# Prepare dataloader
dataset = Dataset(args.input, args.seq_size)
dataloader = DataLoader(dataset, batch_size=args.batch_size)
print(len(dataloader))


# Prepare model
model = Model(dataset, args.embedding_dim, args.lstm_size, args.layers, args.dropout).to(args.device)
print(model)


loss_fn = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=1e-3)


for t in range(args.epochs):
    state_h, state_c = model.zero_state(args.batch_size)
    state_h = state_h.to(args.device)
    state_c = state_c.to(args.device)
    
    iteration = 0

    for batch, (X, y) in enumerate(dataloader):
        model.train()
        
        iteration += 1

        optimizer.zero_grad()

        X = torch.tensor(X).to(args.device)
        y = torch.tensor(y).to(args.device)

        # Compute prediction error
        logits, (state_h, state_c) = model(X, (state_h, state_c))
        loss = loss_fn(logits.transpose(1, 2), y)

        loss_value = loss.item()
        
        # Backpropogation
        loss.backward()

        state_h = state_h.detach()
        state_c = state_c.detach()

        _ = torch.nn.utils.clip_grad_norm_(
            model.parameters(), args.max_norm)

        optimizer.step()

        if iteration % 1 == 0:
            print('Epoch: {}/{}'.format(t, args.epochs),
                  'Iteration: {}'.format(iteration),
                  'Loss: {}'.format(loss_value))

        if iteration % 1000 == 0:
            predict(args.device, dataset, model, 100, 2)
            torch.save(net.state_dict(),
                       'checkpoint/model-{}.pth'.format(iteration))