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
83. 83
84. 84
85. 85
86. 86
87. 87
88. 88
89. 89
90. 90
91. 91
92. 92
93. 93
94. 94
95. 95
96. 96
97. 97
98. 98
99. 99
100. 100
101. 101
102. 102
103. 103
104. 104
105. 105
106. 106
107. 107
108. 108
109. 109
110. 110
111. 111
112. 112
113. 113
114. 114
115. 115
116. 116
117. 117
118. 118
119. 119
120. 120
121. 121
122. 122
123. 123
124. 124
125. 125
126. 126
127. 127
128. 128
129. 129
130. 130
131. 131
132. 132
133. 133
134. 134
135. 135
136. 136
137. 137
138. 138
139. 139
140. 140
141. 141
142. 142
143. 143
144. 144
145. 145
146. 146
147. 147
148. 148
149. 149
150. 150
151. 151
152. 152
153. 153
154. 154
155. 155
156. 156
157. 157
158. 158
159. 159
160. 160
161. 161
162. 162
163. 163
164. 164
165. 165
166. 166
167. 167
168. 168
169. 169
170. 170
171. 171

#!/usr/bin/python3
# https://github.com/ChunML/NLP/blob/master/text_generation/train_pt.py

import torch
import torch.nn as nn
import torch.nn.functional as F

import numpy as np
from collections import Counter
import os
from argparse import Namespace


flags = Namespace(
    train_file='data',
    seq_size=32,
    batch_size=256,
    embedding_size=64,
    lstm_size=64,
    gradients_norm=5,
    initial_words=['i', 'am'],
    predict_top_k=3,
    checkpoint_path='checkpoint',
)


def get_data_from_file(train_file, batch_size, seq_size):
    with open(train_file, 'r', encoding='utf-8') as f:
        text = f.read()
    text = text.split()

    word_counts = Counter(text)
    sorted_vocab = sorted(word_counts, key=word_counts.get, reverse=True)
    int_to_vocab = {k: w for k, w in enumerate(sorted_vocab)}
    vocab_to_int = {w: k for k, w in int_to_vocab.items()}
    n_vocab = len(int_to_vocab)

    print('Vocabulary size', n_vocab)

    int_text = [vocab_to_int[w] for w in text]
    num_batches = int(len(int_text) / (seq_size * batch_size))
    in_text = int_text[:num_batches * batch_size * seq_size]
    out_text = np.zeros_like(in_text)
    out_text[:-1] = in_text[1:]
    out_text[-1] = in_text[0]
    in_text = np.reshape(in_text, (batch_size, -1))
    out_text = np.reshape(out_text, (batch_size, -1))
    return int_to_vocab, vocab_to_int, n_vocab, in_text, out_text


def get_batches(in_text, out_text, batch_size, seq_size):
    num_batches = np.prod(in_text.shape) // (seq_size * batch_size)
    for i in range(0, num_batches * seq_size, seq_size):
        yield in_text[:, i:i+seq_size], out_text[:, i:i+seq_size]


class RNNModule(nn.Module):
    def __init__(self, n_vocab, seq_size, embedding_size, lstm_size):
        super(RNNModule, self).__init__()
        self.seq_size = seq_size
        self.lstm_size = lstm_size
        self.embedding = nn.Embedding(n_vocab, embedding_size)
        self.lstm = nn.LSTM(embedding_size,
                            lstm_size,
                            batch_first=True)
        self.dense = nn.Linear(lstm_size, n_vocab)

    def forward(self, x, prev_state):
        embed = self.embedding(x)
        output, state = self.lstm(embed, prev_state)
        logits = self.dense(output)

        return logits, state

    def zero_state(self, batch_size):
        return (torch.zeros(1, batch_size, self.lstm_size),
                torch.zeros(1, batch_size, self.lstm_size))


def get_loss_and_train_op(net, lr=0.001):
    criterion = nn.CrossEntropyLoss()
    optimizer = torch.optim.Adam(net.parameters(), lr=lr)

    return criterion, optimizer


def predict(device, net, words, n_vocab, vocab_to_int, int_to_vocab, top_k=5):
    net.eval()
    words = ['i', 'am']

    state_h, state_c = net.zero_state(1)
    state_h = state_h.to(device)
    state_c = state_c.to(device)
    for w in words:
        ix = torch.tensor([[vocab_to_int[w]]]).to(device)
        output, (state_h, state_c) = net(ix, (state_h, state_c))

    _, top_ix = torch.topk(output[0], k=top_k)
    choices = top_ix.tolist()
    choice = np.random.choice(choices[0])

    words.append(int_to_vocab[choice])

    for _ in range(100):
        ix = torch.tensor([[choice]]).to(device)
        output, (state_h, state_c) = net(ix, (state_h, state_c))

        _, top_ix = torch.topk(output[0], k=top_k)
        choices = top_ix.tolist()
        choice = np.random.choice(choices[0])
        words.append(int_to_vocab[choice])

    print(' '.join(words).encode('utf-8'))


def main():
    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
    int_to_vocab, vocab_to_int, n_vocab, in_text, out_text = get_data_from_file(
        flags.train_file, flags.batch_size, flags.seq_size)

    net = RNNModule(n_vocab, flags.seq_size,
                    flags.embedding_size, flags.lstm_size)
    net = net.to(device)

    criterion, optimizer = get_loss_and_train_op(net, 0.01)

    iteration = 0

    for e in range(200):
        batches = get_batches(in_text, out_text, flags.batch_size, flags.seq_size)
        state_h, state_c = net.zero_state(flags.batch_size)
        state_h = state_h.to(device)
        state_c = state_c.to(device)
        for x, y in batches:
            iteration += 1
            net.train()

            optimizer.zero_grad()

            x = torch.tensor(x).to(device)
            y = torch.tensor(y).to(device)

            logits, (state_h, state_c) = net(x, (state_h, state_c))
            loss = criterion(logits.transpose(1, 2), y)

            loss_value = loss.item()

            loss.backward()

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

            _ = torch.nn.utils.clip_grad_norm_(
                net.parameters(), flags.gradients_norm)

            optimizer.step()

            if iteration % 1 == 0:
                print('Epoch: {}/{}'.format(e, 20),
                      'Iteration: {}'.format(iteration),
                      'Loss: {}'.format(loss_value))

            if iteration % 1000 == 0:
                predict(device, net, flags.initial_words, n_vocab,
                        vocab_to_int, int_to_vocab, top_k=3)
                torch.save(net.state_dict(),
                           'checkpoint/model-{}.pth'.format(iteration))


if __name__ == '__main__':
    main()