File size: 9,041 Bytes
e8ca4ee
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
import os
import sys
import time
import random
import json
import numpy as np
import matplotlib.pyplot as plt
from model import MusicLSTM as MusicRNN
import torch
import torch.nn as nn
import torch.optim as optim
from torch.autograd import Variable
from utils import seq_to_tensor, load_vocab, save_vocab


def logger(active=True):
    """Simple logging utility."""
    def log(*args, **kwargs):
        if active:
            print(*args, **kwargs)
    return log

# Configuration
class Config:
    SAVE_EVERY = 20
    SEQ_SIZE = 25
    RANDOM_SEED = 11
    VALIDATION_SIZE = 0.15
    LR = 1e-3
    N_EPOCHS = 100
    NUM_LAYERS = 1
    HIDDEN_SIZE = 150
    DROPOUT_P = 0
    MODEL_TYPE = 'lstm'
    INPUT_FILE = 'data/music.txt'
    RESUME = False
    BATCH_SIZE = 1

# Utility functions
def tic():
    """Start timer."""
    return time.time()

def toc(start_time, msg=None):
    """Calculate elapsed time."""
    s = time.time() - start_time
    m = int(s / 60)
    if msg:
        return f'{m}m {int(s - (m * 60))}s {msg}'
    return f'{m}m {int(s - (m * 60))}s'

class DataLoader:
    def __init__(self, input_file, config):
        self.config = config
        self.char_idx, self.char_list = self._load_chars(input_file)
        self.data = self._load_data(input_file)
        self.train_idxs, self.valid_idxs = self._split_data()
        log = logger(True)
        log(f"Total songs: {len(self.data)}")
        log(f"Training songs: {len(self.train_idxs)}")
        log(f"Validation songs: {len(self.valid_idxs)}")
    
    def _load_chars(self, input_file):
        """Load unique characters from the input file."""
        with open(input_file, 'r') as f:
            char_idx = ''.join(set(f.read()))
        return char_idx, list(char_idx)

    def _load_data(self, input_file):
        """Load song data from input file."""
        with open(input_file, "r") as f:
            data, buffer = [], ''
            for line in f:
                if line == '<start>\n':
                    buffer += line
                elif line == '<end>\n':
                    buffer += line
                    data.append(buffer)
                    buffer = ''
                else:
                    buffer += line

        # Filter songs shorter than sequence size
        data = [song for song in data if len(song) > self.config.SEQ_SIZE + 10]
        return data

    def _split_data(self):
        """Split data into training and validation sets."""
        num_train = len(self.data)
        indices = list(range(num_train))

        np.random.seed(self.config.RANDOM_SEED)
        np.random.shuffle(indices)

        split_idx = int(np.floor(self.config.VALIDATION_SIZE * num_train))
        train_idxs = indices[split_idx:]
        valid_idxs = indices[:split_idx]

        return train_idxs, valid_idxs

    def rand_slice(self, data, slice_len=None):
        """Get a random slice of data."""
        if slice_len is None:
            slice_len = self.config.SEQ_SIZE

        d_len = len(data)
        s_idx = random.randint(0, d_len - slice_len)
        e_idx = s_idx + slice_len + 1
        return data[s_idx:e_idx]

    def seq_to_tensor(self, seq):
        """Convert sequence to tensor."""
        out = torch.zeros(len(seq)).long()
        for i, c in enumerate(seq):
            out[i] = self.char_idx.index(c)
        return out

    def song_to_seq_target(self, song):
        """Convert a song to sequence and target."""
        try:
            a_slice = self.rand_slice(song)
            seq = self.seq_to_tensor(a_slice[:-1])
            target = self.seq_to_tensor(a_slice[1:])
            return seq, target
        except Exception as e:
            print(f"Error in song_to_seq_target: {e}")
            print(f"Song length: {len(song)}")
            raise    

def train_model(config, data_loader, model, optimizer, loss_function):
    """Training loop for the model."""
    log = logger(True)
    time_since = tic()
    losses, v_losses = [], []

    for epoch in range(config.N_EPOCHS):
        # Training phase
        epoch_loss = 0
        model.train()

        for i, song_idx in enumerate(data_loader.train_idxs):
            try:
                seq, target = data_loader.song_to_seq_target(data_loader.data[song_idx])

                # Reset hidden state and gradients
                model.init_hidden()
                optimizer.zero_grad()

                # Forward pass
                outputs = model(seq)
                loss = loss_function(outputs, target)

                # Backward pass and optimization
                loss.backward()
                optimizer.step()

                epoch_loss += loss.item()

                msg = f'\rTraining Epoch: {epoch}, {(i+1)/len(data_loader.train_idxs)*100:.2f}% iter: {i} Time: {toc(time_since)} Loss: {loss.item():.4f}'
                sys.stdout.write(msg)
                sys.stdout.flush()

            except Exception as e:
                log(f"Error processing song {song_idx}: {e}")
                continue

        print()
        losses.append(epoch_loss / len(data_loader.train_idxs))

        # Validation phase
        model.eval()
        val_loss = 0
        with torch.no_grad():
            for i, song_idx in enumerate(data_loader.valid_idxs):
                try:
                    seq, target = data_loader.song_to_seq_target(data_loader.data[song_idx])

                    # Reset hidden state
                    model.init_hidden()

                    # Forward pass
                    outputs = model(seq)
                    loss = loss_function(outputs, target)

                    val_loss += loss.item()

                    msg = f'\rValidation Epoch: {epoch}, {(i+1)/len(data_loader.valid_idxs)*100:.2f}% iter: {i} Time: {toc(time_since)} Loss: {loss.item():.4f}'
                    sys.stdout.write(msg)
                    sys.stdout.flush()

                except Exception as e:
                    log(f"Error processing validation song {song_idx}: {e}")
                    continue

        print()
        v_losses.append(val_loss / len(data_loader.valid_idxs))

        # Checkpoint saving
        if epoch % config.SAVE_EVERY == 0 or epoch == config.N_EPOCHS - 1:
            log('=======> Saving..')
            state = {
                'model': model.state_dict(),
                'optimizer': optimizer.state_dict(),
                'loss': losses[-1],
                'v_loss': v_losses[-1],
                'losses': losses,
                'v_losses': v_losses,
                'epoch': epoch,
            }
            os.makedirs('checkpoint', exist_ok=True)
            torch.save(model, f'checkpoint/ckpt_mdl_{config.MODEL_TYPE}_ep_{config.N_EPOCHS}_hsize_{config.HIDDEN_SIZE}_dout_{config.DROPOUT_P}.t{epoch}')

    return losses, v_losses

def plot_losses(losses, v_losses):
    """Plot training and validation losses."""
    plt.figure(figsize=(10, 5))
    plt.plot(losses, label='Training Loss')
    plt.plot(v_losses, label='Validation Loss')
    plt.xlabel('Epoch')
    plt.ylabel('Loss')
    plt.title('Loss per Epoch')
    plt.legend()
    plt.show()

def generate_song(model, data_loader, prime_str='<start>', max_len=1000, temp=0.8):
    """Generate a new song using the trained model."""
    model.eval()
    model.init_hidden()
    creation = prime_str
    char_idx, char_list = load_vocab()

    # Build up hidden state
    prime = seq_to_tensor(creation, char_idx)

    with torch.no_grad():
        for _ in range(len(prime)-1):
            _ = model(prime[_:_+1])

        # Generate rest of sequence
        for _ in range(max_len):
            last_char = prime[-1:]
            out = model(last_char).squeeze()

            out = torch.exp(out/temp)
            dist = out / torch.sum(out)

            # Sample from distribution
            next_char_idx = torch.multinomial(dist, 1).item()
            next_char = char_idx[next_char_idx]

            creation += next_char
            prime = torch.cat([prime, torch.tensor([next_char_idx])], dim=0)

            if creation[-5:] == '<end>':
                break

    return creation

def main():
    """Main execution function."""
    # Set up configuration and data
    global model, data_loader
    config = Config()
    data_loader = DataLoader(config.INPUT_FILE, config)

    # Model setup
    in_size = out_size = len(data_loader.char_idx)
    model = MusicRNN(
        in_size,
        config.HIDDEN_SIZE,
        out_size,
        config.MODEL_TYPE,
        config.NUM_LAYERS,
        config.DROPOUT_P
    )

    # Optimizer and loss
    optimizer = torch.optim.Adam(model.parameters(), lr=config.LR)
    loss_function = nn.CrossEntropyLoss()

    # Train the model
    losses, v_losses = train_model(config, data_loader, model, optimizer, loss_function)

    # Plot losses
    plot_losses(losses, v_losses)
    save_vocab(data_loader)

    # Generate a song
    generated_song = generate_song(model, data_loader)
    print("Generated Song:", generated_song)

if __name__ == "__main__":
    main()