-
Notifications
You must be signed in to change notification settings - Fork 2
/
Transformer_cls.py
257 lines (223 loc) · 10.4 KB
/
Transformer_cls.py
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
# -*- coding: utf-8 -*-
from gensim.models import Word2Vec
from torch.utils.data.dataset import Dataset
from torch.utils.data import DataLoader, WeightedRandomSampler
import torch
import os
import numpy as np
import pandas as pd
import pickle
import torch.optim as optim
import torch.nn as nn
import argparse
from torch.autograd import Variable
import pdb
def adjust_learning_rate(optimizer, epoch):
lr = opt.learning_rate * (0.1 ** (epoch // 10))
for param_group in optimizer.param_groups:
param_group['lr'] = lr
return optimizer
class TxtDatasetProcessing(Dataset):
def __init__(self, dataset, seqlen, syllable_vocabulary, music_vocabulary):
self.dataset = dataset
lyrics = []
musics = []
labels = []
for i in range(dataset.shape[0]):
musics += list(dataset[i, 0, :])
lyrics += list(dataset[i, 1, :])
labels += list(dataset[i, 2, :])
#把negative samples记录下来,多两倍
index = []
for i in range(len(labels)):
if labels[i] == 'negative':
index.append(i)
negative_musics = []
negative_lyrics = []
negative_labels = []
for i in index:
negative_musics.append(musics[i])
negative_lyrics.append(lyrics[i])
negative_labels.append(labels[i])
self.lyrics = lyrics+negative_lyrics
self.musics = musics+negative_musics
self.labels = labels+negative_labels
self.seqlen = seqlen
self.word_vocabulary = word_vocabulary
self.syllable_vocabulary = syllable_vocabulary
self.music_vocabulary = music_vocabulary
def __getitem__(self, index):
lyric = self.lyrics[index]
music = self.musics[index]
la = self.labels[index]
if la == 'positive':
label = torch.ones(1)
else:
label = torch.zeros(1)
#txt = torch.zeros((seqlen, 23), dtype = torch.float64)
txt = torch.LongTensor(np.zeros(self.seqlen, dtype=np.int64))
mus = torch.LongTensor(np.zeros(self.seqlen, dtype=np.int64))
txt_len = 0
for i in range(len(lyric)):
word = ''
for syll in lyric[i]:
word += syll
if word in self.word_vocabulary:
word2idx = self.word_vocabulary[word]
else:
continue
for j in range(len(lyric[i])):
syll = lyric[i][j]
#note = music[i][j]
if syll in self.syllable_vocabulary:
syll2idx = self.syllable_vocabulary[syll]
else:
continue
#syllWordVec = (syll,word,note)
music_note = 'p_'+str(music[i][j][0])+'^'+'d_'+str(music[i][j][1])+'^'+'r_'+str(music[i][j][2])
if music_note in self.music_vocabulary:
music_note2idx = self.music_vocabulary[music_note]
else:
continue
syllWordVec = (syll2idx,music_note2idx)
if txt_len<self.seqlen:
txt[txt_len] = syll2idx
mus[txt_len] = music_note2idx
txt_len += 1
else:
break
if txt_len >= self.seqlen:
break
return txt, mus, label.type(torch.int64)
def __len__(self):
return len(self.labels)
class TransformerClassifier(nn.Module):
def __init__(self, input_txt_size, input_mus_size, nhead, num_layers, num_classes, syllable_vocabulary, music_vocabulary):
super(TransformerClassifier, self).__init__()
self.nhead = nhead
self.num_layers = num_layers
input_size = input_txt_size+input_mus_size
self.encoder_layer = nn.TransformerEncoderLayer(d_model=input_size, nhead=8, batch_first=True)
self.encoder = nn.TransformerEncoder(self.encoder_layer, num_layers=self.num_layers)
#self.lstm = nn.LSTM(input_size, hidden_size, num_layers, batch_first=True, bidirectional=True)
self.fc = nn.Linear(input_size, num_classes)
syllable_vocab_size = len(syllable_vocabulary)
self.word_embeddings = nn.Embedding(syllable_vocab_size, input_txt_size)
music_vocab_size = len(music_vocabulary)
self.music_embeddings = nn.Embedding(music_vocab_size, input_mus_size)
def forward(self, txt, mus):
#h0 = torch.zeros(self.num_layers*2, txt.size(0), self.hidden_size).to(device) # 同样考虑向前层和向后层
#c0 = torch.zeros(self.num_layers*2, txt.size(0), self.hidden_size).to(device)
#lens = len(txt)
#batch_size = txt.size(0)
txt = self.word_embeddings(txt)
mus = self.music_embeddings(mus)
X = torch.cat((txt, mus), 2)
#X = X.view(lens, batch_size, -1)
Y = self.encoder(X)
out = torch.mean(Y, dim=1)
out = self.fc(out)
#out, _ = self.lstm(X, (h0, c0)) # LSTM输出大小为 (batch_size, seq_length, hidden_size*2)
#out = self.fc(out[:, -1, :])
#out = self.Sigmoid(out)
return out
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='CLF.py')
parser.add_argument('--data', type=str, default='lyrics_datasets_v3/dataset_50_v3_clf.npy', help="Dnd data.")
parser.add_argument('--dataset_len', type=str, default=50)
parser.add_argument('--batch_size', type=str, default=32)
parser.add_argument('--seqlen', type=str, default=70)
parser.add_argument('--learning_rate', type=str, default=0.0001)
parser.add_argument('--num_epochs', type=str, default=30)
opt = parser.parse_args()
#dataset_len = 20
#batch_size = 64
#seqlen = 30
#learning_rate = 0.0001
#num_epochs = 30
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# save np.load
np_load_old = np.load
# modify the default parameters of np.load
np.load = lambda *a,**k: np_load_old(*a, allow_pickle=True, **k)
dataset = np.load(opt.data) #shape (8, 4, 4722)
word_vocabulary_file = 'saved_model/word_vocabulary.npy'
word_vocabulary = np.load(word_vocabulary_file)
word_vocabulary = word_vocabulary.item()
syllable_vocabulary_file = 'saved_model/syllable_vocabulary.npy'
syllable_vocabulary = np.load(syllable_vocabulary_file)
syllable_vocabulary = syllable_vocabulary.item()
music_vocabulary_file = 'saved_model/music_vocabulary_'+str(opt.dataset_len)+'.npy'
music_vocabulary = np.load(music_vocabulary_file)
music_vocabulary = music_vocabulary.item()
model = TransformerClassifier(input_txt_size=118, input_mus_size=10, nhead=8, num_layers=6, num_classes=2, syllable_vocabulary=syllable_vocabulary, music_vocabulary=music_vocabulary)
model = model.to(device)
for i in range(8):
train_dataset = np.concatenate((dataset[0:i], dataset[i+1:]), axis=0)
test_dataset = dataset[i:i+1]
dtrain_set = TxtDatasetProcessing(train_dataset, opt.seqlen, syllable_vocabulary, music_vocabulary)
train_loader = DataLoader(dtrain_set, batch_size=opt.batch_size, shuffle=True, num_workers=4)
# data[0] torch.Size([64, 30]) data[1] torch.Size([64, 30]) data[2] torch.Size([64, 1])
#for data in train_loader:
# print(data)
dtest_set = TxtDatasetProcessing(test_dataset, opt.seqlen, syllable_vocabulary, music_vocabulary)
test_loader = DataLoader(dtrain_set, batch_size=opt.batch_size, shuffle=True, num_workers=4)
optimizer = optim.Adam(model.parameters(), lr=opt.learning_rate)
loss_function = nn.CrossEntropyLoss()
train_loss_ = []
test_loss_ = []
train_acc_ = []
test_acc_ = []
for epoch in range(opt.num_epochs):
optimizer = adjust_learning_rate(optimizer, epoch)
total_acc = 0.0
total_loss = 0.0
total = 0.0
for iter, traindata in enumerate(train_loader):
train_txt, train_mus ,train_labels = traindata
train_labels = torch.squeeze(train_labels)
train_txt = Variable(train_txt.to(device))
train_mus = Variable(train_mus.to(device))
train_labels = Variable(train_labels.to(device))
model.zero_grad()
output = model(train_txt, train_mus)
output = output.squeeze(dim=-1)
loss = loss_function(output, Variable(train_labels))
loss.backward()
optimizer.step()
_, predicted = torch.max(output.data, 1)
total_acc += (predicted == train_labels).sum()
total_acc = total_acc.type(torch.float32)
total += torch.tensor(len(train_labels))
total = total.type(torch.float32)
total_loss += loss.data
train_loss_.append(total_loss / total)
train_acc_.append(total_acc / total)
# testing epoch
total_acc = 0.0
total_loss = 0.0
total = 0.0
for iter, testdata in enumerate(test_loader):
test_txt, test_mus, test_labels = testdata
test_labels = torch.squeeze(test_labels)
test_txt = Variable(test_txt.to(device))
test_mus = Variable(test_mus.to(device))
test_labels = Variable(test_labels.to(device))
output = model(test_txt, test_mus)
output = output.squeeze(dim=-1)
loss = loss_function(output, Variable(test_labels))
# calc testing acc
_, predicted = torch.max(output.data, 1)
#predicted = 1*(output>0.5)
total_acc += (predicted == test_labels).sum()
total_acc = total_acc.type(torch.float32)
total += torch.tensor(len(test_labels))
total = total.type(torch.float32)
total_loss += loss.data
test_loss_.append(total_loss / total)
test_acc_.append(total_acc / total)
print('[Epoch: %3d/%3d] Training Loss: %.3f, Testing Loss: %.3f, Training Acc: %.3f, Testing Acc: %.3f'
% (epoch, opt.num_epochs, train_loss_[epoch], test_loss_[epoch], test_acc_[epoch], train_acc_[epoch]))
filename = 'Transformer_datasetlen_'+str(opt.dataset_len)+'_fold_'+str(i)+'_clf.pkl'
torch.save(model.state_dict(), filename)
print('File %s is saved.' % filename)