-
Notifications
You must be signed in to change notification settings - Fork 0
/
dataload.py
278 lines (233 loc) · 9.84 KB
/
dataload.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
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
"""
file : dataload
authors : 21112254, 16008937, 20175911, 21180859
"""
import torch
from torch.utils.data import DataLoader, Dataset
from torchtext.data.utils import get_tokenizer
from torchtext.vocab import build_vocab_from_iterator
import torch.nn as nn
import nltk
from nltk import word_tokenize
from nltk.stem import WordNetLemmatizer
import re
nltk.download('punkt')
nltk.download('wordnet')
nltk.download('omw-1.4')
from transformers import BertTokenizer
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
####### Data untils functions #############
def get_dataset(config,df):
if config['Model'] == "Conv":
dataset = ConvDataset(config,df)
if config['Model'] == "BERT":
PRE_TRAINED_MODEL_NAME = 'bert-base-cased'
tokenizer = BertTokenizer.from_pretrained(PRE_TRAINED_MODEL_NAME)
dataset = BERTDataset(config,df,tokenizer,300)
return dataset
def gen_map(col_):
unique = col_.unique()
return {unique[i]:i for i in range(len(unique))}
def create_mappings(df,config):
return {task:gen_map(df[task]) for task in config['Tasks']}
########## Dataset classes ######################
class BERTDataset(Dataset):
"""
This class is adopted from the pytorch Dataset class. We use it for loading the data for the BERT model.
"""
def __init__(self,config, df, tokenizer, max_len):
self.lyrics = df['lyrics']
self.genre = df['genre']
self.tokenizer = tokenizer
self.max_len = max_len
self.violence_bool = 'violence' in config['Tasks']
self.dance_bool = 'danceability' in config['Tasks']
self.sadness_bool = 'sadness' in config['Tasks']
self.romantic_bool = 'romantic' in config['Tasks']
self.topic_bool = 'topic' in config['Tasks']
if self.violence_bool:
self.violence = df['violence']
if self.dance_bool:
self.dance = df['danceability']
if self.sadness_bool:
self.sadness = df['sadness']
if self.romantic_bool:
self.romantic = df['romantic']
if self.topic_bool:
self.topic = df['topic']
def __len__(self):
return len(self.lyrics)
def __getitem__(self, idx):
sample = {}
lyrics = str(self.lyrics[idx])
genre = self.genre[idx]
encoding = self.tokenizer.encode_plus(
lyrics,
add_special_tokens=True,
max_length=self.max_len,
return_token_type_ids=False,
padding='max_length',
return_attention_mask=True,
return_tensors='pt',
)
sample['lyrics'] = lyrics
sample['lyrics_ids'] = encoding['input_ids'].flatten()
sample['attention_mask'] = encoding['attention_mask'].flatten()
sample['genre'] = genre
if self.violence_bool:
sample['violence'] = self.violence[idx]
if self.dance_bool:
sample['danceability'] = self.dance[idx]
if self.sadness_bool:
sample['sadness'] = self.sadness[idx]
if self.romantic_bool:
sample['romantic'] = self.romantic[idx]
if self.topic_bool:
sample['topic'] = self.topic[idx]
return sample
class ConvDataset(Dataset):
def __init__(self,config,df):
self.lyrics = df['lyrics']
self.genre_bool = 'genre' in config['Tasks']
self.violence_bool = 'violence' in config['Tasks']
self.dance_bool = 'danceability' in config['Tasks']
self.energy_bool = 'energy' in config['Tasks']
self.topic_bool = 'topic' in config['Tasks']
self.feelings_bool = 'feelings' in config['Tasks']
self.sadness_bool = 'sadness' in config['Tasks']
self.romantic_bool = 'romantic' in config['Tasks']
if self.genre_bool:
self.genre = df['genre']
if self.violence_bool:
self.violence = df['violence']
if self.dance_bool:
self.dance = df['danceability']
if self.energy_bool:
self.energy = df['energy']
if self.topic_bool:
self.topic = df['topic']
if self.feelings_bool:
self.feelings = df['feelings']
if self.sadness_bool:
self.sadness = df['sadness']
if self.romantic_bool:
self.romantic = df['romantic']
def __len__(self):
return len(self.lyrics)
def __getitem__(self,idx):
if idx >= len(self.lyrics):
return None
sample = {}
sample['lyrics'] = self.lyrics[idx]
if self.genre_bool:
sample['genre'] = self.genre[idx]
if self.violence_bool:
sample['violence'] = self.violence[idx]
if self.dance_bool:
sample['danceability'] = self.dance[idx]
if self.energy_bool:
sample['energy'] = self.energy[idx]
if self.topic_bool:
sample['topic'] = self.topic[idx]
if self.feelings_bool:
sample['feelings'] = self.feelings[idx]
if self.sadness_bool:
sample['sadness'] = self.sadness[idx]
if self.romantic_bool:
sample['romantic'] = self.romantic[idx]
if self.topic_bool:
sample['topic'] = self.topic[idx]
return sample
############ Dataloaders ################
tokenizer = get_tokenizer('basic_english')
def yield_tokens(data_iter):
# pytorch implementation
for sample in data_iter:
if sample == None:
break
else:
yield tokenizer(sample['lyrics'])
def get_vocab(dataset):
vocab = build_vocab_from_iterator(yield_tokens(dataset), specials=["<unk>"])
vocab.set_default_index(vocab["<unk>"])
return vocab
def get_bert_dataloader(ds,batch_size,mapping):
genre_pipeline = lambda x: mapping['genre'][x]
topic_pipeline = lambda x: mapping['topic'][x]
def collate_batch(batch):
genre_ids = torch.Tensor([genre_pipeline(sample['genre']) for sample in batch]).long()
#attention_mask = torch.Tensor([sample['attention_mask'] for sample in batch])
attention_mask = torch.Tensor([sample['attention_mask'].flatten().tolist()[:300] for sample in batch]).to(torch.int64)
lyrics_ids = torch.Tensor([sample['lyrics_ids'].tolist()[:300] for sample in batch]).to(torch.int64)
batch_output = {'lyrics': lyrics_ids,
'genre': genre_ids,
'attention_mask': attention_mask}
if 'violence' in batch[0]:
violence_ids = torch.Tensor([sample['violence'] for sample in batch]).to(torch.float32).unsqueeze(1)
batch_output['violence'] = violence_ids
if 'romantic' in batch[0]:
romantic_ids = torch.Tensor([sample['romantic'] for sample in batch]).to(torch.float32).unsqueeze(1)
batch_output['romantic'] = romantic_ids
if 'sadness' in batch[0]:
sadness_ids = torch.Tensor([sample['sadness'] for sample in batch]).to(torch.float32).unsqueeze(1)
batch_output['sadness'] = sadness_ids
if 'romantic' in batch[0]:
romantic_ids = torch.Tensor([sample['romantic'] for sample in batch]).to(torch.float32).unsqueeze(1)
batch_output['romantic'] = romantic_ids
if 'danceability' in batch[0]:
dance_ids = torch.Tensor([sample['danceability'] for sample in batch]).to(torch.float32).unsqueeze(1)
batch_output['danceability'] = dance_ids
if 'topic' in batch[0]:
topic_ids = torch.Tensor([topic_pipeline(sample['topic']) for sample in batch]).long()
batch_output['topic'] = topic_ids
return batch_output
return DataLoader(
ds,
batch_size=batch_size,
#num_workers=4,
shuffle=True,
collate_fn=collate_batch
)
def get_dataloader(dataset, batch_size,mapping,vocab):
def preprocess(passage):
remove_punc = re.sub(r'[^\w\s]', ' ', passage)
remove_und_sc = remove_punc.replace('_', ' ')
remove_non_eng = re.sub(r'[^\x00-\x7F]+',' ', remove_und_sc)
digi = r'[0-9]'
remove_num = re.sub(digi, ' ', remove_non_eng)
lower_text = remove_num.lower()
tokenization = word_tokenize(lower_text)
lemma = WordNetLemmatizer()
tokens = []
for word in tokenization:
lemmatized_word = lemma.lemmatize(word)
tokens.append(lemmatized_word)
return tokens
lyrics_pipeline = lambda x: vocab(preprocess(x))
genre_pipeline = lambda x: mapping['genre'][x]
topic_pipeline = lambda x: mapping['topic'][x]
def collate_batch(batch):
lyrics_ids = [torch.Tensor(lyrics_pipeline(sample['lyrics'])).long() for sample in batch]
lyrics_ids = nn.utils.rnn.pad_sequence(lyrics_ids, padding_value=vocab['<pad>'], batch_first=True)
genre_ids = torch.Tensor([genre_pipeline(sample['genre']) for sample in batch]).long()
text_len = torch.Tensor([len(sample['lyrics']) for sample in batch])
batch_output = {'lyrics': lyrics_ids,
'genre': genre_ids,
'attention_mask': text_len}
if 'violence' in batch[0]:
violence_ids = torch.Tensor([sample['violence'] for sample in batch]).to(torch.float32).unsqueeze(1)
batch_output['violence'] = violence_ids
if 'romantic' in batch[0]:
romantic_ids = torch.Tensor([sample['romantic'] for sample in batch]).to(torch.float32).unsqueeze(1)
batch_output['romantic'] = romantic_ids
if 'sadness' in batch[0]:
sadness_ids = torch.Tensor([sample['sadness'] for sample in batch]).to(torch.float32).unsqueeze(1)
batch_output['sadness'] = sadness_ids
if 'romantic' in batch[0]:
romantic_ids = torch.Tensor([sample['romantic'] for sample in batch]).to(torch.float32).unsqueeze(1)
batch_output['romantic'] = romantic_ids
if 'danceability' in batch[0]:
dance_ids = torch.Tensor([sample['danceability'] for sample in batch]).to(torch.float32).unsqueeze(1)
batch_output['danceability'] = dance_ids
return batch_output
return DataLoader(dataset, batch_size=batch_size, shuffle=True, collate_fn=collate_batch)