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
| import torch
from datasets import load_dataset
from torch.utils.data import Dataset
from torch.utils.data import DataLoader
import torch.nn as nn
from transformers import BertTokenizer, BertModel
import torch.optim as optim
from torch.nn.functional import one_hot
import pytorch_lightning as pl
from pytorch_lightning import Trainer
from torchmetrics.functional import accuracy, recall, precision, f1_score
from pytorch_lightning.callbacks.early_stopping import EarlyStopping
from pytorch_lightning.callbacks import ModelCheckpoint
class MydataSet(Dataset):
def __init__(self, path, split):
self.dataset = load_dataset('csv', data_files=path, split=split)
def __getitem__(self, item):
text = self.dataset[item]['text']
label = self.dataset[item]['label']
return text, label
def __len__(self):
return len(self.dataset)
def collate_fn(data):
sents = [i[0] for i in data]
labels = [i[1] for i in data]
data = token.batch_encode_plus(
batch_text_or_text_pairs=sents,
truncation=True,
padding='max_length',
max_length=200,
return_tensors='pt',
return_length=True,
)
input_ids = data['input_ids']
attention_mask = data['attention_mask']
token_type_ids = data['token_type_ids']
labels = torch.LongTensor(labels)
return input_ids, attention_mask, token_type_ids, labels
class BiLSTMClassifier(nn.Module):
def __init__(self, drop, hidden_dim, output_dim):
super(BiLSTMClassifier, self).__init__()
self.drop = drop
self.hidden_dim = hidden_dim
self.output_dim = output_dim
self.embedding = BertModel.from_pretrained('bert-base-chinese')
for param in self.embedding.parameters():
param.requires_grad_(False)
self.lstm = nn.LSTM(input_size=768, hidden_size=self.hidden_dim, num_layers=2, batch_first=True,
bidirectional=True, dropout=self.drop)
self.fc = nn.Linear(self.hidden_dim * 2, self.output_dim)
def forward(self, input_ids, attention_mask, token_type_ids):
embedded = self.embedding(input_ids=input_ids, attention_mask=attention_mask, token_type_ids=token_type_ids)
embedded = embedded.last_hidden_state
out, (h_n, c_n) = self.lstm(embedded)
output = torch.cat((h_n[-2, :, :], h_n[-1, :, :]), dim=1)
output = self.fc(output)
return output
class BiLSTMLighting(pl.LightningModule):
def __init__(self, drop, hidden_dim, output_dim):
super(BiLSTMLighting, self).__init__()
self.model = BiLSTMClassifier(drop, hidden_dim, output_dim)
self.criterion = nn.CrossEntropyLoss()
self.train_dataset = MydataSet('./data/archive/train_clean.csv', 'train')
self.val_dataset = MydataSet('./data/archive/val_clean.csv', 'train')
self.test_dataset = MydataSet('./data/archive/test_clean.csv', 'train')
def configure_optimizers(self):
optimizer = optim.AdamW(self.parameters(), lr=lr)
return optimizer
def forward(self, input_ids, attention_mask, token_type_ids):
return self.model(input_ids, attention_mask, token_type_ids)
def train_dataloader(self):
train_loader = DataLoader(dataset=self.train_dataset, batch_size=batch_size, collate_fn=collate_fn,
shuffle=True)
return train_loader
def training_step(self, batch, batch_idx):
input_ids, attention_mask, token_type_ids, labels = batch
y = one_hot(labels + 1, num_classes=3)
y = y.to(dtype=torch.float)
y_hat = self.model(input_ids, attention_mask, token_type_ids)
y_hat = y_hat.squeeze()
loss = self.criterion(y_hat, y)
self.log('train_loss', loss, prog_bar=True, logger=True, on_step=True, on_epoch=True)
return loss
def val_dataloader(self):
val_loader = DataLoader(dataset=self.val_dataset, batch_size=batch_size, collate_fn=collate_fn, shuffle=False)
return val_loader
def validation_step(self, batch, batch_idx):
input_ids, attention_mask, token_type_ids, labels = batch
y = one_hot(labels + 1, num_classes=3)
y = y.to(dtype=torch.float)
y_hat = self.model(input_ids, attention_mask, token_type_ids)
y_hat = y_hat.squeeze()
loss = self.criterion(y_hat, y)
self.log('val_loss', loss, prog_bar=False, logger=True, on_step=True, on_epoch=True)
return loss
def test_dataloader(self):
test_loader = DataLoader(dataset=self.test_dataset, batch_size=batch_size, collate_fn=collate_fn, shuffle=False)
return test_loader
def test_step(self, batch, batch_idx):
input_ids, attention_mask, token_type_ids, labels = batch
target = labels + 1
y = one_hot(target, num_classes=3)
y = y.to(dtype=torch.float)
y_hat = self.model(input_ids, attention_mask, token_type_ids)
y_hat = y_hat.squeeze()
pred = torch.argmax(y_hat, dim=1)
acc = (pred == target).float().mean()
loss = self.criterion(y_hat, y)
self.log('loss', loss)
re = recall(pred, target, task="multiclass", num_classes=class_num, average=None)
pre = precision(pred, target, task="multiclass", num_classes=class_num, average=None)
f1 = f1_score(pred, target, task="multiclass", num_classes=class_num, average=None)
def log_score(name, scores):
for i, score_class in enumerate(scores):
self.log(f"{name}_class{i}", score_class)
log_score("recall", re)
log_score("precision", pre)
log_score("f1", f1)
self.log('acc', accuracy(pred, target, task="multiclass", num_classes=class_num))
self.log('avg_recall', recall(pred, target, task="multiclass", num_classes=class_num, average="weighted"))
self.log('avg_precision', precision(pred, target, task="multiclass", num_classes=class_num, average="weighted"))
self.log('avg_f1', f1_score(pred, target, task="multiclass", num_classes=class_num, average="weighted"))
def test():
model = BiLSTMLighting.load_from_checkpoint(checkpoint_path=PATH,
drop=dropout, hidden_dim=rnn_hidden, output_dim=class_num)
trainer = Trainer(fast_dev_run=False)
result = trainer.test(model)
print(result)
|
