-
Notifications
You must be signed in to change notification settings - Fork 0
/
CustomInLawBERT_RR.py
245 lines (184 loc) · 6.46 KB
/
CustomInLawBERT_RR.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
# -*- coding: utf-8 -*-
"""finetune_InLegalBERT.ipynb
Automatically generated by Colaboratory.
Original file is located at
https://colab.research.google.com/drive/1d4bwab1Ke5JPYh_nUHOuzxIlkO0Prvu2
"""
import numpy as np
import torch
import torch.nn as nn
import torchmetrics
import os
os.environ["CUDA_VISIBLE_DEVICES"]="0"
#root="/export/home/nlp1/rachit/Inlegalbert"
from transformers import AutoTokenizer, AutoModel
tokenizer = AutoTokenizer.from_pretrained("law-ai/CustomInLawBERT")
# text = "Replace this string with yours"
# encoded_input = tokenizer(text, return_tensors="pt")
Inlegal_model = AutoModel.from_pretrained("law-ai/CustomInLawBERT")
# output = model(**encoded_input)
# last_hidden_state = output.last_hidden_state
# import json
# import numpy as np
# with open('/content/drive/MyDrive/nlp course/legaldatatrain.json', 'r') as f:
# data = json.load(f)
# texts = []
# labels = []
# for item in data:
# annotations = item.get('annotations', [])
# for annotation in annotations:
# result = annotation.get('result', [])
# for r in result:
# text = r.get('value', {}).get('text', '')
# label = r.get('value', {}).get('labels', [])
# if text and label:
# texts.append(text)
# labels.append(label[0])
# texts_array = np.array(texts)
# labels_array = np.array(labels)
# texts_array.shape,labels_array.shape
# np.unique(labels_array)
# label_mapping = {
# 'ANALYSIS': 0,
# 'ARG_PETITIONER': 1,
# 'ARG_RESPONDENT': 2,
# 'FAC': 3,
# 'ISSUE': 4,
# 'NONE': 5,
# 'PREAMBLE': 6,
# 'PRE_NOT_RELIED': 7,
# 'PRE_RELIED': 8,
# 'RATIO': 9,
# 'RLC': 10,
# 'RPC': 11,
# 'STA': 12
# }
# labels_int = np.array([label_mapping[label] for label in labels_array])
# np.save('/content/drive/MyDrive/nlp course/labels_train.npy', labels_int)
# np.save('/content/drive/MyDrive/nlp course/texts_train.npy', texts_array)
X_train=np.load(f'texts_train.npy')
X_test=np.load(f'texts_dev.npy')
y_train=np.load(f'labels_train.npy')
y_test=np.load(f'labels_dev.npy')
print("Length of X_train is ", len(X_train))
print("Length of X_test is ", len(X_test))
print("Length of y_train is ", len(y_train))
print("Length of y_test is ", len(y_test))
print(np.unique(y_train))
tokenized_text_train=[]
tokenized_text_test=[]
for text in X_train:
encoded_input = tokenizer.encode_plus(text,max_length=32,padding='max_length', truncation=True,return_tensors="pt")
tokenized_text_train.append(encoded_input)
for text in X_test:
encoded_input = tokenizer.encode_plus(text,max_length=32,padding='max_length',truncation=True, return_tensors="pt")
tokenized_text_test.append(encoded_input)
num_classes=13
classifier=nn.Sequential(
nn.Linear(768,512),
nn.ReLU(),
nn.Dropout(0.4),
nn.Linear(512,128),
nn.ReLU(),
nn.Dropout(0.4),
nn.Linear(128,num_classes),
nn.Softmax()
)
for param in Inlegal_model.parameters():
param.requires_grad=False
for p in classifier.parameters():
p.requires_grad=True
class custom(nn.Module):
def __init__(self):
super().__init__()
self.m1=Inlegal_model
self.m2=classifier
def forward(self,inp,tti,mask):
x=self.m1(input_ids=inp,token_type_ids=tti,attention_mask=mask).pooler_output
x=torch.squeeze(x, 0)
x=self.m2(x)
return x
model=custom()
inp_train=[]
masks_train=[]
tti_train=[]
for data in tokenized_text_train:
inp=data['input_ids']
token_type_id=data['token_type_ids']
mask=data['attention_mask']
inp_train.append(inp)
masks_train.append(mask)
tti_train.append(token_type_id)
inp_test=[]
masks_test=[]
tti_test=[]
for data in tokenized_text_test:
inp=data['input_ids']
token_type_id=data['token_type_ids']
mask=data['attention_mask']
inp_test.append(inp)
masks_test.append(mask)
tti_test.append(token_type_id)
inp_train=torch.stack(inp_train)
masks_train=torch.stack(masks_train)
tti_train=torch.stack(tti_train)
inp_test=torch.stack(inp_test)
masks_test=torch.stack(masks_test)
tti_test=torch.stack(tti_test)
y_train=torch.LongTensor(y_train)
y_test=torch.LongTensor(y_test)
y_train=y_train.to('cuda')
y_test=y_test.to('cuda')
inp_train=inp_train.to('cuda')
masks_train=masks_train.to('cuda')
tti_train=tti_train.to('cuda')
inp_test=inp_test.to('cuda')
masks_test=masks_test.to('cuda')
tti_test=tti_test.to('cuda')
model=model.to("cuda")
from torch.utils.data import DataLoader,TensorDataset
train_dataset=TensorDataset(inp_train,tti_train,masks_train,y_train)
val_dataset=TensorDataset(inp_test,tti_test,masks_test,y_test)
train = DataLoader(train_dataset, batch_size=64, shuffle=True)
val=DataLoader(val_dataset, batch_size=64, shuffle=True)
epochs=5
print(len(val),len(train))
criterion=nn.CrossEntropyLoss()
opt=torch.optim.Adam(model.parameters(), lr=0.1)
accuracy = torchmetrics.Accuracy(task="multiclass", num_classes=num_classes).to('cuda')
for epoch in range(epochs):
avg_val_acc=0
count=0
avg_train_loss=0
avg_val_loss=0
for batch in train:
count=count+1
if count%100==0:
print(f'Epoch {epoch} Batch no.: {count}')
X_batch_in,X_batch_tti,X_batch_mask,label_batch = batch
X_batch_in=torch.squeeze(X_batch_in)
X_batch_tti=torch.squeeze(X_batch_tti)
X_batch_mask=torch.squeeze(X_batch_mask)
preds=model(X_batch_in,X_batch_tti,X_batch_mask)
loss=criterion(preds,label_batch)
avg_train_loss=avg_train_loss+loss
opt.zero_grad()
loss.backward()
opt.step()
with torch.no_grad():
for batch in val:
X_val_batch_in,X_val_batch_tti,X_val_batch_mask,label_val_batch = batch
X_val_batch_in=torch.squeeze(X_val_batch_in)
X_val_batch_tti=torch.squeeze(X_val_batch_tti)
X_val_batch_mask=torch.squeeze(X_val_batch_mask)
val_preds=model(X_val_batch_in,X_val_batch_tti,X_val_batch_mask)
val_loss=criterion(val_preds,label_val_batch)
val_acc=accuracy.update(val_preds,label_val_batch)
print(val_preds)
print(label_val_batch)
avg_val_loss=avg_val_loss+val_loss
#print(f'Epoch {epoch}: Training Loss: Training accuracy: , Validation Loss: Validation accuracy:')
if epoch%1==0:
print(f"| Epoch={epoch} | Validation Accuracy={accuracy.compute()} | Training Loss={avg_train_loss/len(train)} | Validation_Loss={avg_val_loss/len(val)} |")
print('-------------------------------------------------------------------------------------------------------------------------------------------------------------------------')
torch.save(model,f'weights/inlegalbert_{epochs}epochs.pt')