File size: 6,682 Bytes
d5062c8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# -*- coding: utf-8 -*-
"""
Created on Mon Aug 30 19:54:17 2021

@author: luol2
"""



import os, sys
import numpy as np
from tensorflow.keras.preprocessing.sequence import pad_sequences
from transformers import AutoTokenizer


class Hugface_RepresentationLayer(object):
    
    
    def __init__(self, tokenizer_name_or_path, label_file,lowercase=True):
        

        #load vocab

        self.tokenizer = AutoTokenizer.from_pretrained(tokenizer_name_or_path, use_fast=True,do_lower_case=lowercase)
        self.label_2_index={}
        self.index_2_label={}
        self.label_table_size=0
        self.load_label_vocab(label_file,self.label_2_index,self.index_2_label)
        self.label_table_size=len(self.label_2_index)
        self.vocab_len=len(self.tokenizer)
       
    def load_label_vocab(self,fea_file,fea_index,index_2_label):
        
        fin=open(fea_file,'r',encoding='utf-8')
        all_text=fin.read().strip().split('\n')
        fin.close()
        for i in range(0,len(all_text)):
            fea_index[all_text[i]]=i
            index_2_label[str(i)]=all_text[i]
            

            
    def generate_label_list(self,ori_tokens,labels,word_index): #the lable of subtoken is the same with the label of first subtoken
        label_list=['O']*len(word_index)

        label_list_index=[]
        old_new_token_map=[]
        ori_i=0
        for i in range(0,len(word_index)):
            if word_index[i]==None:
                label_list_index.append(self.label_2_index[label_list[i]])
            else:
                label_list[i]=labels[word_index[i]]
                label_list_index.append(self.label_2_index[label_list[i]])
                if word_index[i]==ori_i:
                    old_new_token_map.append(i)
                    ori_i+=1
        

        bert_text_label=[]
        for i in range(0,len(ori_tokens)):
            bert_text_label.append([ori_tokens[i],labels[i],old_new_token_map[i]])

        return label_list_index,bert_text_label
    
    def generate_label_list_B(self,ori_tokens,labels,word_index): #tonly first subtoken is B, other is I
        label_list=['O']*len(word_index)

        label_list_index=[]
        old_new_token_map=[]
        ori_i=0
        first_index=-1
        i=0
        while i <len(word_index):
            if word_index[i]==None:
                label_list_index.append(self.label_2_index[label_list[i]])
                i+=1
            else:
                first_index=word_index[i]
                if first_index==ori_i:
                    old_new_token_map.append(i)
                    ori_i+=1
                label_list[i]=labels[word_index[i]]
                label_list_index.append(self.label_2_index[label_list[i]])
                i+=1
                while word_index[i]==first_index and word_index[i]!=None:
                    #print(first_index)
                    if labels[first_index].startswith("B-"):
                        label_list[i]='I-'+labels[first_index][2:]
                        label_list_index.append(self.label_2_index[label_list[i]])
                    else:
                        label_list[i]=labels[word_index[i]]
                        label_list_index.append(self.label_2_index[label_list[i]])
                    i+=1
                        

        

        bert_text_label=[]
        #print(len(old_new_token_map))
        for i in range(0,len(ori_tokens)):
            if i<len(old_new_token_map):
                bert_text_label.append([ori_tokens[i],labels[i],old_new_token_map[i]])
            else: # after token > max len
                break
        return label_list_index,bert_text_label
    
    def load_data_hugface(self,instances,  word_max_len=100, label_type='softmax'):
    
        x_index=[]
        x_seg=[]
        x_mask=[]
        y_list=[]
        bert_text_labels=[]
        max_len=0
        over_num=0
        maxT=word_max_len
        ave_len=0

        #print('instances:', instances)
        #print('labels:',labels)
        
        
        for sentence in instances:                           
            sentence_text_list=[]
            label_list=[]
            for j in range(0,len(sentence)):
                sentence_text_list.append(sentence[j][0])
                label_list.append(sentence[j][-1])

            token_result=self.tokenizer(
                sentence_text_list,
                max_length=word_max_len,
                truncation=True,is_split_into_words=True)
            
            bert_tokens=self.tokenizer.convert_ids_to_tokens(token_result['input_ids'])
            word_index=token_result.word_ids(batch_index=0)
            ave_len+=len(bert_tokens)
            if len(sentence_text_list)>max_len:
                max_len=len(sentence_text_list)
            if len(bert_tokens)==maxT:
                over_num+=1

            x_index.append(token_result['input_ids'])
            x_seg.append(token_result['token_type_ids'])
            x_mask.append(token_result['attention_mask'])
            
            #print('\nsentence_text_list:',len(sentence_text_list),sentence_text_list)
            #print('\nlabel:',len(label_list),label_list)
            #print('\nword_index:',len(word_index),word_index)
            #print('\nbert_tokens:',len(bert_tokens),bert_tokens)
            label_list,bert_text_label=self.generate_label_list_B(sentence_text_list,label_list,word_index) # the label list after bert token, ori token/lable/new index
            #print('\nlabel list:',len(label_list),label_list)
            #print('\nbert_text_label:',len(bert_text_label),bert_text_label)
            #sys.exit()
            y_list.append(label_list)
            #print(y_list)
            bert_text_labels.append(bert_text_label)

        
        x1_np = pad_sequences(x_index, word_max_len, value=0, padding='post',truncating='post')  # right padding
        x2_np = pad_sequences(x_seg, word_max_len, value=0, padding='post',truncating='post')
        x3_np = pad_sequences(x_mask, word_max_len, value=0, padding='post',truncating='post')
        y_np = pad_sequences(y_list, word_max_len, value=0, padding='post',truncating='post')
        #print('x1_np:',x1_np)
        #print('\nx2_np:',x2_np)
        #print('\ny_np:',y_np)
        #print('\nbert_text:',bert_text_labels)
        # print('bert max len:',max_len,',Over',maxT,':',over_num,'ave len:',ave_len/len(instances),'total:',len(instances))

        if label_type=='softmax':
            y_np = np.expand_dims(y_np, 2)
        elif label_type=='crf':
            pass
        
        return [x1_np, x2_np,x3_np], y_np,bert_text_labels  
         

if __name__ == '__main__':
    pass