sdnet/fewshotreader.py at master · chen700564/sdnet · GitHub

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
from typing import List
from allennlp.data import DatasetReader
from allennlp.data.fields import LabelField, ArrayField, ListField, MetadataField
from allennlp.data.instance import Instance
import json
from transformers import AutoTokenizer
import numpy as np
import random
import copy


class FSNERreader(DatasetReader):
    def __init__(self, file=None,pretrainedfile=None, mapping=None, mode='prompt', lazy=False) -> None:
        super().__init__(lazy)
        self.tokenizer = AutoTokenizer.from_pretrained(pretrainedfile)
        self.mapping = mapping
        self.mode = mode
        if file is not None:
            self.init(file)

    def entity2text(self,entitys,labelmap=None):
        generated = []
        for entity in entitys:
            type = entity['type']
            if self.target_classes is None or type in self.target_classes:
                type = labelmap[type]
                generated.append( entity['text'] + ' is' + ' ' + type + '.')
        return generated

    def prefix_generator(self,instances,typing=False,labelmap=None):
        '''
        instances = labels / spans
        labels: [[Label1,[finelabel1,finelabel2],[Label2,[finelabel3,finelabel4]]
        spans: [span1,span2]
        FET: {Typing | &span1 &span2}
        NER: {NER | & Label1: finelabel1, finelabel2 & Label2: finelabel3, finelabel4}
        { <extra_id_0> 32099
        } <extra_id_1> 32098
        | <extra_id_2> 32097
        : <extra_id_3> 32096
        , <extra_id_4> 32095
        & <extra_id_5> 32094
        NER <extra_id_6> 32093
        FET <extra_id_7> 32092
        '''
        prefix = [32099]
        if typing:
            prefix.append(32092)
            prefix.append(32097)
            for span in instances:
                spanid = [32094] + self.tokenizer.encode(span,add_special_tokens=False)
                prefix = prefix + spanid
        else:
            prefix.append(32093)
            prefix.append(32097)
            for label in instances:
                if self.info.withoutconcepts:
                    labelid = [32094] + self.tokenizer.encode(labelmap[label],add_special_tokens=False)
                    prefix = prefix + labelid
                    continue
                labelid = [32094] + self.tokenizer.encode(labelmap[label[0]],add_special_tokens=False) + [32096]
                labeltoken = [[self.tokenizer.encode(finelabel,add_special_tokens=False),index] for index,finelabel in enumerate(label[1])]
                labeltoken = sorted(labeltoken, key=lambda k:len(k[0]))
                cum = 0
                finelabelids = []
                for finelabel in labeltoken:
                    if len(labelid) + cum + min(len(finelabelids) - 1, 0) + len(finelabel[0]) <= 15:
                        cum += len(finelabel[0])
                        finelabelids.append(finelabel)
                    else:
                        break
                finelabelids = sorted(finelabelids,key = lambda k:k[1])
                finelabelids = [finelabel[0] for finelabel in finelabelids]
                for index,finelabel in enumerate(finelabelids):
                    if index > 0:
                        labelid.append(32095)
                    labelid = labelid + finelabel
                prefix = prefix + labelid
        prefix.append(32098)
        return prefix

    def getinstance(self,data,pred=False,labels=None,typing=False,index=0):
        if (not pred) and self.mode == 'prompt':
            if self.info.withoutconcepts:
                self.target_classes = [i for i in labels]
            else:
                self.target_classes = [i[0] for i in labels]
        text = ' '.join(data['tokens'])
        labelmap = copy.deepcopy(self.mapping)
        if typing:
            entities = data['entity']
            spans = [entity['text'] for entity in entities]
            prefix = self.prefix_generator(spans,True)
            generated = self.entity2text(entities,labelmap)
            generated = ' '.join(generated)
        else:
            generated = self.entity2text(data['entity'],labelmap)
            generated = ' '.join(generated)
            if self.mode == 'prompt':
                prefix = self.prefix_generator(labels,labelmap=labelmap)

        if self.mode == 'prompt' or typing:
            inputid = self.tokenizer.encode(text,max_length=511-len(prefix),truncation=True)
            inputid = prefix + inputid
        else:
            inputid = self.tokenizer.encode(text,max_length=511,truncation=True)
        outputid = [self.tokenizer.pad_token_id] + self.tokenizer.encode(generated,max_length=511,truncation=True)
        labels = outputid[1:]
        inputmask = [1] * len(inputid)
        generated = self.tokenizer.decode(outputid)
        generated = generated.replace('<pad>','')
        generated = generated.replace('</s>','')
        generated = generated.strip()
        outputmask = [1] * (len(outputid) - 1)
        field = {
            'inputid':ArrayField(np.array(inputid)),
            'mask':ArrayField(np.array(inputmask)),
            'outputid':ArrayField(np.array(outputid[:-1])),
            'outmask':ArrayField(np.array(outputmask)),
            'text':MetadataField(text),
            'tokens':MetadataField(data['tokens']),
            'generated':MetadataField(generated),
            'entity':MetadataField(data['entity']),
        }
        if not pred:
            field['label'] = ListField([LabelField(int(i),skip_indexing=True) for i in labels])
        else:
            field['index'] = MetadataField(index)
        if typing:
            field['span'] = MetadataField([[self.tokenizer.decode(self.tokenizer.encode(i['text'],add_special_tokens=False)),i['type']] for i in data['entity']])
        return Instance(field)

    def obtaindatawithtype(self,dataset,target_classes):
        newdataset = []
        for data in dataset:
            entities = []
            for entity in data['entity']:
                if entity['type'] in target_classes:
                    entities.append(entity)
            if len(entities) > 0:
                newdataset.append({'tokens':data['tokens'],'entity':entities})
        return newdataset

    def sampleOneEpoch(self,dataset,typing=False,labels=None,pred=False):
        results = []
        for i in range(len(dataset)):
            if typing:
                results.append(self.getinstance(dataset[i],pred=True,typing=True,index=i))
            else:
                if pred:
                    for singlelabels in labels:
                        results.append(self.getinstance(dataset[i],pred=True,labels=singlelabels,index=i))
                else:
                    for singlelabels in labels:
                        sampledlabel = copy.deepcopy(singlelabels)
                        num = random.choice(list(range(self.info.sample[0],self.info.sample[1] + 1)))
                        sampledlabel = random.sample(sampledlabel,num)
                        results.append(self.getinstance(dataset[i],pred=False,labels=sampledlabel,index=i))
                    random.shuffle(results)
        return results

    def init(self,file):
        self.dataset = []
        with open(file) as f:
            for line in f:
                line = json.loads(line)
                self.dataset.append(line)

    def setinfo(self,info):
        self.info = info

    def _read(self,file):
        if self.info.dataset is not None:
            dataset = copy.deepcopy(self.info.dataset)
        else:
            dataset = copy.deepcopy(self.dataset)
        self.target_classes = self.info.target_classes
        if self.info.pred:
            epochdata = self.sampleOneEpoch(dataset,typing=self.info.typing,labels=self.info.labels,pred=True)
            for data in epochdata:
                yield data
        else:
            while True:
                epochdata = self.sampleOneEpoch(dataset,typing=self.info.typing,labels=self.info.labels,pred=False)
                for data in epochdata:
                    yield data