text_match/data_loader.py at main · luogan1234/text_match · 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
import numpy as np
import os
import json
import tqdm
import torch
import torch.nn as nn
import torch.nn.functional as F
import pickle
import random
import copy
from torch.utils.data import Dataset, DataLoader, random_split

class MyDataset(Dataset):
    def __init__(self, file, is_train):
        self.data, self.labels, self.is_train = [], [], is_train
        self.vocabs, self.vocab_num = set(), 0
        with open('data/{}.tsv'.format(file), 'r') as f:
            for line in f:
                items = line.split('\t')
                seq1 = self.to_index(items[0])
                seq2 = self.to_index(items[1])
                self.data.append([seq1, seq2])
                self.vocabs = self.vocabs.union(seq1+seq2)
                self.vocab_num = max([self.vocab_num]+seq1+seq2)
                if is_train:
                    self.labels.append(int(items[2]))
                else:
                    self.labels.append(-1)
        self.vocab_num += 1
        print('Vocab number:', self.vocab_num)

    def to_index(self, seq):
        # [PAD], [unused1]...[unused99], [UNK], [CLS], [SEP], [MASK]
        # 0, 1..99, 100, 101, 102, 103
        seq = [int(v)+104 for v in seq.split(' ')]
        return seq

    def __len__(self):
        return len(self.data)

    def __getitem__(self, idx):
        item = {'seq1': self.data[idx][0], 'seq2': self.data[idx][1], 'label': self.labels[idx]}
        return item

class MyCollation:
    def __init__(self, config, is_train):
        self.config = config
        self.is_train = is_train

    def mask(self, seq):
        res, label_res = [], []
        for token in seq:
            p = random.random()
            if self.is_train and p < 0.15:
                p /= 0.15
                label_res.append(token)
                if p < 0.8:
                    res.append(103)  # [MASK]
                elif p < 0.9:
                    res.append(random.sample(self.config.vocabs, 1)[0])
                else:
                    res.append(token)
            else:
                res.append(token)
                label_res.append(0)
        return res, label_res

    def __call__(self, data):
        inputs, segs, mask_labels, cls_labels = [], [], [], []
        max_len = 0
        for datum in data:
            max_len = max(max_len, len(datum['seq1'])+len(datum['seq2'])+3)
        for datum in data:
            seq1, label_seq1 = self.mask(datum['seq1'])
            seq2, label_seq2 = self.mask(datum['seq2'])
            if not self.is_train or random.randint(0, 1):
                input = [101]+seq1+[102]+seq2+[102]
                seg = [0]*(len(seq1)+2)+[1]*(len(seq2)+1)
                mask_label = [0]+label_seq1+[0]+label_seq2+[0]
            else:
                input = [101]+seq2+[102]+seq1+[102]
                seg = [0]*(len(seq2)+2)+[1]*(len(seq1)+1)
                mask_label = [0]+label_seq2+[0]+label_seq1+[0]
            input += [0]*(max_len-len(input))
            seg += [1]*(max_len-len(seg))
            mask_label += [0]*(max_len-len(mask_label))
            inputs.append(input)
            segs.append(seg)
            mask_labels.append(mask_label)
            cls_labels.append(datum['label'])
        inputs = torch.tensor(inputs, dtype=torch.long).to(self.config.device)
        segs = torch.tensor(segs, dtype=torch.long).to(self.config.device)
        res = {'inputs': inputs, 'segs': segs, 'mask_labels': mask_labels, 'cls_labels': cls_labels}
        return res

class InfiniteDataLoader(DataLoader):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self.iterator = super().__iter__()

    def __iter__(self):
        return self

    def __next__(self):
        try:
            batch = next(self.iterator)
        except StopIteration:
            self.iterator = super().__iter__()
            batch = next(self.iterator)
        return batch

class MyDataLoader:
    def __init__(self, config):
        self.train = MyDataset(config.train, True)
        self.test = MyDataset(config.test, False)
        config.vocabs = self.train.vocabs.union(self.test.vocabs)
        config.vocab_num = max(self.train.vocab_num, self.test.vocab_num, 25000)
        print('Unknown token number in test:', len(self.test.vocabs-self.train.vocabs))
        self.config = config
        self.fn_train = MyCollation(config, True)
        self.fn_eval = MyCollation(config, False)

    def get_train(self):
        n = len(self.train)
        d1, d2 = int(n*0.9), n-int(n*0.9)
        train, valid = random_split(self.train, [d1, d2])
        [test] = random_split(self.test, [len(self.test)])
        valid_unlabel = []
        for datum in valid:
            datum_new = copy.deepcopy(datum)
            datum_new['label'] = -1
            valid_unlabel.append(datum_new)
        train += test+valid_unlabel
        train = InfiniteDataLoader(train, self.config.batch_size(True), shuffle=True, collate_fn=self.fn_train)
        valid = DataLoader(valid, self.config.batch_size(False), shuffle=False, collate_fn=self.fn_eval)
        return train, valid

    def get_all(self):
        data = DataLoader(self.train+self.test, self.config.batch_size(False), shuffle=False, collate_fn=self.fn_eval)
        return data

    def get_predict(self):
        data = DataLoader(self.test, self.config.batch_size(False), shuffle=False, collate_fn=self.fn_eval)
        return data