Multilingual-Transformers/preprocessing.py at main · codewithzichao/Multilingual-Transformers · 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
import os
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import codecs
import re
from collections import Counter
import json
import math
from transformers import XLMRobertaTokenizer

'''
lang_id={"sanwu":0,"tamil":1,"mal":2,"kan":3}
'''

# 去除emoji
def demoji(text):
    emoji_pattern = re.compile("["
                               u"\U0001F600-\U0001F64F"  # emoticons
                               u"\U0001F300-\U0001F5FF"  # symbols & pictographs
                               u"\U0001F680-\U0001F6FF"  # transport & map symbols
                               u"\U0001F1E0-\U0001F1FF"  # flags (iOS)
                               u"\U00002702-\U000027B0"
                               u"\U000024C2-\U0001F251"
                               u"\U00010000-\U0010ffff"
                               "]+", flags=re.UNICODE)
    return (emoji_pattern.sub(r'', text))


def merge_all_train(train_file_path_list, output_file_path, label_path, label_weight_path,label_freq_path):
    all_label_dict = {}

    idx=1
    for path in train_file_path_list:
        train_df = pd.read_csv(path, encoding="utf-8", header=None, sep="\t", names=["text", "label", "NaN"])
        train_df["text"] = train_df["text"].astype(str)
        train_df["text"] = train_df["text"].apply(lambda x: demoji(x))
        train_df = train_df[["text", "label"]]
        lang_id=[idx if x not in ["not-Tamil","not-malayalam","not-Kannada"] else 0 for x in train_df["label"]]
        idx+=1
        train_df["lang_id"]=lang_id
        print(output_file_path)
        train_df.to_csv(output_file_path, mode="a", encoding="utf-8", header=None, sep="\t", index=False)

        # 处理label
        label_np = train_df["label"].values
        label_dict = Counter(label_np)

        for label, count in label_dict.items():
            if label not in all_label_dict:
                all_label_dict[label] = count
            else:
                all_label_dict[label] += count

    total_num = sum([count for label, count in all_label_dict.items()])
    all_label_weight = {label: math.log(total_num / count) for label, count in all_label_dict.items()}
    all_label_weight["Not-in-indented-language"] = math.log( \
        total_num / (all_label_dict["not-Tamil"] + all_label_dict["not-Kannada"] + all_label_dict["not-malayalam"]))

    all_label_freq={label:(1.0*count/total_num) for label,count in all_label_dict.items()}
    all_label_freq["Not-in-indented-language"]=1.0*(all_label_dict["not-Tamil"] + all_label_dict["not-Kannada"] + all_label_dict["not-malayalam"])/total_num

    with codecs.open(label_path, "w", encoding="utf-8") as f:
        json.dump(all_label_dict, f)
    with codecs.open(label_weight_path, "w", encoding="utf-8") as f:
        json.dump(all_label_weight, f)
    with codecs.open(label_freq_path,"w",encoding="utf-8") as f:
        json.dump(all_label_freq,f)

def merge_all_dev(dev_file_path_list,output_file_path):

    idx=0
    for path in dev_file_path_list:
        dev_df=pd.read_csv(path,sep="\t",encoding="utf-8",header=None,names=["text","label"])
        lang_id=[idx if x not in ["not-Tamil","not-malayalam","not-Kannada"] else 0 for x in dev_df["label"]]
        idx+=1
        dev_df["lang_id"]=lang_id

        dev_df.to_csv(output_file_path,sep="\t",mode="a",encoding="utf-8",header=None,index=False)


def merge_train_dev(train_file_path,dev_file_path,output_file_path):

    train_df=pd.read_csv(train_file_path,sep="\t",encoding="utf-8",header=None)
    train_df.to_csv(output_file_path,mode="a",sep="\t",encoding="utf-8",header=None,index=False)
    dev_df=pd.read_csv(dev_file_path,sep="\t",encoding="utf-8",header=None)
    dev_df.to_csv(output_file_path,mode="a",sep="\t",encoding="utf-8",header=None,index=False)


def longest_text(data_path,name="train"):
    tokenizer=XLMRobertaTokenizer(vocab_file="/Users/codewithzichao/Desktop/competitions/EACL2021/xlm-roberta-base/sentencepiece.bpe.model")
    final_data=dict()

    train_df = pd.read_csv(data_path, encoding="utf-8", header=None, sep="\t", names=["text", "label","NaN"])
    text = train_df["text"].values
    for item in text:
        input_ids=tokenizer.tokenize(item)
        if len(input_ids) in final_data:
            final_data[len(input_ids)]+=1
        else:
            final_data[len(input_ids)]=1

    final_data=dict(sorted(final_data.items(),key=lambda x:x[0],reverse=True))
    with codecs.open("%s_longest.json"%name,"w",encoding="utf-8") as f:
        json.dump(final_data,f,ensure_ascii=False,indent=4)


def convert_label_to_id(label_path):
    with codecs.open(label_path, "r", encoding="utf-8") as f:
        label_dict = json.load(f)

    final_label_dict = {}
    for idx, (label, count) in enumerate(label_dict.items()):
        final_label_dict[label] = idx

    final_label_dict["not-Tamil"] = 1
    final_label_dict["not-malayalam"] = 1
    final_label_dict["not-Kannada"] = 1
    final_label_dict["Not-in-indented-language"] = 1

    return final_label_dict


def process_dev_data(dev_path, output_path):
    dev_df = pd.read_csv(dev_path, sep="\t", encoding="utf-8", header=None, names=["text", "label", "NaN"])
    dev_df = dev_df[["text", "label"]]
    dev_df["text"] = dev_df["text"].astype(str)
    dev_df["text"] = dev_df["text"].apply(lambda x: demoji(x))

    dev_df.to_csv(output_path, sep="\t", encoding="utf-8", header=None, index=False)


def split_train_dev(data,dev_num=5000):

    np.random.shuffle(data)

    dev_data=data[:dev_num]
    train_data=data[dev_num:]

    return train_data,dev_data


if __name__ == "__main__":
    base_path = "/Users/codewithzichao/Desktop/competitions/EACL2021"

    tamil_train_path = base_path + "/data/tamil_offensive_full_train.csv"
    tamil_dev_path = base_path + "/data/tamil_offensive_full_dev.csv"

    mal_train_path = base_path + "/data/mal_full_offensive_train.csv"
    mal_dev_path = base_path + "/data/mal_full_offensive_dev.csv"

    kan_train_path = base_path + "/data/kannada_offensive_train.csv"
    kan_dev_path = base_path + "/data/kannada_offensive_dev.csv"

    total_train_path = base_path + "/data/train.csv"
    tamil_dev_path_output = base_path + "/data/tamil_dev.csv"
    mal_dev_path_output = base_path + "/data/mal_dev.csv"
    kan_dev_path_output = base_path + "/data/kan_dev.csv"

    label_path = base_path + "/data/label.json"
    label_weight_path = base_path + "/data/label_weight.json"
    label_freq_path=base_path+"/data/label_freq.json"

    train_file_path_list = [tamil_train_path, mal_train_path, kan_train_path]
    merge_all_train(train_file_path_list,total_train_path,label_path,label_weight_path,label_freq_path)

    label_dict = convert_label_to_id(label_path)
    print(label_dict)

    process_dev_data(tamil_dev_path, tamil_dev_path_output)
    process_dev_data(mal_dev_path, mal_dev_path_output)
    process_dev_data(kan_dev_path, kan_dev_path_output)

    dev_list=[tamil_dev_path_output,mal_dev_path_output,kan_dev_path_output]
    all_dev_output=base_path+"/data/dev.csv"
    merge_all_dev(dev_list,all_dev_output)

    final_data_path=base_path+"/data/data.csv"
    merge_train_dev(total_train_path,all_dev_output,final_data_path)