File size: 6,640 Bytes
ceed500
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# coding=utf-8
#
# This script applies AraBERT's cleaning process and segmentation to ARCD or
# any SQUAD-like structured files and "naively" re-alligns the answers start positions

import sys

sys.path.append("..")

import tensorflow.compat.v1 as tf

import json

from fuzzysearch import find_near_matches
from pyarabic import araby
from tqdm import tqdm

from arabert.arabert.tokenization import BasicTokenizer
from arabert.preprocess import ArabertPreprocessor

flags = tf.flags
FLAGS = flags.FLAGS


## Required parameters
flags.DEFINE_string(
    "input_file", None, "The input json file with a SQUAD like structure."
)

flags.DEFINE_string(
    "output_file", None, "The ouput json file with AraBERT preprocessing applied."
)

flags.DEFINE_string("model_name", None, "Model name same as HuggingFace library")

flags.DEFINE_bool(
    "filter_tydiqa",
    False,
    "If the input dataset is tydiqa, then only process arabic examples",
)


bt = BasicTokenizer(do_lower_case=False)


def clean_preprocess(text, processor):
    text = " ".join(bt._run_split_on_punc(text))
    text = processor.preprocess(text)
    text = " ".join(text.split())  # removes extra whitespaces
    return text


def get_start_pos(old_context, old_answer_start, processor):
    new_context = clean_preprocess(old_context[:old_answer_start], processor)
    num_of_pluses = new_context.count("+")
    return old_answer_start + num_of_pluses * 2 - 20


def main(_):
    tf.logging.set_verbosity(tf.logging.INFO)
    logger = tf.get_logger()
    logger.propagate = False

    print(FLAGS.model_name)
    arabert_prep = ArabertPreprocessor(
        model_name=FLAGS.model_name, remove_html_markup=False
    )

    with tf.gfile.Open(FLAGS.input_file, "r") as reader:
        input_data = json.load(reader)["data"]

    new_answers_count = 0
    no_answers_found_count = 0
    trunc_ans_count = 0
    for entry in tqdm(input_data):
        for paragraph in entry["paragraphs"]:

            if FLAGS.filter_tydiqa:
                # this will only apply farasa segmentation to Arabic Data
                if "arabic" not in paragraph["qas"][0]["id"]:
                    continue
            old_context = paragraph["context"]
            paragraph["context"] = clean_preprocess(paragraph["context"], arabert_prep)
            for qas in paragraph["qas"]:
                qas["question"] = clean_preprocess(qas["question"], arabert_prep)

                for i in range(len(qas["answers"])):
                    temp_text = clean_preprocess(
                        qas["answers"][i]["text"], arabert_prep
                    )

                    if temp_text == "":
                        temp_text = qas["answers"][i]["text"]

                    answer_location = paragraph["context"].find(temp_text)
                    if answer_location == -1:

                        search_start_pos = get_start_pos(
                            old_context, qas["answers"][i]["answer_start"], arabert_prep
                        )
                        search_end_pos = min(
                            len(paragraph["context"]),
                            search_start_pos + len(temp_text) + 20,
                        )
                        answer_match = find_near_matches(
                            temp_text,
                            paragraph["context"][search_start_pos:search_end_pos],
                            max_l_dist=min(10, len(temp_text) // 2),
                        )
                        if len(answer_match) > 0:
                            tf.logging.warning(
                                "Found new answer for question '%s' :\n '%s' \nvs old.\n '%s'\norig:\n'%s'\ncontext:\n'%s'\n==================",
                                qas["id"],
                                answer_match[i].matched,
                                temp_text,
                                qas["answers"][i]["text"],
                                paragraph["context"],
                            )
                            temp_text = answer_match[i].matched
                            qas["answers"][i]["answer_start"] = answer_match[i].start
                            new_answers_count += 1

                        else:
                            tf.logging.warning(
                                "Could not find answer for question '%s' :\n '%s' \nvs.\n '%s'\norig answer:\n '%s'\n==================",
                                qas["id"],
                                paragraph["context"],
                                temp_text,
                                qas["answers"][i]["text"],
                            )
                            qas["answers"][i]["answer_start"] = -1
                            no_answers_found_count += 1
                    else:
                        qas["answers"][i]["answer_start"] = answer_location

                    if len(temp_text) + qas["answers"][i]["answer_start"] < (
                        len(paragraph["context"]) + 1
                    ):
                        qas["answers"][i]["text"] = temp_text
                    else:
                        tf.logging.warning(
                            "answer truncated for question '%s' :\n context:\n'%s' \nanswer:\n '%s'\n orig_answer:\n'%s'\nanswer start: %d\nlength of answer: %d\nlength of paragraph: %d\n=================================",
                            qas["id"],
                            paragraph["context"],
                            temp_text,
                            qas["answers"][i]["text"],
                            qas["answers"][0]["answer_start"],
                            len(temp_text),
                            len(paragraph["context"]),
                        )
                        qas["answers"][0]["text"] = temp_text[
                            0 : len(paragraph["context"])
                            - (len(temp_text) + qas["answers"][0]["answer_start"])
                        ]
                        trunc_ans_count += 1

    tf.logging.warning("Found %d new answers: ", new_answers_count)
    tf.logging.warning("Found %d with no answers: ", no_answers_found_count)
    tf.logging.warning("Found %d with trunc answers: ", trunc_ans_count)

    input_data = {
        "data": input_data,
        "version": "1.1",
        "preprocess": "True",
    }
    with tf.gfile.Open(FLAGS.output_file, "w") as writer:
        json.dump(input_data, writer)


if __name__ == "__main__":
    flags.mark_flag_as_required("input_file")
    flags.mark_flag_as_required("output_file")
    flags.mark_flag_as_required("model_name")
    tf.app.run()