Hugging Face's logo

Spaces:
nvidia
/
Plan2Align-NV
Sleeping

App Files Community
1
Plan2Align-NV
File size: 4,988 Bytes 
05d3571
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
import argparse
import sys
import typing as tp
import unicodedata

import xxhash
from sacremoses import MosesPunctNormalizer

from .demojizer import Demojizer, legacy_demojizer
from .remove_non_printing_char import \
    get_replacer as non_printing_char_replacer
from .sentence_split import get_split_algo

demojizer = Demojizer()


class SentenceSplitClean:
    def __init__(self, splitter_lang: str, split_algo: str):
        # setup sentence splitter
        self.splitter = get_split_algo(splitter_lang, split_algo=split_algo)

        # setup "moses" normalization
        self.mpn = MosesPunctNormalizer(lang="en", perl_parity=True)  # TODO
        self.replace_nonprint = non_printing_char_replacer(" ")

    def __call__(self, line):
        sentence_splits = self.splitter(line)
        line_hash = xxhash.xxh3_64_intdigest(line)

        for sent in sentence_splits:
            # normalize -- moses equivalent
            clean = self.mpn.normalize(sent)
            clean = self.replace_nonprint(clean)
            # replace π“•π”―π”žπ”«π” π”’π”°π” π”ž by Francesca
            clean = unicodedata.normalize("NFKC", clean)

            yield (line_hash, sent, clean)


def remove_on_unicode_category(x: str) -> str:
    return "".join(filter(lambda ch: not unicodedata.category(ch) in {"So"}, x))


def get_replacer_unicode_category(
    skip_min: int, max_num: int, replace_by: str = " "
) -> str:
    def replace_by_unicode_category(x: str) -> str:
        total_counter = 0
        skip_counter = 0

        def flt(ch):
            nonlocal total_counter
            nonlocal skip_counter
            if max_num == 0 or total_counter < max_num:
                if unicodedata.category(ch) in {"So"}:
                    if skip_counter < skip_min:
                        skip_counter += 1
                        return ch
                    total_counter += 1
                    return replace_by
            return ch

        return "".join(map(flt, x))

    return replace_by_unicode_category


# to map with previous versions of the pipeline
def get_sentence_candidate_modifiers() -> tp.List[tp.Callable]:
    return [
        lambda x: x,
        lambda x: x + " ",
        lambda x: " " + x,
        lambda x: " " + x + " ",
        lambda x: "  " + x,
        lambda x: x.rstrip(),
        lambda x: x.lstrip(),
        lambda x: " " + x.rstrip(),
        lambda x: x.strip(),
        lambda x: demojizer(x, ""),
        lambda x: demojizer(x, "").strip(),
        lambda x: " " + demojizer(x, ""),
        legacy_demojizer,
        remove_on_unicode_category,
        get_replacer_unicode_category(1, 1),
        get_replacer_unicode_category(0, 0),
    ]


def reach_sentence_from_paragraph(
    paragraph: str,
    expected_paragraph_digest: int,
    expected_sentence_digest: int,
    lang: str,
    sentence_splitters: tp.Dict[str, "SentenceSplitClean"],
    debug_candidates: bool,
):
    if lang not in sentence_splitters:
        sentence_splitters[lang] = SentenceSplitClean(lang, "default")

    def no_splitter(paragraph):
        line_h = xxhash.xxh3_64_intdigest(paragraph)
        return [(line_h, paragraph, paragraph)]

    sentence_splitter = sentence_splitters[lang]
    splitter_candidates = [sentence_splitter, no_splitter]
    for duct_candidate in get_sentence_candidate_modifiers():
        for split_cand in splitter_candidates:
            for line_hash, sent, clean in split_cand(paragraph):
                assert line_hash == expected_paragraph_digest
                clean_cand = duct_candidate(clean)
                reached_sentence_digest = xxhash.xxh3_64_intdigest(clean_cand)
                if debug_candidates:
                    print(f"{reached_sentence_digest}::\t::{clean_cand}::")
                if reached_sentence_digest == expected_sentence_digest:
                    return clean_cand

    return None


def split_clean():
    split_algo = "default"
    sentence_splitters = {}

    for line in sys.stdin:
        line_stripped = line.rstrip("\n")
        metadata, paragraph = line_stripped.split("\t")
        (
            _,
            _,
            _,
            _,
            paragraph_digest,
            sentence_digest,
            _,
            _,
            _,
            lang,
            _,
        ) = metadata.split()
        paragraph_digest = int(paragraph_digest)
        sentence_digest = int(sentence_digest)

        sentence = reach_sentence_from_paragraph(
            paragraph,
            paragraph_digest,
            sentence_digest,
            lang,
            sentence_splitters,
            False,
        )

        if sentence is not None:
            print(f"{line_stripped}\t{sentence}")
        else:
            print(
                f"Couldn't match sentence for paragraph: {paragraph_digest} sentence: {sentence_digest} lang: {lang}",
                file=sys.stderr,
            )


def main():
    split_clean()


if __name__ == "__main__":
    main()