aboutsummaryrefslogtreecommitdiffstats
path: root/fuzzycat/cluster.py
blob: d859507245aeab7e2fcd6ba193ebe4b870bdd171 (plain)
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
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
# pylint: disable=C0103
"""
Clustering stage.

* [x] verify needs whole document
* [ ] parallelization misses groups
* [ ] cached match key store (tsv, sqlite3), something ~/.cache/...
* [x] reproducibly run tests
* [x] place for put md/tsv record tests

----

* [x] hadoop -> py (bn)
* [ ] gnu parallel, share command line -- note (bn)

----

Ideas:

* lookup potential matches; TSV [key, ...]; sort
* maybe new "schema" - size vs "common schema" -- key <TAB> {"bibjson": ...}
* merge-join

```
$ python -m fuzzycat keygen -s "algo" < ours | sort -k1,1 > a.tsv
$ python -m fuzzycat keygen -s "algo" < other | sort -k1,1 > b.tsv
$ merge-join a.tsv b.tsv
```

A couple of "keygen" algos.

> 10k/s, 1B, ~day

Partial fields should be ok.

Q:

* nysiis

Deps.

* pydantic; json "omitempty" -- get rid of it?
* orjson (serialize datetime) -- maybe enough w/ dataclasses w/ dataclasses

fuzzycat.main -> `__main__.py`

* elasticsearch-py >> elasticsearch

Matching releases to non-release entities.

----

Features and integration.

* work grouping at import time; random pdfs; requires strong verification (vs cgraph)
* email out to OCI

"""

import collections
import fileinput
import itertools
import json
import logging
import multiprocessing
import operator
import os
import re
import string
import subprocess
import sys
import tempfile
import unicodedata
from dataclasses import dataclass
from typing import IO, Any, Callable, Dict, Generator, List, Optional, Tuple

import fuzzy
import regex
import zstandard

from fuzzycat.utils import cut, shellout, slugify_string, zstdlines

__all__ = [
    "release_key_title",
    "release_key_title_normalized",
    "release_key_title_nysiis",
    "release_key_title_sandcrawler",
    "Cluster",
]


@dataclass
class KeyDoc:
    """
    A document from which we can derive a key, e.g. a release entity.
    """
    ident: str
    title: str


get_ident_title = operator.itemgetter("ident", "title")
ws_replacer = str.maketrans({"\t": " ", "\n": " "})
non_word_re = re.compile(r'[\W_]+', re.UNICODE)

# Notes: untie from release_entity, as we are only using a few fields. Maybe
# it's a jsob blob, with a pydantic spec and schema.


def release_key_title(doc: KeyDoc) -> Tuple[str, str]:
    ident, title = get_ident_title(doc)
    if not title:
        raise ValueError('title missing for {}'.format(ident))
    title = title.translate(ws_replacer).strip()
    return (ident, title)


def release_key_title_normalized(doc: KeyDoc) -> Tuple[str, str]:
    ident, title = release_key_title(doc)
    title = re.sub(r'[ ]{2,}', ' ', title).lower()
    return (ident, non_word_re.sub('', title))


def release_key_title_nysiis(doc: KeyDoc) -> Tuple[str, str]:
    """
    Use NYSIIS New York State Identification and Intelligence System.
    """
    ident, title = release_key_title(doc)
    return (ident, fuzzy.nysiis(title))


# from http://zderadicka.eu/removing-diacritics-marks-from-strings/
SANDCRAWLER_CHAR_MAP = {
    '\N{Latin capital letter AE}': 'AE',
    '\N{Latin small letter ae}': 'ae',
    '\N{Latin capital letter Eth}': 'D',
    '\N{Latin small letter eth}': 'd',
    '\N{Latin capital letter O with stroke}': 'O',
    '\N{Latin small letter o with stroke}': 'o',
    '\N{Latin capital letter Thorn}': 'Th',
    '\N{Latin small letter thorn}': 'th',
    '\N{Latin small letter sharp s}': 's',
    '\N{Latin capital letter D with stroke}': 'D',
    '\N{Latin small letter d with stroke}': 'd',
    '\N{Latin capital letter H with stroke}': 'H',
    '\N{Latin small letter h with stroke}': 'h',
    '\N{Latin small letter dotless i}': 'i',
    '\N{Latin small letter kra}': 'k',
    '\N{Latin capital letter L with stroke}': 'L',
    '\N{Latin small letter l with stroke}': 'l',
    '\N{Latin capital letter Eng}': 'N',
    '\N{Latin small letter eng}': 'n',
    '\N{Latin capital ligature OE}': 'Oe',
    '\N{Latin small ligature oe}': 'oe',
    '\N{Latin capital letter T with stroke}': 'T',
    '\N{Latin small letter t with stroke}': 't',

    # bnewbold additions
    '\N{MICRO SIGN}': 'u',
    '\N{LATIN SMALL LETTER C}': 'c',
    '\N{LATIN SMALL LETTER F WITH HOOK}': 'f',
    # bnewbold map-to-null (for non-printing stuff not in the regex)
    '\N{PARTIAL DIFFERENTIAL}': '',
    '\N{LATIN LETTER INVERTED GLOTTAL STOP}': '',
    '\N{N-ARY SUMMATION}': '',
    '\N{N-ARY PRODUCT}': '',
    '\N{MODIFIER LETTER CIRCUMFLEX ACCENT}': '',
    '\N{SNOWMAN}': '',
    '\N{CARON}': '',
}

SANDCRAWLER_PREFIX_REMOVE = [
    "original article: ",
    "original article ",
    "article: ",
    "title: ",
]

# regex that matches all characters which should be removed
SANDCRAWLER_REMOVE_CHAR_REGEX = regex.compile(
    r"[\s\p{Punctuation}\p{M}\p{InCombiningDiacriticalMarks}\u2000-\u206F\u2E00-\u2E7F’·“”‘’“”«»「」¿–±§_`°ʖ©®¤=<>|+$^~≈√∫≤≥÷ƒ∆¬£¢∞¥◊€]"
)


def sandcrawler_slugify(raw: str) -> str:
    """
    Python re-implementation of sandcrawler Scala code for string comparison
    ("scorable" strings)
    """
    slug = raw.strip().lower()

    # transforms before running regex
    for prefix in SANDCRAWLER_PREFIX_REMOVE:
        if slug.startswith(prefix):
            slug = slug[:len(prefix)]

    slug = slug.replace("&apos;", "'")

    # iterate over all chars and replace from map, if in map; then lower-case again
    slug = ''.join([SANDCRAWLER_CHAR_MAP.get(c, c) for c in slug])

    # early bailout before executing regex
    if not slug:
        return ""

    slug = unicodedata.normalize('NFKD', slug)
    slug = SANDCRAWLER_REMOVE_CHAR_REGEX.sub('', slug)

    return slug.lower()


def test_sandcrawler_slugify() -> None:
    test_cases = [
        ("", ""),
        ("asdf", "asdf"),
        ("'Hello World!'", "helloworld"),
        ("ASDF", "asdf"),
        ("as\n  df", "asdf"),
        ("as\u0142  bb \u00f8", "aslbbo"),
        ("`hello¿", "hello"),
        ("علمية", "علمية"),
        ("期刊的数字", "期刊的数字"),
        ("les pré-impressions explorées à partir", "lespreimpressionsexploreesapartir"),

        # "MICRO SIGN"
        ("\xb5meter", "umeter"),
        # "GREEK SMALL LETTER MU"
        ("\u03bcmeter", "\u03bcmeter"),

        # TODO: ("salt &and; pepper", "saltpepper"),
        # TODO: ("new <b>and</b> improved", "newandimproved"),

        # some via https://github.com/minimaxir/big-list-of-naughty-strings/blob/master/blns.txt
        ("-9223372036854775808/-1", "92233720368547758081"),
        (r",./;'[]\-= <>?:\"{}|_+ !@#$%^&*()`~", ""),
        (" \n\r \x85 \u1680\u2002\u2003\u2002\u2003\u2004\u2005\u2006\u2007\u2008\u2009\u200a\u200b\u202f\u205f\u3000",
         ""),
        (r"Ω≈ç√∫˜≤≥÷", "ωc"),
        (r"åß∂ƒ©˙∆˚¬…æ", "asfae"),
        (r"œ∑´®†¥¨ˆøπ“‘", "oeoπ"),
        (r"¡™£¢∞§¶•ªº–≠ ", "tmao"),
        (r"¸˛Ç◊ı˜Â¯˘¿", "cia"),
        (r"ÅÍÎÏ˝ÓÔÒÚÆ☃", "aiiiooouae"),
        (r"Œ„´‰ˇÁ¨ˆØ∏”’", "oeao"),
        (r"`⁄€‹›fifl‡°·‚—±", "fifl"),
        (r"ЁЂЃЄЅІЇЈЉЊЋЌЍЎЏАБВГДЕЖЗИЙКЛМНОПРСТУФХЦЧШЩЪЫЬЭЮЯабвгдежзийклмнопрстуфхцчшщъыьэюя",
         "еђгєѕііјљњћкиуџабвгдежзииклмнопрстуфхцчшщъыьэюяабвгдежзииклмнопрстуфхцчшщъыьэюя"),
        (r"⁰⁴⁵₀₁₂", "045012"),
        (r"社會科學院語學研究所", "社會科學院語學研究所"),
        # TODO: ("パーティーへ行かないか", "パーティーへ行かないか"),
        # TODO: ("表ポあA鷗ŒéB逍Üߪąñ丂㐀𠀀", "表ポあa鷗oeebB逍usaan丂㐀𠀀"),
        (r"( ͡° ͜ʖ ͡°)", ""),
        # emoji ok? I guess
        (r"👾 🙇 💁 🙅 🙆 🙋 🙎 🙍", "👾🙇💁🙅🙆🙋🙎🙍"),
        (r"2️⃣ 3️⃣ 4️⃣ 5️⃣", "2345"),
        (r"﷽ ", "﷽"),
        (r"\"̗̺͖̹̯͓Ṯ̤͍̥͇͈h̲́e͏͓̼̗̙̼̣͔ ͇̜̱̠͓͍ͅN͕͠e̗̱z̘̝̜̺͙p̤̺̹͍̯͚e̠̻̠͜r̨̤͍̺̖͔̖̖d̠̟̭̬̝͟i̦͖̩͓͔̤a̠̗̬͉̙n͚͜ ̻̞̰͚ͅh̵͉i̳̞v̢͇ḙ͎͟-҉̭̩̼͔m̤̭̫i͕͇̝̦n̗͙ḍ̟ ̯̲͕͞ǫ̟̯̰̲͙̻̝f ̪̰̰̗̖̭̘͘c̦͍̲̞͍̩̙ḥ͚a̮͎̟̙͜ơ̩̹͎s̤.̝̝ ҉Z̡̖̜͖̰̣͉̜a͖̰͙̬͡l̲̫̳͍̩g̡̟̼̱͚̞̬ͅo̗͜.̟",
         "thenezperdianhivemindofchaoszalgo"),
        (r"The quick brown fox jumps over the lazy dog", "thequickbrownfoxjumpsoverthelazydog"),
        (r"The quick brown fox jumps over the lazy dog", "thequickbrownfoxjumpsoverthelazydog"),
        (r"𝕋𝕙𝕖 𝕢𝕦𝕚𝕔𝕜 𝕓𝕣𝕠𝕨𝕟 𝕗𝕠𝕩 𝕛𝕦𝕞𝕡𝕤 𝕠𝕧𝕖𝕣 𝕥𝕙𝕖 𝕝𝕒𝕫𝕪 𝕕𝕠𝕘 ", "thequickbrownfoxjumpsoverthelazydog"),
    ]

    for in_str, out_str in test_cases:
        if sandcrawler_slugify(in_str) != out_str:
            for c in list(sandcrawler_slugify(in_str)):
                try:
                    print(unicodedata.name(c))
                except ValueError:
                    print(ord(c))
                #print(ord(c))
            print("----")
            for c in list(out_str):
                print(unicodedata.name(c))
            print(in_str)
        assert sandcrawler_slugify(in_str) == out_str


def release_key_title_sandcrawler(doc: KeyDoc) -> Tuple[str, str]:
    ident, title = release_key_title(doc)
    slug = sandcrawler_slugify(title)
    return (ident, slug)


def release_key_title_ngram(doc: KeyDoc, n=3) -> Tuple[str, str]:
    """
    Derive a key from title.

    Tokenize title, remote stopwords, lookup first three, lookup last three,
    plus authors. TODO(miku): authors.
    """
    ident, title = get_ident_title(doc)
    slug_title = slugify_string(title)
    tokens = slug_title.split()
    if len(tokens) < 2 * n:
        key = ''.join(tokens)
    else:
        key = ''.join(tokens[:3] + tokens[-3:])
    return (ident, key)


class Cluster:
    """
    Setup and run clustering over a potentially large (100m) number of records.

    Two main options are iterable (TODO: work on parsed docs), and the key
    function to apply to value to group by.

    TODO: We want compression.
    """
    def __init__(self,
                 iterable: collections.abc.Iterable,
                 key: Callable[[Any], Tuple[str, str]],
                 output: IO[str] = sys.stdout,
                 key_denylist: Optional[List[str]] = None,
                 prefix: str = "fuzzycat-",
                 tmpdir: str = tempfile.gettempdir(),
                 strict: bool = False,
                 min_cluster_size: int = 2,
                 max_cluster_size: int = 100,
                 compress=False,
                 verbose=True):
        self.iterable: collections.abc.Iterable = iterable
        self.key: Callable[[Any], Tuple[str, str]] = key
        self.output: IO[str] = output
        self.prefix: str = prefix
        self.tmpdir: str = tmpdir
        self.strict = strict
        self.key_denylist = key_denylist
        self.min_cluster_size = min_cluster_size
        self.max_cluster_size = max_cluster_size
        self.verbose = verbose
        self.compress = compress
        self.counter: Dict[str, int] = collections.Counter({
            "key_fail": 0,
            "key_ok": 0,
            "key_empty": 0,
            "key_denylist": 0,
            "num_clusters": 0,
        })

    def run(self):
        """
        First map documents to keys, then group by keys, outline: json -> tsv
        -> sort -> group -> json.
        """
        with tempfile.NamedTemporaryFile(delete=False, mode="wb", prefix=self.prefix) as tf:
            if self.compress:
                zc = zstandard.ZstdCompressor(level=9, threads=multiprocessing.cpu_count())
                writer = zc.stream_writer(tf)
            else:
                writer = tf
            for i, line in enumerate(self.iterable):
                if self.verbose and i % 100000 == 0:
                    print("@{}".format(i), file=sys.stderr)
                try:
                    doc = json.loads(line)
                    id, key = self.key(doc)
                except (KeyError, ValueError):
                    if self.strict:
                        raise
                    self.counter["key_fail"] += 1
                    continue
                if not key:
                    self.counter["key_empty"] += 1
                    continue
                if self.key_denylist and key in self.key_denylist:
                    self.counter["key_denylist"] += 1
                    continue
                self.counter["key_ok"] += 1
                # XXX: if the line itself contains tabs, we need to remove
                # them here; maybe offer TSV and JSON output and extra flag
                # XXX: this needs to be compressed (e.g. with 2B records, we
                # fill up disk too quickly)
                data = bytes("{}\t{}\t{}\n".format(id, key,
                                                   line.replace("\t", " ").strip()),
                             encoding="utf-8")
                writer.write(data)
            if self.compress:
                writer.flush(zstandard.FLUSH_FRAME)

        sf = self.sort(tf.name, opts='-k 2')
        if self.compress:
            f = zstdlines(sf)
        else:
            f = open(sf)

        for doc in self.group_by(f, key=cut(f=1)):
            if len(doc["v"]) < self.min_cluster_size:
                continue
            self.counter["num_clusters"] += 1
            json.dump(doc, self.output)
            self.output.write("\n")

        os.remove(sf)
        os.remove(tf.name)
        return self.counter

    def sort(self, filename: str, opts: str = "-k 2", fast: bool = True, mode: str = "w"):
        """
        Sort tabular file with sort(1), returns the filename of the sorted
        file. Options to sort can be passed in via opts keyword argument.
        """
        with tempfile.NamedTemporaryFile(delete=False, mode=mode, prefix=self.prefix) as tf:
            env = os.environ.copy()
            env["TMPDIR"] = self.tmpdir
            if fast:
                env["LC_ALL"] = "C"
            if self.compress:
                output = shellout(
                    "zstdcat -T0 {input} | LC_ALL=C TMPDIR={tmpdir} sort {opts} | zstd -T0 -c9 > {output}",
                    input=filename,
                    tmpdir=self.tmpdir,
                    opts=opts)
            else:
                subprocess.run(["sort"] + opts.split() + [filename], stdout=tf, env=env, check=True)
                output = tf.name

        return output

    def group_by(self,
                 seq: collections.abc.Iterable,
                 key: Callable[[Any], str] = None) -> Generator[Any, None, None]:
        """
        Extract a key from elements of an iterable and group them. Just as
        uniq(1), the input iterable must be ordered (by the key that is
        extracted) for this to work.

        There might be large clusters, which would currently exceed memory,
        hence the max_cluster_size option.
        """
        for k, g in itertools.groupby(seq, key=key):
            payload = []
            for i, line in enumerate(g):
                if i > 0 and i == self.max_cluster_size:
                    print('max cluster size cut off for: {}'.format(k), file=sys.stderr)
                    break
                # XXX: This is a bit too much "serde", get rid of this.
                fields = line.split("\t")
                if len(fields) < 3:
                    continue
                payload.append(json.loads(fields[2]))
            doc = {
                "k": k.strip(),
                "v": payload,
            }
            yield doc