File: lm_wrapper.py

package info (click to toggle)
onboard 1.4.1-5
  • links: PTS, VCS
  • area: main
  • in suites: bookworm, bullseye
  • size: 31,548 kB
  • sloc: python: 29,215; cpp: 5,965; ansic: 5,735; xml: 1,026; sh: 163; makefile: 39
file content (698 lines) | stat: -rw-r--r-- 22,008 bytes parent folder | download | duplicates (4)
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
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
#!/usr/bin/python3

# Copyright © 2009-2010, 2012-2014 marmuta <marmvta@gmail.com>
#
# This file is part of Onboard.
#
# Onboard is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 3 of the License, or
# (at your option) any later version.
#
# Onboard is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.

from __future__ import division, print_function, unicode_literals

import sys
import re
import codecs
from math import log

import pypredict.lm as lm
from pypredict.lm import overlay, linint, loglinint  # exported symbols

class _BaseModel:

    modified = False
    load_error = False
    load_error_msg = ""

    def learn_tokens(self, tokens, allow_new_words=True):
        """ Extract n-grams from tokens and count them. """
        for ngram in self._extract_ngrams(tokens):
            self.count_ngram(ngram, 1, allow_new_words)

        self.modified = True

    def _extract_ngrams(self, tokens):
        """
        Extract n-grams from tokens.

        Doctests:
        >>> m = DynamicModel(3)
        >>> list(m._extract_ngrams(["word1", "word2", "<unk>", "word3"]))
        [['word1'], ['word1', 'word2'], ['word2'], ['word3']]
        >>> list(m._extract_ngrams(["word1", "word2", "<s>", "word3"]))
        [['word1'], ['word1', 'word2'], ['word2'], ['<s>'], ['<s>', 'word3'], ['word3']]
        >>> list(m._extract_ngrams(["a", "b", "c", "d"]))
        [['a'], ['a', 'b'], ['a', 'b', 'c'], ['b'], ['b', 'c'], ['b', 'c', 'd'], ['c'], ['c', 'd'], ['d']]
        """
        token_sections = []

        # Don't let <unk> enter the model.
        # Split the token stream into sections between <unk>s.
        unk_sections = split_tokens(tokens, "<unk>")
        for section in unk_sections:
            # Don't learn across sentence marks.
            token_sections.extend(split_tokens(section, "<s>", True))

        # Run a window of size <order> along the section and return n-grams.
        for token_section in token_sections:
            section = token_section

            for i,token in enumerate(section):
                for n in range(self.order):
                    if i+n+1 <= len(section):
                        ngram = section[i:i+n+1]
                        assert(n == len(ngram)-1)
                        yield ngram

    def get_counts(self):
        """
        Return number of n-gram types and total occurances
        for each n-gram level.
        """
        counts = [0]*self.order
        totals = [0]*self.order
        for ng in self.iter_ngrams():
            counts[len(ng[0])-1] +=  1
            totals[len(ng[0])-1] += ng[1]
        return counts, totals

    def copy(self, model):
        """
        Copy contents of self to model. The order of the destination
        stays unchanged.
        """
        if hasattr(self, "smoothing"): # not for UnigramModel
            model.smoothing = self.smoothing

        for it in self.iter_ngrams():
            ngram = it[0]
            count = it[1]
            model.count_ngram(ngram, count)

        return model

    def prune(self, prune_counts):
        """
        Return a copy of self with all ngrams removed whose
        count is less or equal to <prune_count>.

        prune_count==-1  # prune all frequencies
        prune_count=0    # prune nothing
        prune_count>0    # prune frequencies below or equal prune_count
        """
        # drop order for to be emptied n-gram levels
        order = self.order
        for prune_count in reversed(prune_counts):
            if prune_count != -1:
                break
            order -= 1

        order = max(order, 2)
        model = self.__class__(order)

        if hasattr(self, "smoothing"): # not for UnigramModel
            model.smoothing = self.smoothing

        for it in self.iter_ngrams():
            ngram = it[0]
            count = it[1]

            level = len(ngram)
            k = min(len(prune_counts), level) - 1
            prune_count = prune_counts[k]

            if count > prune_count and  prune_count != -1:
                model.count_ngram(ngram, count)

        return model

    def load(self, filename):
        self.load_error = False
        self.load_error_msg = ""
        self.modified = False
        try:
            super(_BaseModel, self).load(filename)
        except IOError as e:
            self.load_error = True
            raise e

    def remove_context(self, context):
        """
        Remove word context[-1] where it appears after history context[:-1]
        from the model. If the history is empty all n-grams containing word
        will be removed.
        """
        changes = self.get_remove_context_changes(context)
        if changes:
            for ngram, count in changes.items():
                self.count_ngram(ngram, count)

            self.modified = True

        return changes

    def get_remove_context_changes(self, context):
        """
        Simulate removal of context.
        Returns a dict of affected n-grams and their count changes (negative).
        """
        changes = {}

        for it in self.iter_ngrams():
            ngram = it[0]
            count = it[1]

            # find intersection with context
            for i in range(len(ngram)):
                for j in range(min(len(context), i+1)):
                    if ngram[i-j] != context[-j-1]:
                        break
                else:
                    if j == len(context) - 1:
                        changes[ngram] = -count
                        break

        return changes


class LanguageModel(_BaseModel, lm.LanguageModel):
    """
    Abstract class representing the base class of all models.
    Keep this for access to class constants.
    """
    def __init__(self):
        raise NotImplementedError()

class UnigramModel(_BaseModel, lm.UnigramModel):
    pass


class DynamicModel(_BaseModel, lm.DynamicModel):
    pass


class DynamicModelKN(_BaseModel, lm.DynamicModelKN):
    pass


class CachedDynamicModel(_BaseModel, lm.CachedDynamicModel):
    pass


def split_tokens(tokens, separator, keep_separator = False):
    """
    Split list of tokens at separator token.

    Doctests:
    # excluding separator
    >>> split_tokens(["<unk>", "word1", "word2", "word3"], "<unk>")
    [['word1', 'word2', 'word3']]
    >>> split_tokens(["word1", "<unk>", "word2", "word3"], "<unk>")
    [['word1'], ['word2', 'word3']]
    >>> split_tokens(["word1", "word2", "word3", "<unk>"], "<unk>")
    [['word1', 'word2', 'word3']]

    # including separator
    >>> split_tokens(["<unk>", "word1", "word2", "word3"], "<unk>", True)
    [['<unk>', 'word1', 'word2', 'word3']]
    >>> split_tokens(["word1", "<unk>", "word2", "word3"], "<unk>", True)
    [['word1'], ['<unk>', 'word2', 'word3']]
    >>> split_tokens(["word1", "word2", "word3", "<unk>"], "<unk>", True)
    [['word1', 'word2', 'word3']]
    """
    token_sections = []
    token_section = []
    for token in tokens:
        if token == separator:
            if token_section:
                token_sections.append(token_section)

            if keep_separator:
                token_section = [separator]
            else:
                token_section = []
        else:
            token_section.append(token)

    if len(token_section) > 1 or \
       (token_section and token_section[0] != separator):
        token_sections.append(token_section)

    return token_sections

def split_tokens_at(tokens, split_indices):
    """
    Patition tokens with splits at the given indices.
    split_indices must be sorted in ascending order.

    Doctests:
    >>> test = split_tokens_at

    >>> test(["word0", "word1", "word2"], [])
    [['word0', 'word1', 'word2']]

    >>> test(["word0", "word1", "word2"], [0])
    [['word1', 'word2']]

    >>> test(["word0", "word1", "word2"], [1])
    [['word0'], ['word2']]

    >>> test(["word0", "word1", "word2"], [2])
    [['word0', 'word1']]

    >>> test(["word0", "word1", "word2"], [0, 2])
    [['word1']]

    >>> test(["word0", "word1", "word2"], [0, 1, 2])
    []

    >>> test(["word0", "word1", "word2", "word3", "word4"], [0, 2, 4])
    [['word1'], ['word3']]

    # out of range indices
    >>> test(["word0", "word1", "word2"], [100, 1000])
    [['word0', 'word1', 'word2']]
    """
    token_sections = []
    remaining = 0

    for i in split_indices:
        section = tokens[remaining:i]
        if section:
            token_sections.append(section)
        remaining = i+1

    section = tokens[remaining:]
    if section:
        token_sections.append(section)

    return token_sections


SENTENCE_PATTERN = re.compile( \
    """ .*?
           (?:
                 (?:[.;:!?](?:(?=[\s]) | \")) # punctuation
               | (?:\\s*\\n\\s*)+(?=[\\n])    # multiples newlines
               | <s>                          # sentence end mark
           )
         | .+$                                # last sentence fragment
    """, re.UNICODE|re.DOTALL|re.VERBOSE)

def split_sentences(text, disambiguate=False):
    """ Split text into sentences. """

    # Remove carriage returns from Moby Dick.
    # Don't change the text's length, keep it in sync with spans.
    filtered = text.replace("\r"," ")

    # split into sentence fragments
    matches = SENTENCE_PATTERN.finditer(filtered)

    # filter matches
    sentences = []
    spans = []
    for match in matches:
        sentence = match.group()
        # not only newlines? remove fragments with only double newlines
        if True: #not re.match("^\s*\n+\s*$", sentence, re.UNICODE):
            begin = match.start()
            end   = match.end()

            # strip whitespace including newlines
            l = len(sentence)
            sentence = sentence.lstrip()
            begin += l - len(sentence)

            l = len(sentence)
            sentence = sentence.rstrip()
            end -= l - len(sentence)

            # remove <s>
            sentence = re.sub("<s>", "   ", sentence)

            # remove newlines and double spaces - no, invalidates spans
            #sentence = re.sub(u"\s+", u" ", sentence)

            # strip whitespace from the cuts, remove carriage returns
            l = len(sentence)
            sentence = sentence.rstrip()
            end -= l - len(sentence)
            l = len(sentence)
            sentence = sentence.lstrip()
            begin += l - len(sentence)

            # add <s> sentence separators if the end of the sentence is
            # ambiguous - required by the split_corpus tool where the
            # result of split_sentences is saved to a text file and later
            # fed back to split_sentences again.
            if disambiguate:
                if not re.search("[.;:!?]\"?$", sentence, re.UNICODE):
                    sentence += " <s>"

            sentences.append(sentence)
            spans.append([begin, end])

    return sentences, spans


tokenize_pattern = """
    (                                     # <unk>
      (?:^|(?<=\s))
        \S*(\S)\\2{{3,}}\S*               # char repeated more than 3 times
        | [-]{{3}}                        # dash repeated more than 2 times
      (?=\s|$)
      | :[^\s:@]+?@                       # password in URL
    ) |
    (                                     # <num>
      (?:[-+]?\d+(?:[.,]\d+)*)            # anything numeric looking
      | (?:[.,]\d+)
    ) |
    (                                     # word
      (?:[-]{{0,2}}                       # allow command line options
        [^\W\d]\w*(?:[-'´΄][\w]+)*        # word, not starting with a digit
        [{trailing_characters}'´΄]?)
      | <unk> | <s> | </s> | <num>        # pass through control words
      | <bot:[a-z]*>                      # pass through begin of text merkers
      | (?:^|(?<=\s))
          (?:
            \| {standalone_operators}     # common space delimited operators
          )
        (?=\s|$)
    )
    """
# Don't learn "-" or "--" as standalone tokens...
TEXT_PATTERN = re.compile(tokenize_pattern.format(
                          trailing_characters = "",
                          standalone_operators = ""),
                          re.UNICODE|re.DOTALL|re.VERBOSE)
# ...but recognize them in a prediction context as start of a cmd line option.
CONTEXT_PATTERN = re.compile(tokenize_pattern.format(
                          trailing_characters = "-",
                          standalone_operators = "| [-]{1,2}"),
                          re.UNICODE|re.DOTALL|re.VERBOSE)

def tokenize_sentence(sentence, is_context = False):

    if is_context:
        matches = CONTEXT_PATTERN.finditer(sentence)
    else:
        matches = TEXT_PATTERN.finditer(sentence)
    tokens = []
    spans = []

    for match in matches:
        groups = match.groups()
        if groups[3]:
            tokens.append(groups[3])
            spans.append(match.span())
        elif groups[2]:
            tokens.append("<num>")
            spans.append(match.span())
        elif groups[0]:
            tokens.append("<unk>")
            spans.append(match.span())

    return tokens, spans

def tokenize_text(text, is_context = False):
    """ Split text into word tokens.
        The result is ready for use in learn_tokens().

        Sentence begins, if detected, are marked with "<s>".
        Numbers are replaced with the number marker <num>.
        Other tokens that could confuse the prediction are
        replaced with the unknown word marker "<unk>".

        Examples, text -> tokens:
            "We saw whales"  -> ["We", "saw", "whales"]
            "We saw whales " -> ["We", "saw", "whales"]
            "Hello there! We saw 5 whales "
                             -> ["Hello", "there", "<s>",
                                 "We", "saw", "<num>", "whales"]
    """

    tokens = []
    spans = []
    sentences, sentence_spans = split_sentences(text)
    for i, sentence in enumerate(sentences):
        ts, ss = tokenize_sentence(sentence, is_context)

        sbegin = sentence_spans[i][0]
        ss = [[s[0]+sbegin, s[1]+sbegin] for s in ss]

        # sentence begin?
        if i > 0:
            tokens.append("<s>")      # prepend sentence begin marker
            spans.append([sbegin, sbegin]) # empty span
        tokens.extend(ts)
        spans.extend(ss)

    return tokens, spans

def tokenize_context(text):
    """ Split text into word tokens + completion prefix.
        The result is ready for use in predict().
    """
    tokens, spans = tokenize_text(text, is_context = True)
    if not re.match("""
                  ^$                             # empty string?
                | .*[-'´΄\w]$                    # word at the end?
                | (?:^|.*\s)[|]=?$               # recognized operator?
                | .*(\S)\\1{3,}$                 # anything repeated > 3 times?
                """, text, re.UNICODE|re.DOTALL|re.VERBOSE):
        tokens.append("")
        tend = len(text)
        spans.append([tend, tend]) # empty span

    return tokens, spans

def read_order(filename, encoding=None):
    """
    Read the order from the header of the given file.
    Encoding may be 'utf-8', 'latin-1'.
    """
    order = None

    if sys.version_info >= (3, 3):
        ex_class = FileNotFoundError
    else:
        ex_class = IOError

    try:
        text = read_corpus(filename, encoding, 20)
    except ex_class as ex:
        return None

    lines = text.split("\n")
    data = False
    for line in lines:
        if line.startswith("\\data\\"):
            data = True
            continue

        if data:  # data section?
            result = re.search("ngram (\d+)=\d+", line)
            if result:
                if order is None:
                    order = 0
                order = max(order, int(result.groups()[0]))

            if line.startswith("\\"):  # end of data section?
                break

    return order

def read_corpus(filename, encoding=None, num_lines = None):
    """ Read corpus, encoding may be 'utf-8', 'latin-1'. """

    if encoding:
        encodings = [encoding]
    else:
        encodings = ['utf-8', 'latin-1']

    for i,enc in enumerate(encodings):
        try:
            if num_lines is None:
                text = codecs.open(filename, encoding=enc).read()
            else:
                text = ""
                with codecs.open(filename, encoding=enc) as f:
                    for i in range(num_lines):
                        t = f.readline()
                        if not t:
                            break
                        text += t
        except UnicodeDecodeError as err:
            if i == len(encodings)-1: # all encodings failed?
                raise err
            continue   # silently retry with the next encoding
        break

    return text

def read_vocabulary(filename, encoding=None):
    """
    Read vocabulary with one word per line.
    Encoding may be 'utf-8', 'latin-1', like read_corpus.
    """
    text = read_corpus(filename, encoding)
    vocabulary = text.split("\n")

    for ctrl_word in ["<unk>", "<s>", "</s>", "</num>"]:
        if not ctrl_word in vocabulary:
            vocabulary.append(ctrl_word)

    return vocabulary


def extract_vocabulary(tokens, min_count=1, max_words=0):
    """ Extract the most frequent <max_words> words from <tokens>. """
    m = {}
    for t in tokens:
        m[t] = m.get(t, 0) + 1
    items = [x for x in list(m.items()) if x[1] >= min_count]
    items = sorted(items, key=lambda x: x[1], reverse=True)
    if max_words:
        return items[:max_words]
    else:
        return items

def filter_tokens(tokens, vocabulary):
    v = set(vocabulary)
    return [t if t in v else "<unk>" for t in tokens]

def entropy(model, tokens, order=None):

    if not order:
        order = model.order  # fails for non-ngram models, specify order manually

    ngram_count = 0
    entropy = 0
    word_count = len(tokens)

    # extract n-grams of maximum length
    for i in range(len(tokens)):
        b = max(i-(order-1),0)
        e = min(i-(order-1)+order, len(tokens))
        ngram = tokens[b:e]
        if len(ngram) != 1:
            p = model.get_probability(ngram)
            if p == 0:
                print(word_count, ngram,p)
            e = log(p, 2) if p else float("infinity")
            entropy += e
            ngram_count += 1

    entropy = -entropy/word_count if word_count else 0
    try:
        perplexity = 2 ** entropy
    except:
        perplexity = 0

    return entropy, perplexity


def ksr(query_model, learn_model, sentences, limit, progress=None):
    """ Calculate keystroke savings rate from simulated typing. """
    total_chars, pressed_keys = simulate_typing(query_model, learn_model, sentences, limit, progress)
    saved_keystrokes = total_chars - pressed_keys
    return saved_keystrokes * 100.0 / total_chars if total_chars else 0

def simulate_typing(query_model, learn_model, sentences, limit, progress=None):

    total_chars = 0
    pressed_keys = 0

    for i,sentence in enumerate(sentences):
        inputline = ""

        cursor = 0
        while cursor < len(sentence):
            context, spans = tokenize_context(". " + inputline) # simulate sentence begin
            prefix = context[len(context)-1] if context else ""
            prefix_to_end = sentence[len(inputline)-len(prefix):]
            target_word = re.search("^([\w]|[-'])*", prefix_to_end, re.UNICODE).group()
            choices = query_model.predict(context, limit)

            if 0:  # step mode for debugging
                print("cursor=%d total_chars=%d pressed_keys=%d" % (cursor, total_chars, pressed_keys))
                print("sentence= '%s'" % sentence)
                print("inputline='%s'" % inputline)
                print("prefix='%s'" % prefix)
                print("prefix_to_end='%s'" % prefix_to_end)
                print("target_word='%s'" % (target_word))
                print("context=", context)
                print("choices=", choices)
                input()

            if target_word in choices:
                added_chars = len(target_word) - len(prefix)
                if added_chars == 0: # still right after insertion point?
                    added_chars = 1  # continue with next character
            else:
                added_chars = 1

            for k in range(added_chars):
                inputline += sentence[cursor]
                cursor += 1
                total_chars += 1

            pressed_keys += 1

        # learn the sentence
        if learn_model:
            tokens, spans = tokenize_context(sentence)
            learn_model.learn_tokens(tokens)

        # progress feedback
        if progress:
            progress(i, len(sentences), total_chars, pressed_keys)

    return total_chars, pressed_keys


from contextlib import contextmanager

@contextmanager
def timeit(s, out=sys.stdout):
    import time, gc

    if out:
        gc.collect()
        gc.collect()
        gc.collect()

        t = time.time()
        text = s if s else "timeit"
        out.write("%-15s " % text)
        out.flush()
        yield None
        out.write("%10.3fms\n" % ((time.time() - t)*1000))
    else:
        yield None




if __name__ == '__main__':
    a = [".", ". ", " . ", "a. ", "a. b"]
    for text in a:
        print("split_sentences('%s'): %s" % (text, repr(split_sentences(text))))

    for text in a:
        print("tokenize_text('%s'): %s" % (text, repr(tokenize_text(text))))

    for text in a:
        print("tokenize_context('%s'): %s" % (text, repr(tokenize_context(text))))