-
-
Notifications
You must be signed in to change notification settings - Fork 4.4k
/
tfidfmodel.py
551 lines (463 loc) · 21.3 KB
/
tfidfmodel.py
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
#!/usr/bin/env python
# -*- coding: utf-8 -*-
#
# Copyright (C) 2012 Radim Rehurek <radimrehurek@seznam.cz>
# Copyright (C) 2017 Mohit Rathore <mrmohitrathoremr@gmail.com>
# Licensed under the GNU LGPL v2.1 - http://www.gnu.org/licenses/lgpl.html
"""This module implements functionality related to the `Term Frequency - Inverse Document Frequency
<https://en.wikipedia.org/wiki/Tf%E2%80%93idf>` vector space bag-of-words models.
For a more in-depth exposition of TF-IDF and its various SMART variants (normalization, weighting schemes),
see the blog post at https://rare-technologies.com/pivoted-document-length-normalisation/
"""
import logging
from functools import partial
import re
import numpy as np
from gensim import interfaces, matutils, utils
from gensim.utils import deprecated
logger = logging.getLogger(__name__)
def resolve_weights(smartirs):
"""Check the validity of `smartirs` parameters.
Parameters
----------
smartirs : str
`smartirs` or SMART (System for the Mechanical Analysis and Retrieval of Text)
Information Retrieval System, a mnemonic scheme for denoting tf-idf weighting
variants in the vector space model. The mnemonic for representing a combination
of weights takes the form ddd, where the letters represents the term weighting of the document vector.
for more information visit `SMART Information Retrieval System
<https://en.wikipedia.org/wiki/SMART_Information_Retrieval_System>`_.
Returns
-------
str of (local_letter, global_letter, normalization_letter)
local_letter : str
Term frequency weighing, one of:
* `b` - binary,
* `t` or `n` - raw,
* `a` - augmented,
* `l` - logarithm,
* `d` - double logarithm,
* `L` - log average.
global_letter : str
Document frequency weighting, one of:
* `x` or `n` - none,
* `f` - idf,
* `t` - zero-corrected idf,
* `p` - probabilistic idf.
normalization_letter : str
Document normalization, one of:
* `x` or `n` - none,
* `c` - cosine,
* `u` - pivoted unique,
* `b` - pivoted character length.
Raises
------
ValueError
If `smartirs` is not a string of length 3 or one of the decomposed value
doesn't fit the list of permissible values.
"""
if isinstance(smartirs, str) and re.match(r"...\....", smartirs):
match = re.match(r"(?P<ddd>...)\.(?P<qqq>...)", smartirs)
raise ValueError(
"The notation {ddd}.{qqq} specifies two term-weighting schemes, "
"one for collection documents ({ddd}) and one for queries ({qqq}). "
"You must train two separate tf-idf models.".format(
ddd=match.group("ddd"),
qqq=match.group("qqq"),
)
)
if not isinstance(smartirs, str) or len(smartirs) != 3:
raise ValueError("Expected a string of length 3 got " + smartirs)
w_tf, w_df, w_n = smartirs
if w_tf not in 'btnaldL':
raise ValueError("Expected term frequency weight to be one of 'btnaldL', got {}".format(w_tf))
if w_df not in 'xnftp':
raise ValueError("Expected inverse document frequency weight to be one of 'xnftp', got {}".format(w_df))
if w_n not in 'xncub':
raise ValueError("Expected normalization weight to be one of 'xncub', got {}".format(w_n))
# resolve aliases
if w_tf == "t":
w_tf = "n"
if w_df == "x":
w_df = "n"
if w_n == "x":
w_n = "n"
return w_tf + w_df + w_n
def df2idf(docfreq, totaldocs, log_base=2.0, add=0.0):
r"""Compute inverse-document-frequency for a term with the given document frequency `docfreq`:
:math:`idf = add + log_{log\_base} \frac{totaldocs}{docfreq}`
Parameters
----------
docfreq : {int, float}
Document frequency.
totaldocs : int
Total number of documents.
log_base : float, optional
Base of logarithm.
add : float, optional
Offset.
Returns
-------
float
Inverse document frequency.
"""
return add + np.log(float(totaldocs) / docfreq) / np.log(log_base)
def precompute_idfs(wglobal, dfs, total_docs):
"""Pre-compute the inverse document frequency mapping for all terms.
Parameters
----------
wglobal : function
Custom function for calculating the "global" weighting function.
See for example the SMART alternatives under :func:`~gensim.models.tfidfmodel.smartirs_wglobal`.
dfs : dict
Dictionary mapping `term_id` into how many documents did that term appear in.
total_docs : int
Total number of documents.
Returns
-------
dict of (int, float)
Inverse document frequencies in the format `{term_id_1: idfs_1, term_id_2: idfs_2, ...}`.
"""
# not strictly necessary and could be computed on the fly in TfidfModel__getitem__.
# this method is here just to speed things up a little.
return {termid: wglobal(df, total_docs) for termid, df in dfs.items()}
def smartirs_wlocal(tf, local_scheme):
"""Calculate local term weight for a term using the weighting scheme specified in `local_scheme`.
Parameters
----------
tf : int
Term frequency.
local : {'b', 'n', 'a', 'l', 'd', 'L'}
Local transformation scheme.
Returns
-------
float
Calculated local weight.
"""
if local_scheme == "n":
return tf
elif local_scheme == "l":
return 1 + np.log2(tf)
elif local_scheme == "d":
return 1 + np.log2(1 + np.log2(tf))
elif local_scheme == "a":
return 0.5 + (0.5 * tf / tf.max(axis=0))
elif local_scheme == "b":
return tf.astype('bool').astype('int')
elif local_scheme == "L":
return (1 + np.log2(tf)) / (1 + np.log2(tf.mean(axis=0)))
def smartirs_wglobal(docfreq, totaldocs, global_scheme):
"""Calculate global document weight based on the weighting scheme specified in `global_scheme`.
Parameters
----------
docfreq : int
Document frequency.
totaldocs : int
Total number of documents.
global_scheme : {'n', 'f', 't', 'p'}
Global transformation scheme.
Returns
-------
float
Calculated global weight.
"""
if global_scheme == "n":
return 1.0
elif global_scheme == "f":
return np.log2(1.0 * totaldocs / docfreq)
elif global_scheme == "t":
return np.log2((totaldocs + 1.0) / docfreq)
elif global_scheme == "p":
return max(0, np.log2((1.0 * totaldocs - docfreq) / docfreq))
@deprecated("Function will be removed in 4.0.0")
def smartirs_normalize(x, norm_scheme, return_norm=False):
"""Normalize a vector using the normalization scheme specified in `norm_scheme`.
Parameters
----------
x : numpy.ndarray
The tf-idf vector.
norm_scheme : {'n', 'c'}
Document length normalization scheme.
return_norm : bool, optional
Return the length of `x` as well?
Returns
-------
numpy.ndarray
Normalized array.
float (only if return_norm is set)
Norm of `x`.
"""
if norm_scheme == "n":
if return_norm:
_, length = matutils.unitvec(x, return_norm=return_norm)
return x, length
else:
return x
elif norm_scheme == "c":
return matutils.unitvec(x, return_norm=return_norm)
class TfidfModel(interfaces.TransformationABC):
"""Objects of this class realize the transformation between word-document co-occurrence matrix (int)
into a locally/globally weighted TF-IDF matrix (positive floats).
Examples
--------
.. sourcecode:: pycon
>>> import gensim.downloader as api
>>> from gensim.models import TfidfModel
>>> from gensim.corpora import Dictionary
>>>
>>> dataset = api.load("text8")
>>> dct = Dictionary(dataset) # fit dictionary
>>> corpus = [dct.doc2bow(line) for line in dataset] # convert corpus to BoW format
>>>
>>> model = TfidfModel(corpus) # fit model
>>> vector = model[corpus[0]] # apply model to the first corpus document
"""
def __init__(self, corpus=None, id2word=None, dictionary=None, wlocal=utils.identity,
wglobal=df2idf, normalize=True, smartirs=None, pivot=None, slope=0.25):
r"""Compute TF-IDF by multiplying a local component (term frequency) with a global component
(inverse document frequency), and normalizing the resulting documents to unit length.
Formula for non-normalized weight of term :math:`i` in document :math:`j` in a corpus of :math:`D` documents
.. math:: weight_{i,j} = frequency_{i,j} * log_2 \frac{D}{document\_freq_{i}}
or, more generally
.. math:: weight_{i,j} = wlocal(frequency_{i,j}) * wglobal(document\_freq_{i}, D)
so you can plug in your own custom :math:`wlocal` and :math:`wglobal` functions.
Parameters
----------
corpus : iterable of iterable of (int, int), optional
Input corpus
id2word : {dict, :class:`~gensim.corpora.Dictionary`}, optional
Mapping token - id, that was used for converting input data to bag of words format.
dictionary : :class:`~gensim.corpora.Dictionary`
If `dictionary` is specified, it must be a `corpora.Dictionary` object and it will be used.
to directly construct the inverse document frequency mapping (then `corpus`, if specified, is ignored).
wlocals : callable, optional
Function for local weighting, default for `wlocal` is :func:`~gensim.utils.identity`
(other options: :func:`numpy.sqrt`, `lambda tf: 0.5 + (0.5 * tf / tf.max())`, etc.).
wglobal : callable, optional
Function for global weighting, default is :func:`~gensim.models.tfidfmodel.df2idf`.
normalize : {bool, callable}, optional
Normalize document vectors to unit euclidean length? You can also inject your own function into `normalize`.
smartirs : str, optional
SMART (System for the Mechanical Analysis and Retrieval of Text) Information Retrieval System,
a mnemonic scheme for denoting tf-idf weighting variants in the vector space model.
The mnemonic for representing a combination of weights takes the form XYZ,
for example 'ntc', 'bpn' and so on, where the letters represents the term weighting of the document vector.
Term frequency weighing:
* `b` - binary,
* `t` or `n` - raw,
* `a` - augmented,
* `l` - logarithm,
* `d` - double logarithm,
* `L` - log average.
Document frequency weighting:
* `x` or `n` - none,
* `f` - idf,
* `t` - zero-corrected idf,
* `p` - probabilistic idf.
Document normalization:
* `x` or `n` - none,
* `c` - cosine,
* `u` - pivoted unique,
* `b` - pivoted character length.
Default is 'nfc'.
For more information visit `SMART Information Retrieval System
<https://en.wikipedia.org/wiki/SMART_Information_Retrieval_System>`_.
pivot : float or None, optional
In information retrieval, TF-IDF is biased against long documents [1]_. Pivoted document length
normalization solves this problem by changing the norm of a document to `slope * old_norm + (1.0 -
slope) * pivot`.
You can either set the `pivot` by hand, or you can let Gensim figure it out automatically with the following
two steps:
* Set either the `u` or `b` document normalization in the `smartirs` parameter.
* Set either the `corpus` or `dictionary` parameter. The `pivot` will be automatically determined from
the properties of the `corpus` or `dictionary`.
If `pivot` is None and you don't follow steps 1 and 2, then pivoted document length normalization will be
disabled. Default is None.
See also the blog post at https://rare-technologies.com/pivoted-document-length-normalisation/.
slope : float, optional
In information retrieval, TF-IDF is biased against long documents [1]_. Pivoted document length
normalization solves this problem by changing the norm of a document to `slope * old_norm + (1.0 -
slope) * pivot`.
Setting the `slope` to 0.0 uses only the `pivot` as the norm, and setting the `slope` to 1.0 effectively
disables pivoted document length normalization. Singhal [2]_ suggests setting the `slope` between 0.2 and
0.3 for best results. Default is 0.25.
See also the blog post at https://rare-technologies.com/pivoted-document-length-normalisation/.
See Also
--------
~gensim.sklearn_api.tfidf.TfIdfTransformer : Class that also uses the SMART scheme.
resolve_weights : Function that also uses the SMART scheme.
References
----------
.. [1] Singhal, A., Buckley, C., & Mitra, M. (1996). `Pivoted Document Length
Normalization <http://singhal.info/pivoted-dln.pdf>`_. *SIGIR Forum*, 51, 176–184.
.. [2] Singhal, A. (2001). `Modern information retrieval: A brief overview <http://singhal.info/ieee2001.pdf>`_.
*IEEE Data Eng. Bull.*, 24(4), 35–43.
"""
self.id2word = id2word
self.wlocal, self.wglobal, self.normalize = wlocal, wglobal, normalize
self.num_docs, self.num_nnz, self.idfs = None, None, None
self.smartirs = resolve_weights(smartirs) if smartirs is not None else None
self.slope = slope
self.pivot = pivot
self.eps = 1e-12
if smartirs:
n_tf, n_df, n_n = self.smartirs
self.wlocal = partial(smartirs_wlocal, local_scheme=n_tf)
self.wglobal = partial(smartirs_wglobal, global_scheme=n_df)
if dictionary:
# user supplied a Dictionary object, which already contains all the
# statistics we need to construct the IDF mapping. we can skip the
# step that goes through the corpus (= an optimization).
if corpus:
logger.warning(
"constructor received both corpus and explicit inverse document frequencies; ignoring the corpus"
)
self.num_docs, self.num_nnz = dictionary.num_docs, dictionary.num_nnz
self.cfs = dictionary.cfs.copy()
self.dfs = dictionary.dfs.copy()
self.term_lens = {termid: len(term) for termid, term in dictionary.items()}
self.idfs = precompute_idfs(self.wglobal, self.dfs, self.num_docs)
if not id2word:
self.id2word = dictionary
elif corpus:
self.initialize(corpus)
else:
# NOTE: everything is left uninitialized; presumably the model will
# be initialized in some other way
pass
# If smartirs is not None, override pivot and normalize
if not smartirs:
return
if self.pivot is not None:
if n_n in 'ub':
logger.warning("constructor received pivot; ignoring smartirs[2]")
return
if n_n in 'ub' and callable(self.normalize):
logger.warning("constructor received smartirs; ignoring normalize")
if n_n in 'ub' and not dictionary and not corpus:
logger.warning("constructor received no corpus or dictionary; ignoring smartirs[2]")
elif n_n == "u":
self.pivot = 1.0 * self.num_nnz / self.num_docs
elif n_n == "b":
self.pivot = 1.0 * sum(
self.cfs[termid] * (self.term_lens[termid] + 1.0) for termid in dictionary.keys()
) / self.num_docs
@classmethod
def load(cls, *args, **kwargs):
"""Load a previously saved TfidfModel class. Handles backwards compatibility from
older TfidfModel versions which did not use pivoted document normalization.
"""
model = super(TfidfModel, cls).load(*args, **kwargs)
if not hasattr(model, 'pivot'):
model.pivot = None
logger.info('older version of %s loaded without pivot arg', cls.__name__)
logger.info('Setting pivot to %s.', model.pivot)
if not hasattr(model, 'slope'):
model.slope = 0.65
logger.info('older version of %s loaded without slope arg', cls.__name__)
logger.info('Setting slope to %s.', model.slope)
if not hasattr(model, 'smartirs'):
model.smartirs = None
logger.info('older version of %s loaded without smartirs arg', cls.__name__)
logger.info('Setting smartirs to %s.', model.smartirs)
return model
def __str__(self):
return "TfidfModel(num_docs=%s, num_nnz=%s)" % (self.num_docs, self.num_nnz)
def initialize(self, corpus):
"""Compute inverse document weights, which will be used to modify term frequencies for documents.
Parameters
----------
corpus : iterable of iterable of (int, int)
Input corpus.
"""
logger.info("collecting document frequencies")
dfs = {}
numnnz, docno = 0, -1
for docno, bow in enumerate(corpus):
if docno % 10000 == 0:
logger.info("PROGRESS: processing document #%i", docno)
numnnz += len(bow)
for termid, _ in bow:
dfs[termid] = dfs.get(termid, 0) + 1
# keep some stats about the training corpus
self.num_docs = docno + 1
self.num_nnz = numnnz
self.cfs = None
self.dfs = dfs
self.term_lengths = None
# and finally compute the idf weights
self.idfs = precompute_idfs(self.wglobal, self.dfs, self.num_docs)
self.add_lifecycle_event(
"initialize",
msg=(
f"calculated IDF weights for {self.num_docs} documents and {max(dfs.keys()) + 1 if dfs else 0}"
f" features ({self.num_nnz} matrix non-zeros)"
),
)
def __getitem__(self, bow, eps=1e-12):
"""Get the tf-idf representation of an input vector and/or corpus.
bow : {list of (int, int), iterable of iterable of (int, int)}
Input document in the `sparse Gensim bag-of-words format
<https://radimrehurek.com/gensim/intro.html#core-concepts>`_,
or a streamed corpus of such documents.
eps : float
Threshold value, will remove all position that have tfidf-value less than `eps`.
Returns
-------
vector : list of (int, float)
TfIdf vector, if `bow` is a single document
:class:`~gensim.interfaces.TransformedCorpus`
TfIdf corpus, if `bow` is a corpus.
"""
self.eps = eps
# if the input vector is in fact a corpus, return a transformed corpus as a result
is_corpus, bow = utils.is_corpus(bow)
if is_corpus:
return self._apply(bow)
# unknown (new) terms will be given zero weight (NOT infinity/huge weight,
# as strict application of the IDF formula would dictate)
termid_array, tf_array = [], []
for termid, tf in bow:
termid_array.append(termid)
tf_array.append(tf)
tf_array = self.wlocal(np.array(tf_array))
vector = [
(termid, tf * self.idfs.get(termid))
for termid, tf in zip(termid_array, tf_array) if abs(self.idfs.get(termid, 0.0)) > self.eps
]
# and finally, normalize the vector either to unit length, or use a
# user-defined normalization function
if self.smartirs:
n_n = self.smartirs[2]
if n_n == "n" or (n_n in 'ub' and self.pivot is None):
if self.pivot is not None:
_, old_norm = matutils.unitvec(vector, return_norm=True)
norm_vector = vector
elif n_n == "c":
if self.pivot is not None:
_, old_norm = matutils.unitvec(vector, return_norm=True)
else:
norm_vector = matutils.unitvec(vector)
elif n_n == "u":
_, old_norm = matutils.unitvec(vector, return_norm=True, norm='unique')
elif n_n == "b":
old_norm = sum(freq * (self.term_lens[termid] + 1.0) for termid, freq in bow)
else:
if self.normalize is True:
self.normalize = matutils.unitvec
elif self.normalize is False:
self.normalize = utils.identity
if self.pivot is not None:
_, old_norm = self.normalize(vector, return_norm=True)
else:
norm_vector = self.normalize(vector)
if self.pivot is None:
norm_vector = [(termid, weight) for termid, weight in norm_vector if abs(weight) > self.eps]
else:
pivoted_norm = (1 - self.slope) * self.pivot + self.slope * old_norm
norm_vector = [
(termid, weight / float(pivoted_norm))
for termid, weight in vector
if abs(weight / float(pivoted_norm)) > self.eps
]
return norm_vector