-
Notifications
You must be signed in to change notification settings - Fork 2
/
beacon.py
1161 lines (1055 loc) · 46.9 KB
/
beacon.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
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
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
"""
Business Establishment Automated Classification of NAICS (BEACON)
"""
# Authors: Brian Dumbacher <[email protected]>
# Daniel Whitehead <[email protected]>
# Jiseok Jeong <[email protected]>
# Sarah Pfeiff <[email protected]>
import io
import numpy as np
import re
import time
from numbers import Integral, Real
from sklearn.base import BaseEstimator, ClassifierMixin
from sklearn.utils._param_validation import Interval
from sklearn.utils.validation import check_X_y, check_array, check_random_state, check_is_fitted
def load_naics_data(vintage="2017", shuffle=False, random_state=0):
"""
Load NAICS data
This method loads example NAICS datasets for fitting a BEACON classification model.
Parameters
----------
vintage : str, default="2017"
Vintage of NAICS data. Valid values are "2017" and "2022".
shuffle : boolean, default=False
Flag indicating whether to shuffle the observations.
random_state : int or RandomState instance, default=0
The seed of the pseudo random number generator used to shuffle the observations.
Pass an int for reproducible output across multiple function calls.
Used only if shuffle=True.
Returns
-------
X : numpy.ndarray
1D NumPy array of strings representing business descriptions.
y : numpy.ndarray
1D NumPy array of strings representing 6-digit NAICS codes.
sample_weight : numpy.ndarray
1D NumPy array of numerical sample weights.
"""
# Check vintage
if not isinstance(vintage, str):
raise ValueError("Parameter 'vintage' is not a string.")
if vintage not in ("2017", "2022"):
raise ValueError("Parameter 'vintage' is invalid. Valid values are '2017' and '2022'.")
# Prepare for reading data
file = "example_data_{}.txt".format(vintage)
data_tups = []
n_vars = 0
line_number = 0
len_error_flag = False
# Read data line by line
# io.open() checks for the existence of the data file and raises FileNotFoundError accordingly
f = io.open(file, "r")
for line in f:
line_number += 1
line_strip = line.strip()
if line_number == 1:
if line_strip != "":
var_names = line_strip.split("|")
if (var_names != ["TEXT", "NAICS"] and var_names != ["TEXT", "NAICS", "SAMPLE_WEIGHT"]):
raise ValueError("Input data file does not have the expected format: TEXT|NAICS|SAMPLE_WEIGHT (SAMPLE_WEIGHT optional).")
n_vars = len(var_names)
else:
raise ValueError("No variable names appear on the first line of input data file.")
else:
if line_strip != "":
row_data = line_strip.split("|")
if len(row_data) != n_vars:
len_error_flag = True
data_tups.append(tuple(row_data))
f.close()
# Check data
n_obs = len(data_tups)
if n_obs == 0:
raise ValueError("Input data file contains zero observations.")
if len_error_flag:
raise ValueError("Input data file contains observations with inconsistent numbers of variables.")
if shuffle:
# check_random_state() is provided by sklearn.utils.validation
random_state = check_random_state(random_state)
uniform_random = random_state.uniform(0, 1, n_obs)
data_tups_random = [(data_tups[i], uniform_random[i]) for i in range(n_obs)]
data_tups_random.sort(key=lambda z: z[1])
data_tups = [z[0] for z in data_tups_random]
# Create X, y, and sample_weight
X = np.array([tup[0] for tup in data_tups])
y = np.array([tup[1] for tup in data_tups])
if n_vars == 2:
sample_weight = np.ones(X.shape[0])
elif n_vars == 3:
# float() checks whether the sample weights can be converted to type float and raises ValueError accordingly
sample_weight = np.array([float(tup[2]) for tup in data_tups])
return X, y, sample_weight
class BeaconModel(BaseEstimator, ClassifierMixin):
"""
BEACON text classification model for predicting 6-digit NAICS codes
Business Establishment Automated Classification of NAICS (BEACON) was developed by the U.S. Census Bureau to help
respondents to economic surveys self-classify their establishment's primary business activity in real time. BEACON is
based on natural language processing, machine learning, and information retrieval.
The methodology presented here is a simplified version of what the Census Bureau uses in production. The production
version of BEACON employs a detailed text cleaning algorithm with tens of thousands of rules and a large, rich
dataset compiled from various public and confidential sources. This Python program demonstrates how one can
implement a simple BEACON model as an extension of scikit-learn's (Pedregosa et al., 2011) BaseEstimator and
ClassifierMixin classes. Large sections of BEACON's codebase for processing natural language and cleaning text come
from Snowball (Porter, 2001) and NLTK (Bird, 2006).
Parameters
----------
freq_thresh : int, default=1
Training data frequency threshold for determining whether to include a feature in the dictionary. Does not take
sample weights into account.
wt_umb : float, default=0.6
Ensemble weight of the "umbrella" sub-model.
wt_exact : float, default=0.3
Ensemble weight of the "exact" sub-model.
verbose : int, default=0
Verbosity indicator. Any positive value turns on messages during model fitting.
Attributes
----------
naics_ : list
List of unique 6-digit NAICS codes in data
n_naics_ : int
Number of unique 6-digit NAICS codes in data
sectors_ : list
List of unique sectors in data
n_sectors_ : int
Number of unique sectors in data
sample_sizes_ : dict
Dictionary of sample sizes by sector
naics_indices_ : dict
Dictionary of NAICS index mappings by sector
dict_ncombs_props_ : dict
Dictionary of n-comb features by sector
dict_ncombs_weights_ : dict
Dictionary of n-comb purity weights by sector
dict_ems_props_ : dict
Dictionary of exact features by sector
dict_ems_weights_ : dict
Dictionary of exact feature purity weights by sector
References
----------
Bird, S. (2006). NLTK: The Natural Language Toolkit. Proceedings of the COLING/ACL 2006 Interactive Presentation Sessions.
Sydney, Australia: Association for Computational Linguistics, 69-72.
Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R.,
Dubourg, V., Vanderplas, J., Passos, A., Cournapeau, D., Brucher, M., Perrot, M., and Duchesnay, E. (2011).
Scikit-learn: Machine Learning in Python. Journal of Machine Learning Research, 12, 2825-2830.
Porter, M.F. (2001). Snowball: a language for stemming algorithms. <http://snowball.tartarus.org/texts/introduction.html>.
[Online; accessed 11 March 2024].
"""
# Parameter contraints
# Interval is provided by sklearn.utils._param_validation
# Integral and Real are provided by numbers
_parameter_constraints: dict = {
"freq_thresh": [Interval(Integral, 1, None, closed="left")],
"wt_umb": [Interval(Real, 0, 1, closed="both")],
"wt_exact": [Interval(Real, 0, 1, closed="both")],
"verbose": ["verbose"],
}
# Initialize BeaconModel object
def __init__(self, freq_thresh=1, wt_umb=0.6, wt_exact=0.3, verbose=0):
self.freq_thresh = freq_thresh
self.wt_umb = wt_umb
self.wt_exact = wt_exact
self.verbose = verbose
# Validate parameters
def __validate_parameters(self):
# BeaconModel inherits the method _validate_params() from the parent class BaseEstimator
# _validate_params() checks the parameter values against _parameter_constraints defined above
self._validate_params()
if (self.wt_umb + self.wt_exact > 1):
raise ValueError("Sum of parameters 'wt_umb' and 'wt_exact' is greater than 1: {} + {} = {}.".format(self.wt_umb, self.wt_exact, self.wt_umb + self.wt_exact))
return
# Validate data: X, y, and sample_weight
"""
Validate data
This is the main method for validating input data
Parameters
----------
X : 1D array-like data structure of strings representing business descriptions
y : 1D array-like data structure of strings representing 6-digit NAICS codes
sample_weight : 1D array-like data structure of numbers representing sample weights
Returns
-------
X : numpy.ndarray
1D NumPy array of strings representing business descriptions
y : numpy.ndarray
1D NumPy array of strings representing 6-digit NAICS codes
sample_weight : numpy.ndarray
1D NumPy array of numbers representing sample weights
"""
def __validate_data(self, X, y, sample_weight):
# check_X_y() is provided by sklearn.utils.validation
X = self.__validate_data_X(X)
y = self.__validate_data_y(y)
X, y = check_X_y(X, y, dtype=str, ensure_2d=False)
# Check dimensionality of 'X' and 'y' again
if len(X.shape) > 1:
raise ValueError("Input 'X' is not 1D.")
if len(y.shape) > 1:
raise ValueError("Input 'y' is not 1D.")
if not isinstance(sample_weight, type(None)):
sample_weight = self.__validate_data_sample_weight(sample_weight)
if sample_weight.shape[0] != X.shape[0]:
raise ValueError("Input 'sample_weight' has an inconsistent number of observations: {}.".format(sample_weight.shape[0]))
else:
sample_weight = np.ones(X.shape[0])
return X, y, sample_weight
# Validate data: X only
def __validate_data_X(self, X):
if isinstance(X, str):
raise ValueError("Input 'X' must be a 1D array-like data structure of strings.")
# check_array() is provided by sklearn.utils.validation
X = check_array(X, dtype=str, ensure_2d=False)
if len(X.shape) > 1:
raise ValueError("Input 'X' is not 1D.")
return X
# Validate data: y only
def __validate_data_y(self, y):
if isinstance(y, str):
raise ValueError("Input 'y' must be a 1D array-like data structure of strings.")
# check_array() is provided by sklearn.utils.validation
y = check_array(y, dtype=str, ensure_2d=False)
if len(y.shape) > 1:
raise ValueError("Input 'y' is not 1D.")
return y
# Validate data: sample_weight only
def __validate_data_sample_weight(self, sample_weight):
if isinstance(sample_weight, int) or isinstance(sample_weight, float):
raise ValueError("Input 'sample_weight' must be a 1D array-like data structure of numbers.")
# check_array() is provided by sklearn.utils.validation
sample_weight = check_array(sample_weight, dtype="numeric", ensure_2d=False)
if len(sample_weight.shape) > 1:
raise ValueError("Input 'sample_weight' is not 1D.")
return sample_weight
# Stop words
__stop_words = (
# ADD YOUR OWN STOP WORDS HERE
# Below are examples based on NLTK
"a",
"am",
"an",
"and",
"are",
"as",
"but",
"by",
"for",
"from",
"i",
"if",
"in",
"is",
"it",
"on",
"or",
"other",
"since",
"so",
"the",
"this",
"to",
"we",
"with",
"you",
)
# NLTK implementation of the Porter 2/Snowball stemming algorithm with slight modifications
__vowels = "aeiouy"
__double_consonants = ("bb", "dd", "ff", "gg", "mm", "nn", "pp", "rr", "tt")
__li_ending = "cdeghkmnrt"
__step1a_suffixes = ("sses", "ied", "ies", "us", "ss", "s")
__step1b_suffixes = ("eedly", "ingly", "edly", "eed", "ing", "ed")
__step2_suffixes = (
"ization",
"ational",
"fulness",
"ousness",
"iveness",
"tional",
"biliti",
"lessli",
"entli",
"ation",
"alism",
"aliti",
"ousli",
"iviti",
"fulli",
"enci",
"anci",
"abli",
"izer",
"ator",
"alli",
"bli",
"ogi",
"li",
)
__step3_suffixes = (
"ational",
"tional",
"alize",
"icate",
"iciti",
"ative",
"ical",
"ness",
"ful",
)
__step4_suffixes = (
"ement",
"ance",
"ence",
"able",
"ible",
"ment",
"ant",
"ent",
"ism",
"ate",
"iti",
"ous",
"ive",
"ize",
"ion",
"al",
"er",
"ic",
)
__step5_suffixes = ("e", "l")
__step6_suffixes = (
"curist",
"graphi",
"logi",
"logist",
"nomi",
"nomist",
"pathi",
"pathet",
"physicist",
"scopi",
"therapeut",
"therapi",
"therapist",
"tomi",
"tomist",
"tri",
"trist",
"trician",
"turist",
)
__special_words = {
# ADD YOUR OWN STEMMING RULES HERE FOR CORRECTING OVERSTEMMING ERRORS
# Below are examples from Porter (2001) and Bird (2006)
"skis": "ski",
"skies": "sky",
"dying": "die",
"lying": "lie",
"tying": "tie",
"idly": "idl",
"gently": "gentl",
"ugly": "ugli",
"early": "earli",
"only": "onli",
"singly": "singl",
"sky": "sky",
"news": "news",
"howe": "howe",
"atlas": "atlas",
"cosmos": "cosmos",
"bias": "bias",
"andes": "andes",
"inning": "inning",
"innings": "inning",
"outing": "outing",
"outings": "outing",
"canning": "canning",
"cannings": "canning",
"herring": "herring",
"herrings": "herring",
"earring": "earring",
"earrings": "earring",
"proceed": "proceed",
"proceeds": "proceed",
"proceeded": "proceed",
"proceeding": "proceed",
"exceed": "exceed",
"exceeds": "exceed",
"exceeded": "exceed",
"exceeding": "exceed",
"succeed": "success",
"succeeds": "success",
"succeeded": "success",
"succeeding": "success",
}
def __r1r2(self, word, vowels):
r1 = ""
r2 = ""
for i in range(1, len(word)):
if word[i] not in vowels and word[i - 1] in vowels:
r1 = word[i + 1 :]
break
for i in range(1, len(r1)):
if r1[i] not in vowels and r1[i - 1] in vowels:
r2 = r1[i + 1 :]
break
return (r1, r2)
def __suffix_replace(self, original, old, new):
return original[: -len(old)] + new
# Stem words
# Main method of the Porter 2/Snowball stemming algorithm
def __stem(self, word):
if word in self.__special_words:
return self.__special_words[word]
elif len(word) <= 3:
return word
if word.startswith("y"):
word = "".join(("Y", word[1:]))
for i in range(1, len(word)):
if word[i - 1] in self.__vowels and word[i] == "y":
word = "".join((word[:i], "Y", word[i + 1 :]))
step1a_vowel_found = False
step1b_vowel_found = False
r1 = ""
r2 = ""
if word.startswith(("gener", "commun", "arsen")):
if word.startswith(("gener", "arsen")):
r1 = word[5:]
else:
r1 = word[6:]
for i in range(1, len(r1)):
if r1[i] not in self.__vowels and r1[i - 1] in self.__vowels:
r2 = r1[i + 1 :]
break
else:
r1, r2 = self.__r1r2(word, self.__vowels)
# STEP 1a
for suffix in self.__step1a_suffixes:
if word.endswith(suffix):
if suffix == "sses":
word = word[:-2]
r1 = r1[:-2]
r2 = r2[:-2]
elif suffix in ("ied", "ies"):
if len(word[: -len(suffix)]) > 1:
word = word[:-2]
r1 = r1[:-2]
r2 = r2[:-2]
else:
word = word[:-1]
r1 = r1[:-1]
r2 = r2[:-1]
elif suffix == "s":
for letter in word[:-2]:
if letter in self.__vowels:
step1a_vowel_found = True
break
if step1a_vowel_found:
word = word[:-1]
r1 = r1[:-1]
r2 = r2[:-1]
break
# STEP 1b
for suffix in self.__step1b_suffixes:
if word.endswith(suffix):
if suffix in ("eed", "eedly"):
if r1.endswith(suffix):
word = self.__suffix_replace(word, suffix, "ee")
if len(r1) >= len(suffix):
r1 = self.__suffix_replace(r1, suffix, "ee")
else:
r1 = ""
if len(r2) >= len(suffix):
r2 = self.__suffix_replace(r2, suffix, "ee")
else:
r2 = ""
else:
for letter in word[: -len(suffix)]:
if letter in self.__vowels:
step1b_vowel_found = True
break
if step1b_vowel_found:
word = word[: -len(suffix)]
r1 = r1[: -len(suffix)]
r2 = r2[: -len(suffix)]
if word.endswith(("at", "bl", "iz")):
word = "".join((word, "e"))
r1 = "".join((r1, "e"))
if len(word) > 5 or len(r1) >= 3:
r2 = "".join((r2, "e"))
elif word.endswith(self.__double_consonants):
word = word[:-1]
r1 = r1[:-1]
r2 = r2[:-1]
elif (
r1 == ""
and len(word) >= 3
and word[-1] not in self.__vowels
and word[-1] not in "wxY"
and word[-2] in self.__vowels
and word[-3] not in self.__vowels
) or (
r1 == ""
and len(word) == 2
and word[0] in self.__vowels
and word[1] not in self.__vowels
):
word = "".join((word, "e"))
if len(r1) > 0:
r1 = "".join((r1, "e"))
if len(r2) > 0:
r2 = "".join((r2, "e"))
break
# STEP 1c
if len(word) > 2 and word[-1] in "yY" and word[-2] not in self.__vowels:
word = "".join((word[:-1], "i"))
if len(r1) >= 1:
r1 = "".join((r1[:-1], "i"))
else:
r1 = ""
if len(r2) >= 1:
r2 = "".join((r2[:-1], "i"))
else:
r2 = ""
# STEP 2
for suffix in self.__step2_suffixes:
if word.endswith(suffix):
if r1.endswith(suffix):
if (
suffix in ("entli", "fulli", "lessli", "tional")
or (suffix == "li" and word[-3] in self.__li_ending)
):
word = word[:-2]
r1 = r1[:-2]
r2 = r2[:-2]
elif suffix in ("enci", "anci", "abli"):
word = "".join((word[:-1], "e"))
if len(r1) >= 1:
r1 = "".join((r1[:-1], "e"))
else:
r1 = ""
if len(r2) >= 1:
r2 = "".join((r2[:-1], "e"))
else:
r2 = ""
elif suffix in ("izer", "ization"):
word = self.__suffix_replace(word, suffix, "ize")
if len(r1) >= len(suffix):
r1 = self.__suffix_replace(r1, suffix, "ize")
else:
r1 = ""
if len(r2) >= len(suffix):
r2 = self.__suffix_replace(r2, suffix, "ize")
else:
r2 = ""
elif suffix in ("ational", "ation", "ator"):
word = self.__suffix_replace(word, suffix, "ate")
if len(r1) >= len(suffix):
r1 = self.__suffix_replace(r1, suffix, "ate")
else:
r1 = ""
if len(r2) >= len(suffix):
r2 = self.__suffix_replace(r2, suffix, "ate")
else:
r2 = "e"
elif suffix in ("alism", "aliti", "alli"):
word = self.__suffix_replace(word, suffix, "al")
if len(r1) >= len(suffix):
r1 = self.__suffix_replace(r1, suffix, "al")
else:
r1 = ""
if len(r2) >= len(suffix):
r2 = self.__suffix_replace(r2, suffix, "al")
else:
r2 = ""
elif suffix == "fulness":
word = word[:-4]
r1 = r1[:-4]
r2 = r2[:-4]
elif suffix in ("ousli", "ousness"):
word = self.__suffix_replace(word, suffix, "ous")
if len(r1) >= len(suffix):
r1 = self.__suffix_replace(r1, suffix, "ous")
else:
r1 = ""
if len(r2) >= len(suffix):
r2 = self.__suffix_replace(r2, suffix, "ous")
else:
r2 = ""
elif suffix in ("iveness", "iviti"):
word = self.__suffix_replace(word, suffix, "ive")
if len(r1) >= len(suffix):
r1 = self.__suffix_replace(r1, suffix, "ive")
else:
r1 = ""
if len(r2) >= len(suffix):
r2 = self.__suffix_replace(r2, suffix, "ive")
else:
r2 = "e"
elif suffix in ("biliti", "bli"):
word = self.__suffix_replace(word, suffix, "ble")
if len(r1) >= len(suffix):
r1 = self.__suffix_replace(r1, suffix, "ble")
else:
r1 = ""
if len(r2) >= len(suffix):
r2 = self.__suffix_replace(r2, suffix, "ble")
else:
r2 = ""
elif suffix == "ogi" and word[-4] == "l":
word = word[:-1]
r1 = r1[:-1]
r2 = r2[:-1]
break
# STEP 3
for suffix in self.__step3_suffixes:
if word.endswith(suffix):
if r1.endswith(suffix):
if suffix == "tional":
word = word[:-2]
r1 = r1[:-2]
r2 = r2[:-2]
elif suffix == "ational":
word = self.__suffix_replace(word, suffix, "ate")
if len(r1) >= len(suffix):
r1 = self.__suffix_replace(r1, suffix, "ate")
else:
r1 = ""
if len(r2) >= len(suffix):
r2 = self.__suffix_replace(r2, suffix, "ate")
else:
r2 = ""
elif suffix == "alize":
word = word[:-3]
r1 = r1[:-3]
r2 = r2[:-3]
elif suffix in ("icate", "iciti", "ical"):
word = self.__suffix_replace(word, suffix, "ic")
if len(r1) >= len(suffix):
r1 = self.__suffix_replace(r1, suffix, "ic")
else:
r1 = ""
if len(r2) >= len(suffix):
r2 = self.__suffix_replace(r2, suffix, "ic")
else:
r2 = ""
elif suffix in ("ful", "ness"):
word = word[: -len(suffix)]
r1 = r1[: -len(suffix)]
r2 = r2[: -len(suffix)]
elif suffix == "ative" and r2.endswith(suffix):
word = word[:-5]
r1 = r1[:-5]
r2 = r2[:-5]
break
# STEP 4
for suffix in self.__step4_suffixes:
if word.endswith(suffix):
if r2.endswith(suffix):
if suffix == "ion":
if word[-4] in "st":
word = word[:-3]
r1 = r1[:-3]
r2 = r2[:-3]
else:
word = word[: -len(suffix)]
r1 = r1[: -len(suffix)]
r2 = r2[: -len(suffix)]
break
# STEP 5
if (
(r2.endswith("l") and word[-2] == "l")
or r2.endswith("e")
or (
r1.endswith("e")
and len(word) >= 4
and (
word[-2] in self.__vowels
or word[-2] in "wxY"
or word[-3] not in self.__vowels
or word[-4] in self.__vowels
)
)
):
word = word[:-1]
word = word.replace("Y", "y")
# STEP 6
for suffix in self.__step6_suffixes:
if word.endswith(suffix):
if (
(suffix == "graphi" and len(word) >= 9)
or (suffix == "logi" and len(word) >= 7)
or (suffix == "nomi" and len(word) >= 7)
or (suffix == "pathi" and len(word) >= 6)
or (suffix == "scopi" and len(word) >= 8)
or (suffix == "therapi")
or (suffix == "tomi" and len(word) >= 7)
or (suffix == "tri" and len(word) >= 8 and word[-4] in "ae")
):
word = word[:-1]
elif suffix == "pathet" and len(word) >= 7:
word = word[:-2]
elif (
(suffix == "curist" and len(word) >= 8)
or (suffix == "logist" and len(word) >= 9)
or (suffix == "nomist" and len(word) >= 9)
or (suffix == "therapeut")
or (suffix == "therapist")
or (suffix == "tomist" and len(word) >= 9)
or (suffix == "trist" and len(word) >= 10 and word[-6] in "ae")
or (suffix == "turist" and len(word) >= 8)
):
word = word[:-3]
elif (
(suffix == "physicist")
or (suffix == "trician" and len(word) >= 10 and word[-8] in "ae")
):
word = word[:-5]
break
return word
# Mapping rules
__map_dict = {
# ADD YOUR OWN MAPPING RULES HERE FOR CORRECTING UNDERSTEMMING ERRORS
# Below are a few examples
"auto": "car",
"automobil": "car",
"automot": "car",
}
# Map stemmed words to other stems
def __map(self, word):
return self.__map_dict.get(word, word)
# Clean text
def clean_text(self, text):
text = text.lower()
# ADD YOUR OWN CLEANING RULES HERE
# Below are a few examples
text = re.sub(r"\bcarrepair\b", r" car repair ", text)
text = re.sub(r"\block[ -]+smith", r" locksmith", text)
text = re.sub(r"\(except.*\)", r" ", text)
# Remove non-letters and extraneous whitespace
text = re.sub(r"[^a-z]+", r" " , text)
text = text.strip()
# Remove stop words
text = " ".join(word for word in text.split(" ") if word not in self.__stop_words)
# Stem words
text = " ".join(self.__stem(word) for word in text.split(" ") if word != "")
# Map stemmed words to other stems
text = " ".join(self.__map(word) for word in text.split(" ") if word != "")
return text
# Given tokenized clean text and a value of n, return a list of n-combs
# BeaconModel considers only 1-, 2-, and 3-combs
def __get_ncombs(self, tokens, n):
tokens = sorted(set(tokens))
n_tokens = len(tokens)
ncombs = []
if n == 1 and n_tokens >= 1 and tokens[0] != "":
for i in range(n_tokens):
ncomb = tokens[i]
ncombs.append(ncomb)
elif n == 2 and n_tokens >= 2:
for i in range(n_tokens):
for j in range(i + 1, n_tokens):
ncomb = "_".join(sorted([tokens[i], tokens[j]]))
ncombs.append(ncomb)
elif n == 3 and n_tokens >= 3:
for i in range(n_tokens):
for j in range(i + 1, n_tokens):
for k in range(j + 1, n_tokens):
ncomb = "_".join(sorted([tokens[i], tokens[j], tokens[k]]))
ncombs.append(ncomb)
return ncombs
# Given a 6-digit NAICS code, return the sector
def __get_sector(self, naics):
sector = naics[:2]
# Manufacturing
if sector in ("32", "33"):
return "31"
# Retail Trade
elif sector == "45":
return "44"
# Transportation and Warehousing
elif sector == "49":
return "48"
return sector
# Fit BeaconModel for a given sector
# The method fit() calls __fit_sector() multiple times
def __fit_sector(self, X_clean, y, sample_weight, sector):
# Preparation
sample_size = len(X_clean)
naics_unique = sorted(set(y))
naics_index = {}
for i in range(len(naics_unique)):
naics_index[naics_unique[i]] = i
n_naics = len(naics_unique)
# Dictionary of n-combs
ncombs_list = []
ncombs_freqs_raw = {}
ncombs_freqs_sw = {}
for i in range(len(X_clean)):
x_clean_temp = X_clean[i]
naics_temp = y[i]
sample_weight_temp = sample_weight[i]
if x_clean_temp != "":
tokens = x_clean_temp.split(" ")
ncs = self.__get_ncombs(tokens, 1) + self.__get_ncombs(tokens, 2) + self.__get_ncombs(tokens, 3)
ncs_unique = list(set(ncs))
for nc in ncs_unique:
if nc not in ncombs_freqs_raw:
ncombs_freqs_raw[nc] = [0] * n_naics
ncombs_freqs_sw[nc] = [0] * n_naics
ncombs_freqs_raw[nc][naics_index[naics_temp]] += 1
ncombs_freqs_sw[nc][naics_index[naics_temp]] += sample_weight_temp
ncombs_list.extend(ncs_unique)
ncombs_unique = sorted(set(ncombs_list))
# Apply freq_thresh to determine n-comb features
top_ncombs = [nc for nc in ncombs_unique if sum(ncombs_freqs_raw[nc]) >= self.freq_thresh]
# N-comb proportions
top_ncombs_props = {}
for nc in top_ncombs:
row_total_freq_sw = 1.0 * sum(ncombs_freqs_sw[nc])
top_ncombs_props[nc] = [freq_sw / row_total_freq_sw for freq_sw in ncombs_freqs_sw[nc]]
# N-comb purity weights
top_ncombs_weights = {}
for nc in top_ncombs:
weight_temp = max(top_ncombs_props[nc]) - 1/n_naics
weight_temp_norm = (n_naics/(n_naics - 1)) * weight_temp
weight_temp_round = round(weight_temp_norm, 4)
if weight_temp_round < 0.0001:
weight_temp_round = 0.0001
top_ncombs_weights[nc] = weight_temp_round
# Dictionary of exact features
ems_list = []
ems_freqs_raw = {}
ems_freqs_sw = {}
for i in range(len(X_clean)):
x_clean_temp = X_clean[i]
naics_temp = y[i]
sample_weight_temp = sample_weight[i]
if x_clean_temp != "":
tokens = x_clean_temp.split(" ")
em = "_".join(sorted(set(token for token in tokens if token in top_ncombs_weights)))
if em != "":
if em not in ems_freqs_raw:
ems_freqs_raw[em] = [0] * n_naics
ems_freqs_sw[em] = [0] * n_naics
ems_freqs_raw[em][naics_index[naics_temp]] += 1
ems_freqs_sw[em][naics_index[naics_temp]] += sample_weight_temp
ems_list.append(em)
ems_unique = sorted(set(ems_list))
# Apply freq_thresh to determine exact features
top_ems = [em for em in ems_unique if sum(ems_freqs_raw[em]) >= self.freq_thresh]
# Exact feature proportions
top_ems_props = {}
for em in top_ems:
row_total_freq_sw = 1.0 * sum(ems_freqs_sw[em])
top_ems_props[em] = [freq_sw / row_total_freq_sw for freq_sw in ems_freqs_sw[em]]
# Exact feature purity weights
top_ems_weights = {}
for em in top_ems:
weight_temp = max(top_ems_props[em]) - 1/n_naics
weight_temp_norm = (n_naics/(n_naics - 1)) * weight_temp
weight_temp_round = round(weight_temp_norm, 4)
if weight_temp_round < 0.0001:
weight_temp_round = 0.0001
top_ems_weights[em] = weight_temp_round
# Populate BeaconModel dictionaries with sector-specific information
self.sample_sizes_[sector] = sample_size
self.naics_indices_[sector] = naics_index
self.dict_ncombs_props_[sector] = top_ncombs_props
self.dict_ncombs_weights_[sector] = top_ncombs_weights
self.dict_ems_props_[sector] = top_ems_props
self.dict_ems_weights_[sector] = top_ems_weights
return
# Fit BeaconModel
def fit(self, X, y, sample_weight=None):
if self.verbose:
print("")
print("Parameter and data validation")
# Validate parameters and data
# __validate_parameters() and __validate_data() raise ValueErrors if the input does not have the expected format
self.__validate_parameters()
X, y, sample_weight = self.__validate_data(X, y, sample_weight)
# Determine sectors and NAICS codes
naics_unique = sorted(set(y))
n_naics = len(naics_unique)
y_sector = [self.__get_sector(naics) for naics in y]
sectors_unique = sorted(set(y_sector))
n_sectors = len(sectors_unique)
if n_sectors == 1:
raise ValueError("BeaconModel requires data from at least two sectors.")
in_sector = {sector: [] for sector in sectors_unique}
for i in range(len(y_sector)):
in_sector[y_sector[i]].append(i)
for sector in sectors_unique:
if len(set(y[in_sector[sector]])) == 1:
raise ValueError("BeaconModel requires data from at least two NAICS codes within each sector.")
# Clean text
t0 = time.time()
X_clean = np.array([self.clean_text(x) for x in X])
t1 = time.time()
if self.verbose:
print("Text cleaning (time = {}s)".format(round(t1 - t0, 3)))
# Initialize BeaconModel attributes
# As per Python nomenclature conventions, class attributes estimated from the sample data end in "_"
# The trailing "_" is used to check whether the model has been fitted (see comments on the check_is_fitted() method)
self.naics_ = naics_unique
self.n_naics_ = n_naics
self.sectors_ = sectors_unique
self.n_sectors_ = n_sectors
self.sample_sizes_ = {}
self.naics_indices_ = {}
self.dict_ncombs_props_ = {}