classifier 1.3.5 → 1.4.1
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/lib/classifier/bayes.rb +145 -119
- data/lib/classifier/extensions/string.rb +1 -1
- data/lib/classifier/extensions/vector.rb +66 -72
- data/lib/classifier/extensions/vector_serialize.rb +6 -8
- data/lib/classifier/extensions/word_hash.rb +108 -114
- data/lib/classifier/lsi/content_node.rb +25 -23
- data/lib/classifier/lsi/summary.rb +20 -20
- data/lib/classifier/lsi/word_list.rb +1 -2
- data/lib/classifier/lsi.rb +112 -89
- data/lib/classifier.rb +1 -0
- data/test/test_helper.rb +5 -0
- metadata +7 -21
data/lib/classifier/lsi.rb
CHANGED
@@ -3,15 +3,16 @@
|
|
3
3
|
# License:: LGPL
|
4
4
|
|
5
5
|
begin
|
6
|
-
|
7
|
-
|
8
|
-
require 'gsl' # requires http://rb-gsl.rubyforge.org/
|
9
|
-
require 'classifier/extensions/vector_serialize'
|
10
|
-
$GSL = true
|
6
|
+
# to test the native vector class, try `rake test NATIVE_VECTOR=true`
|
7
|
+
raise LoadError if ENV['NATIVE_VECTOR'] == 'true'
|
11
8
|
|
9
|
+
require 'gsl' # requires https://github.com/SciRuby/rb-gsl/
|
10
|
+
require 'classifier/extensions/vector_serialize'
|
11
|
+
$GSL = true
|
12
12
|
rescue LoadError
|
13
|
-
|
14
|
-
|
13
|
+
warn 'Notice: for 10x faster LSI support, please install https://github.com/SciRuby/rb-gsl/'
|
14
|
+
$GSL = false
|
15
|
+
require 'classifier/extensions/vector'
|
15
16
|
end
|
16
17
|
|
17
18
|
require 'classifier/lsi/word_list'
|
@@ -19,12 +20,10 @@ require 'classifier/lsi/content_node'
|
|
19
20
|
require 'classifier/lsi/summary'
|
20
21
|
|
21
22
|
module Classifier
|
22
|
-
|
23
23
|
# This class implements a Latent Semantic Indexer, which can search, classify and cluster
|
24
24
|
# data based on underlying semantic relations. For more information on the algorithms used,
|
25
25
|
# please consult Wikipedia[http://en.wikipedia.org/wiki/Latent_Semantic_Indexing].
|
26
26
|
class LSI
|
27
|
-
|
28
27
|
attr_reader :word_list
|
29
28
|
attr_accessor :auto_rebuild
|
30
29
|
|
@@ -34,8 +33,10 @@ module Classifier
|
|
34
33
|
#
|
35
34
|
def initialize(options = {})
|
36
35
|
@auto_rebuild = true unless options[:auto_rebuild] == false
|
37
|
-
@word_list
|
38
|
-
@
|
36
|
+
@word_list = WordList.new
|
37
|
+
@items = {}
|
38
|
+
@version = 0
|
39
|
+
@built_at_version = -1
|
39
40
|
end
|
40
41
|
|
41
42
|
# Returns true if the index needs to be rebuilt. The index needs
|
@@ -58,7 +59,7 @@ module Classifier
|
|
58
59
|
# ar = ActiveRecordObject.find( :all )
|
59
60
|
# lsi.add_item ar, *ar.categories { |x| ar.content }
|
60
61
|
#
|
61
|
-
def add_item(
|
62
|
+
def add_item(item, *categories, &block)
|
62
63
|
clean_word_hash = block ? block.call(item).clean_word_hash : item.to_s.clean_word_hash
|
63
64
|
@items[item] = ContentNode.new(clean_word_hash, *categories)
|
64
65
|
@version += 1
|
@@ -69,24 +70,25 @@ module Classifier
|
|
69
70
|
# you are passing in a string with no categorries. item
|
70
71
|
# will be duck typed via to_s .
|
71
72
|
#
|
72
|
-
def <<(
|
73
|
-
add_item
|
73
|
+
def <<(item)
|
74
|
+
add_item(item)
|
74
75
|
end
|
75
76
|
|
76
77
|
# Returns the categories for a given indexed items. You are free to add and remove
|
77
78
|
# items from this as you see fit. It does not invalide an index to change its categories.
|
78
79
|
def categories_for(item)
|
79
80
|
return [] unless @items[item]
|
80
|
-
|
81
|
+
|
82
|
+
@items[item].categories
|
81
83
|
end
|
82
84
|
|
83
85
|
# Removes an item from the database, if it is indexed.
|
84
86
|
#
|
85
|
-
def remove_item(
|
86
|
-
|
87
|
-
|
88
|
-
|
89
|
-
|
87
|
+
def remove_item(item)
|
88
|
+
return unless @items.key?(item)
|
89
|
+
|
90
|
+
@items.delete(item)
|
91
|
+
@version += 1
|
90
92
|
end
|
91
93
|
|
92
94
|
# Returns an array of items that are indexed.
|
@@ -94,13 +96,6 @@ module Classifier
|
|
94
96
|
@items.keys
|
95
97
|
end
|
96
98
|
|
97
|
-
# Returns the categories for a given indexed items. You are free to add and remove
|
98
|
-
# items from this as you see fit. It does not invalide an index to change its categories.
|
99
|
-
def categories_for(item)
|
100
|
-
return [] unless @items[item]
|
101
|
-
return @items[item].categories
|
102
|
-
end
|
103
|
-
|
104
99
|
# This function rebuilds the index if needs_rebuild? returns true.
|
105
100
|
# For very large document spaces, this indexing operation may take some
|
106
101
|
# time to complete, so it may be wise to place the operation in another
|
@@ -115,30 +110,31 @@ module Classifier
|
|
115
110
|
# cutoff parameter tells the indexer how many of these values to keep.
|
116
111
|
# A value of 1 for cutoff means that no semantic analysis will take place,
|
117
112
|
# turning the LSI class into a simple vector search engine.
|
118
|
-
def build_index(
|
113
|
+
def build_index(cutoff = 0.75)
|
119
114
|
return unless needs_rebuild?
|
115
|
+
|
120
116
|
make_word_list
|
121
117
|
|
122
118
|
doc_list = @items.values
|
123
|
-
tda = doc_list.collect { |node| node.raw_vector_with(
|
119
|
+
tda = doc_list.collect { |node| node.raw_vector_with(@word_list) }
|
124
120
|
|
125
121
|
if $GSL
|
126
|
-
|
127
|
-
|
128
|
-
|
129
|
-
|
130
|
-
|
131
|
-
|
132
|
-
|
133
|
-
|
122
|
+
tdm = GSL::Matrix.alloc(*tda).trans
|
123
|
+
ntdm = build_reduced_matrix(tdm, cutoff)
|
124
|
+
|
125
|
+
ntdm.size[1].times do |col|
|
126
|
+
vec = GSL::Vector.alloc(ntdm.column(col)).row
|
127
|
+
doc_list[col].lsi_vector = vec
|
128
|
+
doc_list[col].lsi_norm = vec.normalize
|
129
|
+
end
|
134
130
|
else
|
135
|
-
|
136
|
-
|
131
|
+
tdm = Matrix.rows(tda).trans
|
132
|
+
ntdm = build_reduced_matrix(tdm, cutoff)
|
137
133
|
|
138
|
-
|
139
|
-
|
140
|
-
|
141
|
-
|
134
|
+
ntdm.row_size.times do |col|
|
135
|
+
doc_list[col].lsi_vector = ntdm.column(col) if doc_list[col]
|
136
|
+
doc_list[col].lsi_norm = ntdm.column(col).normalize if doc_list[col]
|
137
|
+
end
|
142
138
|
end
|
143
139
|
|
144
140
|
@built_at_version = @version
|
@@ -152,13 +148,13 @@ module Classifier
|
|
152
148
|
# your dataset's general content. For example, if you were to use categorize on the
|
153
149
|
# results of this data, you could gather information on what your dataset is generally
|
154
150
|
# about.
|
155
|
-
def highest_relative_content(
|
156
|
-
|
151
|
+
def highest_relative_content(max_chunks = 10)
|
152
|
+
return [] if needs_rebuild?
|
157
153
|
|
158
|
-
|
159
|
-
|
154
|
+
avg_density = {}
|
155
|
+
@items.each_key { |x| avg_density[x] = proximity_array_for_content(x).inject(0.0) { |x, y| x + y[1] } }
|
160
156
|
|
161
|
-
|
157
|
+
avg_density.keys.sort_by { |x| avg_density[x] }.reverse[0..max_chunks - 1].map
|
162
158
|
end
|
163
159
|
|
164
160
|
# This function is the primitive that find_related and classify
|
@@ -173,17 +169,17 @@ module Classifier
|
|
173
169
|
# The parameter doc is the content to compare. If that content is not
|
174
170
|
# indexed, you can pass an optional block to define how to create the
|
175
171
|
# text data. See add_item for examples of how this works.
|
176
|
-
def proximity_array_for_content(
|
172
|
+
def proximity_array_for_content(doc, &block)
|
177
173
|
return [] if needs_rebuild?
|
178
174
|
|
179
|
-
content_node = node_for_content(
|
175
|
+
content_node = node_for_content(doc, &block)
|
180
176
|
result =
|
181
177
|
@items.keys.collect do |item|
|
182
|
-
if $GSL
|
183
|
-
|
184
|
-
|
185
|
-
|
186
|
-
|
178
|
+
val = if $GSL
|
179
|
+
content_node.search_vector * @items[item].search_vector.col
|
180
|
+
else
|
181
|
+
(Matrix[content_node.search_vector] * @items[item].search_vector)[0]
|
182
|
+
end
|
187
183
|
[item, val]
|
188
184
|
end
|
189
185
|
result.sort_by { |x| x[1] }.reverse
|
@@ -194,17 +190,17 @@ module Classifier
|
|
194
190
|
# calculated vectors instead of their full versions. This is useful when
|
195
191
|
# you're trying to perform operations on content that is much smaller than
|
196
192
|
# the text you're working with. search uses this primitive.
|
197
|
-
def proximity_norms_for_content(
|
193
|
+
def proximity_norms_for_content(doc, &block)
|
198
194
|
return [] if needs_rebuild?
|
199
195
|
|
200
|
-
content_node = node_for_content(
|
196
|
+
content_node = node_for_content(doc, &block)
|
201
197
|
result =
|
202
198
|
@items.keys.collect do |item|
|
203
|
-
if $GSL
|
204
|
-
|
205
|
-
|
206
|
-
|
207
|
-
|
199
|
+
val = if $GSL
|
200
|
+
content_node.search_norm * @items[item].search_norm.col
|
201
|
+
else
|
202
|
+
(Matrix[content_node.search_norm] * @items[item].search_norm)[0]
|
203
|
+
end
|
208
204
|
[item, val]
|
209
205
|
end
|
210
206
|
result.sort_by { |x| x[1] }.reverse
|
@@ -217,11 +213,12 @@ module Classifier
|
|
217
213
|
#
|
218
214
|
# While this may seem backwards compared to the other functions that LSI supports,
|
219
215
|
# it is actually the same algorithm, just applied on a smaller document.
|
220
|
-
def search(
|
216
|
+
def search(string, max_nearest = 3)
|
221
217
|
return [] if needs_rebuild?
|
222
|
-
|
218
|
+
|
219
|
+
carry = proximity_norms_for_content(string)
|
223
220
|
result = carry.collect { |x| x[0] }
|
224
|
-
|
221
|
+
result[0..max_nearest - 1]
|
225
222
|
end
|
226
223
|
|
227
224
|
# This function takes content and finds other documents
|
@@ -233,11 +230,11 @@ module Classifier
|
|
233
230
|
# This is particularly useful for identifing clusters in your document space.
|
234
231
|
# For example you may want to identify several "What's Related" items for weblog
|
235
232
|
# articles, or find paragraphs that relate to each other in an essay.
|
236
|
-
def find_related(
|
233
|
+
def find_related(doc, max_nearest = 3, &block)
|
237
234
|
carry =
|
238
|
-
proximity_array_for_content(
|
235
|
+
proximity_array_for_content(doc, &block).reject { |pair| pair[0] == doc }
|
239
236
|
result = carry.collect { |x| x[0] }
|
240
|
-
|
237
|
+
result[0..max_nearest - 1]
|
241
238
|
end
|
242
239
|
|
243
240
|
# This function uses a voting system to categorize documents, based on
|
@@ -249,10 +246,17 @@ module Classifier
|
|
249
246
|
# text. A cutoff of 1 means that every document in the index votes on
|
250
247
|
# what category the document is in. This may not always make sense.
|
251
248
|
#
|
252
|
-
def classify(
|
249
|
+
def classify(doc, cutoff = 0.30, &block)
|
250
|
+
votes = vote(doc, cutoff, &block)
|
251
|
+
|
252
|
+
ranking = votes.keys.sort_by { |x| votes[x] }
|
253
|
+
ranking[-1]
|
254
|
+
end
|
255
|
+
|
256
|
+
def vote(doc, cutoff = 0.30, &block)
|
253
257
|
icutoff = (@items.size * cutoff).round
|
254
|
-
carry = proximity_array_for_content(
|
255
|
-
carry = carry[0..icutoff-1]
|
258
|
+
carry = proximity_array_for_content(doc, &block)
|
259
|
+
carry = carry[0..icutoff - 1]
|
256
260
|
votes = {}
|
257
261
|
carry.each do |pair|
|
258
262
|
categories = @items[pair[0]].categories
|
@@ -261,23 +265,46 @@ module Classifier
|
|
261
265
|
votes[category] += pair[1]
|
262
266
|
end
|
263
267
|
end
|
268
|
+
votes
|
269
|
+
end
|
270
|
+
|
271
|
+
# Returns the same category as classify() but also returns
|
272
|
+
# a confidence value derived from the vote share that the
|
273
|
+
# winning category got.
|
274
|
+
#
|
275
|
+
# e.g.
|
276
|
+
# category,confidence = classify_with_confidence(doc)
|
277
|
+
# if confidence < 0.3
|
278
|
+
# category = nil
|
279
|
+
# end
|
280
|
+
#
|
281
|
+
#
|
282
|
+
# See classify() for argument docs
|
283
|
+
def classify_with_confidence(doc, cutoff = 0.30, &block)
|
284
|
+
votes = vote(doc, cutoff, &block)
|
285
|
+
votes_sum = votes.values.inject(0.0) { |sum, v| sum + v }
|
286
|
+
return [nil, nil] if votes_sum.zero?
|
264
287
|
|
265
288
|
ranking = votes.keys.sort_by { |x| votes[x] }
|
266
|
-
|
289
|
+
winner = ranking[-1]
|
290
|
+
vote_share = votes[winner] / votes_sum.to_f
|
291
|
+
[winner, vote_share]
|
267
292
|
end
|
268
293
|
|
269
294
|
# Prototype, only works on indexed documents.
|
270
295
|
# I have no clue if this is going to work, but in theory
|
271
296
|
# it's supposed to.
|
272
|
-
def highest_ranked_stems(
|
273
|
-
raise
|
297
|
+
def highest_ranked_stems(doc, count = 3)
|
298
|
+
raise 'Requested stem ranking on non-indexed content!' unless @items[doc]
|
299
|
+
|
274
300
|
arr = node_for_content(doc).lsi_vector.to_a
|
275
|
-
top_n = arr.sort.reverse[0..count-1]
|
276
|
-
|
301
|
+
top_n = arr.sort.reverse[0..count - 1]
|
302
|
+
top_n.collect { |x| @word_list.word_for_index(arr.index(x)) }
|
277
303
|
end
|
278
304
|
|
279
305
|
private
|
280
|
-
|
306
|
+
|
307
|
+
def build_reduced_matrix(matrix, cutoff = 0.75)
|
281
308
|
# TODO: Check that M>=N on these dimensions! Transpose helps assure this
|
282
309
|
u, v, s = matrix.SV_decomp
|
283
310
|
|
@@ -287,23 +314,21 @@ module Classifier
|
|
287
314
|
s[ord] = 0.0 if s[ord] < s_cutoff
|
288
315
|
end
|
289
316
|
# Reconstruct the term document matrix, only with reduced rank
|
290
|
-
u * ($GSL ? GSL::Matrix : ::Matrix).diag(
|
317
|
+
u * ($GSL ? GSL::Matrix : ::Matrix).diag(s) * v.trans
|
291
318
|
end
|
292
319
|
|
293
320
|
def node_for_content(item, &block)
|
294
|
-
if @items[item]
|
295
|
-
return @items[item]
|
296
|
-
else
|
297
|
-
clean_word_hash = block ? block.call(item).clean_word_hash : item.to_s.clean_word_hash
|
321
|
+
return @items[item] if @items[item]
|
298
322
|
|
299
|
-
|
323
|
+
clean_word_hash = block ? block.call(item).clean_word_hash : item.to_s.clean_word_hash
|
300
324
|
|
301
|
-
|
302
|
-
|
303
|
-
|
325
|
+
cn = ContentNode.new(clean_word_hash, &block) # make the node and extract the data
|
326
|
+
|
327
|
+
unless needs_rebuild?
|
328
|
+
cn.raw_vector_with(@word_list) # make the lsi raw and norm vectors
|
304
329
|
end
|
305
330
|
|
306
|
-
|
331
|
+
cn
|
307
332
|
end
|
308
333
|
|
309
334
|
def make_word_list
|
@@ -312,7 +337,5 @@ module Classifier
|
|
312
337
|
node.word_hash.each_key { |key| @word_list.add_word key }
|
313
338
|
end
|
314
339
|
end
|
315
|
-
|
316
340
|
end
|
317
341
|
end
|
318
|
-
|
data/lib/classifier.rb
CHANGED
data/test/test_helper.rb
ADDED
metadata
CHANGED
@@ -1,14 +1,14 @@
|
|
1
1
|
--- !ruby/object:Gem::Specification
|
2
2
|
name: classifier
|
3
3
|
version: !ruby/object:Gem::Version
|
4
|
-
version: 1.
|
4
|
+
version: 1.4.1
|
5
5
|
platform: ruby
|
6
6
|
authors:
|
7
7
|
- Lucas Carlson
|
8
|
-
autorequire:
|
8
|
+
autorequire:
|
9
9
|
bindir: bin
|
10
10
|
cert_chain: []
|
11
|
-
date:
|
11
|
+
date: 2024-07-31 00:00:00.000000000 Z
|
12
12
|
dependencies:
|
13
13
|
- !ruby/object:Gem::Dependency
|
14
14
|
name: fast-stemmer
|
@@ -24,20 +24,6 @@ dependencies:
|
|
24
24
|
- - "~>"
|
25
25
|
- !ruby/object:Gem::Version
|
26
26
|
version: 1.0.0
|
27
|
-
- !ruby/object:Gem::Dependency
|
28
|
-
name: mathn
|
29
|
-
requirement: !ruby/object:Gem::Requirement
|
30
|
-
requirements:
|
31
|
-
- - ">="
|
32
|
-
- !ruby/object:Gem::Version
|
33
|
-
version: '0'
|
34
|
-
type: :runtime
|
35
|
-
prerelease: false
|
36
|
-
version_requirements: !ruby/object:Gem::Requirement
|
37
|
-
requirements:
|
38
|
-
- - ">="
|
39
|
-
- !ruby/object:Gem::Version
|
40
|
-
version: '0'
|
41
27
|
- !ruby/object:Gem::Dependency
|
42
28
|
name: rake
|
43
29
|
requirement: !ruby/object:Gem::Requirement
|
@@ -99,11 +85,12 @@ files:
|
|
99
85
|
- lib/classifier/lsi/content_node.rb
|
100
86
|
- lib/classifier/lsi/summary.rb
|
101
87
|
- lib/classifier/lsi/word_list.rb
|
88
|
+
- test/test_helper.rb
|
102
89
|
homepage: https://github.com/cardmagic/classifier
|
103
90
|
licenses:
|
104
91
|
- LGPL
|
105
92
|
metadata: {}
|
106
|
-
post_install_message:
|
93
|
+
post_install_message:
|
107
94
|
rdoc_options: []
|
108
95
|
require_paths:
|
109
96
|
- lib
|
@@ -118,9 +105,8 @@ required_rubygems_version: !ruby/object:Gem::Requirement
|
|
118
105
|
- !ruby/object:Gem::Version
|
119
106
|
version: '0'
|
120
107
|
requirements: []
|
121
|
-
|
122
|
-
|
123
|
-
signing_key:
|
108
|
+
rubygems_version: 3.5.9
|
109
|
+
signing_key:
|
124
110
|
specification_version: 4
|
125
111
|
summary: A general classifier module to allow Bayesian and other types of classifications.
|
126
112
|
test_files: []
|