Class: FeatureSet::FeatureBuilders::WordVector

Inherits:
Base
  • Object
show all
Defined in:
lib/feature_set/feature_builders/word_vector.rb

Instance Attribute Summary collapse

Attributes inherited from Base

#options

Instance Method Summary collapse

Constructor Details

#initialize(options = {}) ⇒ WordVector

Options:

:tf_only => true|false, default is false
:idf_cutiff => <cutoff>, default is 10
:word_limit => <word limit>, default is 2000



12
13
14
15
 
# File 'lib/feature_set/feature_builders/word_vector.rb', line 12

def initialize(options = {})
  super
  @idfs = {}
end

Instance Attribute Details

#idfs ⇒ Object

Returns the value of attribute idfs.



6
7
8
 
# File 'lib/feature_set/feature_builders/word_vector.rb', line 6

def idfs
  @idfs
end

Instance Method Details

#before_build_features(dataset) ⇒ Object 



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
 
# File 'lib/feature_set/feature_builders/word_vector.rb', line 17

def before_build_features(dataset)
  @idfs = {}
  dataset.each do |row|
    row.each do |key, datum|
      next if key == :class
      if datum.value.is_a?(String)
        idfs[key] ||= {}
        datum.token_counts.keys.each do |token|
          idfs[key][token] ||= 0
          idfs[key][token] += 1
        end
      end
    end
  end
  
  num_docs = dataset.length
  idf_cutoff = (options[:idf_cutoff] || 10).to_f
  word_limit = options[:word_limit] || 2000
  STDERR.puts "Done building df counts.  The dataset has #{num_docs} documents."

  idfs.each do |feature, freqs|
    pruned = 0
    if options[:tf_only]
      new_freqs = freqs
    else
      new_freqs = {}
      freqs.each do |key, value|
        log = Math.log(num_docs / value.to_f)
        if log < idf_cutoff
          new_freqs[key] = log
        else
          pruned += 1
        end
      end
    end
    if options[:word_limit]
      new_freqs = if options[:tf_only]
                    new_freqs.to_a.sort {|a, b| b.last <=> a.last }
                  else
                    new_freqs.to_a.sort {|a, b| a.last <=> b.last }
                  end
      new_freqs = new_freqs[0...word_limit].inject({}) { |m, (k, v)| m[k] = v; m }
    end
    idfs[feature] = new_freqs
    STDERR.puts "Done calculating idfs for #{feature}.  Pruned #{pruned} rare values, leaving #{idfs[feature].length} values."
  end
end

#build_features(datum, key, row) ⇒ Object 



65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
 
# File 'lib/feature_set/feature_builders/word_vector.rb', line 65

def build_features(datum, key, row)
  return {} unless datum.value.is_a?(String)
  num_words = datum.tokens.length.to_f
  unless idfs[key]
    STDERR.puts "WARNING: build_features called on untrained data in WordVector.  Are you calling 'data_set.build_features_for' without calling 'data_set.build_features_from_data!' first?"
  end
  if options[:tf_only]
    (idfs[key] || {}).inject({}) do |memo, (word, idf)|
      memo["wv_#{word}"] = ((datum.token_counts[word] || 0) / num_words)
      memo
    end
  else
    (idfs[key] || {}).inject({}) do |memo, (word, idf)|
      memo["wv_#{word}"] = ((datum.token_counts[word] || 0) / num_words) * idf
      memo
    end
  end
end