Module: Discretizer

Includes:
Consistency, Entropy
Included in:
FSelector::BaseDiscrete, FSelector::CFS_d, FSelector::KS_CCBF, FSelector::ReliefF_d, FSelector::Relief_d
Defined in:
lib/fselector/discretizer.rb

Overview

discretize continuous feature

Instance Method Summary collapse

Methods included from Entropy

#get_conditional_entropy, #get_information_gain, #get_joint_entropy, #get_marginal_entropy, #get_symmetrical_uncertainty

Methods included from Consistency

#get_IR, #get_IR_by_count, #get_IR_by_feature, #get_instance_count

Instance Method Details

#discretize_by_Chi2!(delta = 0.02)

Note:

Chi2 does some feature reduction if a discretized feature has only one interval. Using delta==0.02 reproduces exactly the same results as that of the original Chi2 algorithm

discretize by Chi2 algorithm

ref: Chi2: Feature Selection and Discretization of Numeric Attributes



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
# File 'lib/fselector/discretizer.rb', line 159

def discretize_by_Chi2!(delta=0.02)    
  # degree of freedom equals one less than number of classes     
  df = get_classes.size-1
      
  #
  # Phase 1
  #
  
  sig_level = 0.5
  sig_level0 = sig_level
  
  inst_cnt = get_instance_count
  inconsis_rate = get_IR_by_count(inst_cnt)
 
  # f2bs = {
    # :'sepal-length' => [4.4],
    # :'sepal-width' => [2.0],
    # :'petal-length' => [1.0, 3.0, 5.0],
    # :'petal-width' => [0.1, 1.0, 1.7],
  # }
  
  while true
    chisq = pval2chisq(sig_level, df)
    f2bs = {} # cut ponts
    
    each_feature do |f|
      bs, cs, qs = chi2_init(f)
      chi2_merge(bs, cs, qs, chisq)
      
      f2bs[f] = bs
    end
    
    inconsis_rate = chi2_get_inconsistency_rate(inst_cnt, f2bs)
    
    if inconsis_rate <= delta
      sig_level -= 0.1
      sig_level0 = sig_level
      
      break if sig_level0 <= 0.2 # phase 1 stop at level == 0.2
    else # data inconsistency
      break
    end 
  end
  
  #
  # Phase 2
  #
  
  try_levels = [0.1, 0.01, 0.001, 1e-4, 
                1e-5, 1e-6, 1e-7, 1e-8, 
                1e-9, 1e-10, 1e-11, 1e-12]           
  mergeble_fs = []
  f2sig_level = {}
  
  each_feature do |f|
    mergeble_fs << f
    f2sig_level[f] = sig_level0
  end
  
  f2bs = {} # cut ponts
  
  while not mergeble_fs.empty?
    mergeble_fs.each do |f|
      #pp f
      bs, cs, qs = chi2_init(f)
      chisq_now = pval2chisq(f2sig_level[f], df)
      chi2_merge(bs, cs, qs, chisq_now)
      
      # backup
      bs_bak = nil
      if f2bs.has_key? f
        bs_bak = f2bs[f]
      end
      f2bs[f] = bs
      
      inconsis_rate = chi2_get_inconsistency_rate(inst_cnt, f2bs)
      
      if (inconsis_rate <= delta)
        # try next level
        next_level = chi2_decrease_sig_level(f2sig_level[f], try_levels)
        f2sig_level[f] = next_level
        
        if not next_level # we've tried all levels
          mergeble_fs.delete(f)
        else
          f2bs[f] = bs # record cut points for this level
        end
      else # cause more inconsistency
        f2bs[f] = bs_bak if bs_bak # restore last cut points
        mergeble_fs.delete(f) # not mergeble
      end
    end
  end
  #pp f2bs
  #pp f2sig_level
  
  # if there is only one interval, remove this feature
  each_sample do |k, s|
    s.delete_if { |f, v| f2bs[f].size <= 1 }
  end
  
  # discretize according to each feature's cut points
  discretize_at_cutpoints!(f2bs)
end

#discretize_by_ChiMerge!(alpha = 0.10)

Note:

data structure will be altered

discretize by ChiMerge algorithm

ref: ChiMerge: Discretization of Numberic Attributes



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
# File 'lib/fselector/discretizer.rb', line 72

def discretize_by_ChiMerge!(alpha=0.10)
  # degree of freedom equals one less than number of classes
  df = get_classes.size-1
  chisq = pval2chisq(alpha, df)
  
  # for intialization
  hzero = {}
  each_class do |k|
    hzero[k] = 0.0
  end
  
  # determine the final boundaries for each feature
  f2bs = {}
  each_feature do |f|
    #f = :"sepal-length"
    # 1a. initialize boundaries
    bs, cs, qs = [], [], []
    fvs = get_feature_values(f).uniq.sort
    fvs.each do |v|
      bs << v
      cs << hzero.dup
    end
    
    # 1b. initialize counts for each interval
    each_sample do |k, s|
      next if not s.has_key? f
      i = bs.rindex { |x| s[f] >= x }
      cs[i][k] += 1.0
    end
    
    # 1c. initialize chi-squared values between two adjacent intervals
    cs.each_with_index do |c, i|
      if i+1 < cs.size
        qs << chisq_calc(c, cs[i+1])
      end
    end
    
    # 2. iteratively merge intervals
    until qs.empty? or qs.min > chisq
      qs.each_with_index do |q, i|
        next if q != qs.min
        
        # update cs for merged two intervals
        cm = {}
        each_class do |k|
          cm[k] = cs[i][k]+cs[i+1][k]
        end
        
        # update qs if necessary
        # before merged intervals
        if i-1 >= 0
          qs[i-1] = chisq_calc(cs[i-1], cm)
        end
        # after merged intervals
        if i+1 < qs.size
          qs[i+1] = chisq_calc(cm, cs[i+2])
        end
        
        # merge up
        bs.delete_at(i+1)
        cs.delete_at(i);cs.delete_at(i);cs.insert(i, cm)
        qs.delete_at(i)
        
        # break out
        break
      end
    end
    
    # 3. record the final boundaries
    f2bs[f] = bs
  end
  
  # discretize according to each feature's boundaries
  discretize_at_cutpoints!(f2bs)
end

#discretize_by_equal_frequency!(n_interval)

Note:

data structure will be altered

discretize by equal-frequency intervals



43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
# File 'lib/fselector/discretizer.rb', line 43

def discretize_by_equal_frequency!(n_interval)
  n_interval = 1 if n_interval < 1 # at least one interval
  
  # first determine the boundaries
  f2bs = Hash.new { |h,k| h[k] = [] }
  each_feature do |f|
    fvs = get_feature_values(f).sort
    # number of samples in each interval
    ns = (fvs.size.to_f/n_interval).round
    fvs.each_with_index do |v, i|
      if i%ns == 0
        f2bs[f] << v
      end
    end
  end
  
  # then discretize based on cut points
  discretize_at_cutpoints!(f2bs)
end

#discretize_by_equal_width!(n_interval)

Note:

data structure will be altered

discretize by equal-width intervals



16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
# File 'lib/fselector/discretizer.rb', line 16

def discretize_by_equal_width!(n_interval)
  n_interval = 1 if n_interval < 1 # at least one interval
  
  # first determine the boundary of each feature
  f2bs = Hash.new { |h,k| h[k] = [] }
  each_feature do |f|
    fvs = get_feature_values(f)
    fmin, fmax = fvs.min, fvs.max
    delta = (fmax-fmin)/n_interval
    
    n_interval.times do |i|
      f2bs[f] << fmin + i*delta
     end
  end
  
  # then discretize based on cut points
  discretize_at_cutpoints!(f2bs)
end

#discretize_by_MID!

Note:

no missing feature value is allowed, and data structure will be altered

discretize by Multi-Interval Discretization (MID) algorithm

ref: Multi-Interval Discretization of Continuous-Valued Attributes for Classification Learning



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
# File 'lib/fselector/discretizer.rb', line 272

def discretize_by_MID!    
  # determine the final boundaries
  f2cp = {} # cut points for each feature
  each_feature do |f|
    cv = get_class_labels
    fv = get_feature_values(f)
    
    n = cv.size
    abort "[#{__FILE__}@#{__LINE__}]: \n"+
          "  missing feature value is not allowed!" if n != fv.size
    
    # sort cv and fv according to ascending order of fv
    sis = (0...n).to_a.sort { |i,j| fv[i] <=> fv[j] }
    cv = cv.values_at(*sis)
    fv = fv.values_at(*sis)
    
    # get initial boundaries
    bs = []
    fv.each_with_index do |v, i|
      # cut point (Ta) for feature A must always be a value between
      # two examples of different classes in the sequence of sorted examples
      # see orginal reference
      if i < n-1 and cv[i] != cv[i+1]
        bs << v
      end
    end
    bs.uniq! # remove duplicates
    
    # main algorithm, iteratively determine cut point
    cp = []
    partition(cv, fv, bs, cp)
    
    # record cut points for feature (f)
    f2cp[f] = cp.sort # sorted cut points
  end
  
  # discretize based on cut points
  discretize_at_cutpoints!(f2cp)
end

#discretize_by_TID!

Note:

no missing feature value is allowed, and data structure will be altered

discretize by Three-Interval Discretization (TID) algorithm

ref: Filter versus wrapper gene selection approaches in DNA microarray domains



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
# File 'lib/fselector/discretizer.rb', line 320

def discretize_by_TID!
  # cut points for each feature
  f2cp = {}
  
  each_feature do |f|
    cv = get_class_labels
    fv = get_feature_values(f)
    
    n = cv.size
    abort "[#{__FILE__}@#{__LINE__}]: \n"+
          "  missing feature value is not allowed!" if n != fv.size
    
    # sort cv and fv according to ascending order of fv
    sis = (0...n).to_a.sort { |i,j| fv[i] <=> fv[j] }
    cv = cv.values_at(*sis)
    fv = fv.values_at(*sis)
    
    # get initial boundaries
    bs = []
    fv_u = fv.uniq
    fv_u.each_with_index do |v, i|
      # cut points are the mean of two adjacent data points
      if i < fv_u.size-1
        bs << (v+fv_u[i+1])/2.0
      end
    end
    
    # test each pair cut point
    s_best, h1_best, h2_best = nil, nil, nil
    
    bs.each_with_index do |h1, i|        
      bs.each_with_index do |h2, j|
        next if j <= i
        
        n_h1 = (0...n).to_a.select { |x| fv[x] < h1 }.size.to_f
        n_h1_h2 = (0...n).to_a.select { |x| fv[x] > h1 and fv[x] < h2 }.size.to_f
        n_h2 = (0...n).to_a.select { |x| fv[x] > h2 }.size.to_f
        
        s = 0.0
        
        each_class do |k|
          n_h1_k = (0...n).to_a.select { |x| fv[x] < h1 and cv[x] == k }.size.to_f
          n_h1_h2_k = (0...n).to_a.select { |x| fv[x] > h1 and fv[x] < h2 and cv[x] == k }.size.to_f
          n_h2_k = (0...n).to_a.select { |x| fv[x] > h2 and cv[x] == k }.size.to_f
          
          s += n_h1_k * Math.log2(n_h1_k/n_h1) if not n_h1_k.zero?
          s += n_h1_h2_k * Math.log2(n_h1_h2_k/n_h1_h2) if not n_h1_h2_k.zero?
          s += n_h2_k * Math.log2(n_h2_k/n_h2) if not n_h2_k.zero?
          
          #pp [s_best, s, h1, h2] + [n_h1, n_h1_k] + [n_h1_h2, n_h1_h2_k] + [n_h2, n_h2_k]
        end
        
        if not s_best or s > s_best
          s_best, h1_best, h2_best = s, h1, h2
          #pp [s_best, h1_best, h2_best]
        end
        
        break if s_best.zero? # allow early temination at maximum value 0.0
      end
      
      break if s_best.zero? # allow early temination at maximum value 0.0
    end
    
    #pp [s_best, h1_best, h2_best]
    f2cp[f] = [h1_best, h2_best]
  end
  
  # discretize based on cut points
  discretize_at_cutpoints!(f2cp, true)
end