Class: TLearn::EM_Gaussian

Inherits:
Object
  • Object
show all
Defined in:
lib/t_learn/em.rb

Instance Attribute Summary collapse

Instance Method Summary collapse

Instance Attribute Details

#conv_list ⇒ Object

Returns the value of attribute conv_list.



12
13
14
 
# File 'lib/t_learn/em.rb', line 12

def conv_list
  @conv_list
end

#data_list ⇒ Object

Returns the value of attribute data_list.



12
13
14
 
# File 'lib/t_learn/em.rb', line 12

def data_list
  @data_list
end

#k_num ⇒ Object

Returns the value of attribute k_num.



12
13
14
 
# File 'lib/t_learn/em.rb', line 12

def k_num
  @k_num
end

#log_likelihood ⇒ Object

Returns the value of attribute log_likelihood.



12
13
14
 
# File 'lib/t_learn/em.rb', line 12

def log_likelihood
  @log_likelihood
end

#mu_list ⇒ Object

Returns the value of attribute mu_list.



12
13
14
 
# File 'lib/t_learn/em.rb', line 12

def mu_list
  @mu_list
end

#pi_list ⇒ Object

Returns the value of attribute pi_list.



12
13
14
 
# File 'lib/t_learn/em.rb', line 12

def pi_list
  @pi_list
end

#real_data_list ⇒ Object

Returns the value of attribute real_data_list.



12
13
14
 
# File 'lib/t_learn/em.rb', line 12

def real_data_list
  @real_data_list
end

Instance Method Details

#calc_ave(k, nk) ⇒ Object 



114
115
116
117
118
119
120
121
122
123
 
# File 'lib/t_learn/em.rb', line 114

def calc_ave(k, nk) 
  mu = Array.new(@dim)
  @dim.times{|i|
    mu[i] = @data_list.each_with_index.inject(0.0){|sum,(data, n)| 
      sum += @gamma[n][k] * data[i] 
    } / nk

  }
  return mu
end

#calc_conv(k, nk) ⇒ Object 



125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
 
# File 'lib/t_learn/em.rb', line 125

def calc_conv(k, nk)
  conv = Array.new(@dim).map{Array.new(@dim, 0)}
  @dim.times{|i|
    @dim.times{|j|
      @data_list.each_with_index{|data, n|
        conv[i][j] += @gamma[n][k] * (data[i]-@mu_list[k][i]) * (data[j]-@mu_list[k][j])
      } 
    }
  }
  conv = conv.map{|arr| 
    arr.map{|v| 
      (v/nk) != 0.0 ? (v/nk) : 0.1
    }
  }
  return conv
end

#calc_first_ave_std(x) ⇒ Object 



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
 
# File 'lib/t_learn/em.rb', line 184

def calc_first_ave_std(x)
  sum_each_vec = []
  ave_list = []
  std_list = []
  x.each{|vec| 
    vec.each_with_index{|data, i|
      sum_each_vec[i] = (sum_each_vec[i] == nil) ? data : sum_each_vec[i] + data
    }
  }
  x[0].size.times{|i| ave_list.push(sum_each_vec[i]/x.size)}

  sum_each_vec = []
  x.each{|vec| 
    vec.each_with_index{|data, i|
      sum_each_vec[i] = (sum_each_vec[i] == nil) ? (ave_list[i]-data)**2 : (sum_each_vec[i]+(ave_list[i]-data)**2)
    }
  }
  x[0].size.times{|i| 
    std = Math.sqrt(sum_each_vec[i]/x.size)
    std = 0.1 if std == 0.0 
    std_list.push(std)
  }

  return {:ave_list => ave_list, :std_list => std_list}
end

#calc_log_likelihood ⇒ Object 



142
143
144
145
146
147
148
149
150
151
152
 
# File 'lib/t_learn/em.rb', line 142

def calc_log_likelihood
  log_likelihood = 0.0
  @data_list.each_with_index{|data, i|
    sum = 0.0
    @k_num.times{|k|
      sum += @pi_list[k] * gauusian(data, @mu_list[k], @conv_list[k]) 
    }
    log_likelihood += Math.log(sum)
  }
  return log_likelihood
end

#create_log(cycle, likelihood) ⇒ Object 



55
56
57
58
59
60
61
62
 
# File 'lib/t_learn/em.rb', line 55

def create_log(cycle, likelihood)
  log = {:cycle => cycle, 
         :likelihood => likelihood,
         :mu => @mu_list.clone, 
         :conv => @conv_list.clone, 
         :pi_list => @pi_list.clone}
  return log
end

#e_step ⇒ Object 



89
90
91
92
93
94
95
96
97
98
99
 
# File 'lib/t_learn/em.rb', line 89

def e_step()
  @data_list.each_with_index{|data, n|
    denominator = 0.0
    @k_num.times{|k|
      denominator += @pi_list[k] * gauusian(data, @mu_list[k], @conv_list[k])
    }
    @k_num.times { |k|
      @gamma[n][k] = @pi_list[k] * gauusian(data, @mu_list[k], @conv_list[k]) / denominator
    }
  }
end

#fit(data_list, k_num) ⇒ Object 



65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
 
# File 'lib/t_learn/em.rb', line 65

def fit(data_list, k_num)
  init(data_list, k_num)
  result = []
  cycle = 0
  last_likelihood = calc_log_likelihood()
  loop do 
    e_step()
    m_step()
    likelihood = calc_log_likelihood()
    diff = (likelihood - last_likelihood).abs
    last_likelihood = likelihood
    puts "likelihood: #{likelihood}"
    result.push(create_log(cycle, likelihood))
    cycle += 1
    break if diff < 0.000001
  end
  puts "===================================="
  puts "pi : #{ @pi_list }"
  puts "mu : #{ @mu_list}"
  puts "conv : #{ @conv_list}"
  return result
end

#gauusian(x, mu, sigma) ⇒ Object

gauusian distribution



157
158
159
160
161
162
163
164
165
166
167
168
 
# File 'lib/t_learn/em.rb', line 157

def gauusian(x, mu, sigma)
  if @dim <= 1
    x = x[0]
    mu = mu[0]
    sigma = sigma[0][0]
    f1 = 1.0/(Math.sqrt(2.0*Math::PI)*Math.sqrt(sigma))
    f2 = Math.exp(-(((x-mu)**2)/((2.0*sigma))))
    return f1 * f2
  else
    return gauusian_over_2dim(x, mu, sigma)
  end
end

#gauusian_over_2dim(x, mu, conv) ⇒ Object

gauusian distribution over 2 dim version



174
175
176
177
178
179
180
181
182
 
# File 'lib/t_learn/em.rb', line 174

def gauusian_over_2dim(x, mu, conv)
  x = Matrix[x]
  mu = Matrix[mu]
  conv = Matrix[*conv]
  f1 = 1.0/(((2.0 * Math::PI)**(@dim/2.0)) * ( conv.det**(0.5) ))
  f2 = Math.exp((-1.0/2.0)*((x-mu) * conv.inverse * (x-mu).transpose)[0, 0])

  return (f1 * f2)
end

#ini_ave(ave_list) ⇒ Object 



27
28
29
30
31
32
33
 
# File 'lib/t_learn/em.rb', line 27

def ini_ave(ave_list)
  array = []
  @dim.times {|i|
    array.push(ave_list[i]*rand())
  }
  return array
end

#ini_conv(std_list) ⇒ Object 



35
36
37
38
39
40
41
 
# File 'lib/t_learn/em.rb', line 35

def ini_conv(std_list)
  conv = []
  @dim.times {|i|
    conv.push(make_array(i, std_list[i]))
  }
  return conv
end

#init(data_list, k_num) ⇒ Object 



14
15
16
17
18
19
20
21
22
23
24
25
 
# File 'lib/t_learn/em.rb', line 14

def init(data_list, k_num)
  @k_num = k_num 
  @data_list = data_list
  @dim = @data_list[0].size
  # @data_list = scale(@data_list)
  data_ave_std = calc_first_ave_std(@data_list)
  @real_data_list = Marshal.load(Marshal.dump(@data_list))
  @mu_list = Array.new(@k_num).map{ini_ave(data_ave_std[:ave_list])}
  @conv_list = Array.new(@k_num).map{ini_conv(data_ave_std[:std_list])}
  @pi_list = @k_num.times.map{rand()}
  @gamma = Array.new(@data_list.size).map{Array.new(@k_num, 0)}
end

#m_step ⇒ Object 



101
102
103
104
105
106
107
108
109
110
111
112
 
# File 'lib/t_learn/em.rb', line 101

def m_step()
  @k_num.times {|k|
    nk = 0.0
    @data_list.each_with_index{|data,  n| 
      nk += @gamma[n][k] 
    }

    @mu_list[k] = calc_ave(k, nk) 
    @conv_list[k] = calc_conv(k, nk)
    @pi_list[k] = nk/@data_list.size
  }
end

#make_array(i, std) ⇒ Object 



43
44
45
46
47
48
49
50
51
52
53
 
# File 'lib/t_learn/em.rb', line 43

def make_array(i, std)
  array = []
  @dim.times {|x|
    if i == x 
      array.push(std**2)
    else
      array.push(0.0)
    end
  }
  return array
end

#scale(x) ⇒ Object 



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
 
# File 'lib/t_learn/em.rb', line 211

def scale(x)
  sum_each_vec = []
  ave_list = []
  std_list = []
  x.each{|vec| 
    vec.each_with_index{|data, i|
      sum_each_vec[i] = (sum_each_vec[i] == nil) ? data : sum_each_vec[i]+data
    }
  }
  x[0].size.times{|i|
    ave_list.push(sum_each_vec[i]/x.size)
  }

  sum_each_vec = []
  x.each{|vec| 
    vec.each_with_index{|data, i|
      sum_each_vec[i] = (sum_each_vec[i] == nil) ? (ave_list[i]-data)**2 : (sum_each_vec[i]+(ave_list[i]-data)**2)
    }
  }
  x[0].size.times{|i|
    std_list.push(Math.sqrt(sum_each_vec[i]/x.size))
  }

  scaled_x = []
  x.each_with_index{|vec, i| 
    scaled_x[i] ||= []
    vec.each_with_index{|data, j|
      scaled_x[i][j] ||= (data-ave_list[j])/std_list[j]
    }
  }
  return scaled_x
end