Class: SvmToolkit::Problem

Inherits:
Object
  • Object
show all
Defined in:
lib/svm_toolkit/problem.rb

Overview

Holds a set of labelled data.

Constant Summary collapse

SvmLight =

To select SvmLight input file format

0
Csv =

To select Csv input file format

1
Arff =

To select ARFF input file format

2

Class Method Summary collapse

Instance Method Summary collapse

Class Method Details

.from_array(instances, labels) ⇒ Object

Support constructing a problem from arrays of numbers (floating-point values).

  • instances - an array of instances, each instance being an array of numbers.

  • labels - an array of numbers, forming the labels for each instance.

An ArgumentError exception is raised if all the following conditions are not met:

  • the number of instances should equal the number of labels,

  • there must be at least one instance, and

  • every instance must have the same number of features.



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
 
# File 'lib/svm_toolkit/problem.rb', line 16

def self.from_array(instances, labels)
  unless instances.size == labels.size
    raise ArgumentError.new "Number of instances must equal number of labels"
  end
  unless instances.size > 0
    raise ArgumentError.new "There must be at least one instance."
  end
  unless instances.collect {|i| i.size}.min == instances.collect {|i| i.size}.max
    raise ArgumentError.new "All instances must have the same size"
  end

  problem = Problem.new
  problem.l = labels.size
  # -- add in the training data
  problem.x = Node[instances.size, instances[0].size].new
  instances.each_with_index do |instance, i|
    instance.each_with_index do |v, j|
      problem.x[i][j] = Node.new(j, v)
    end
  end
  # -- add in the labels
  problem.y = Java::double[labels.size].new
  labels.each_with_index do |v, i| 
    problem.y[i] = v
  end

  return problem
end

.from_file(filename, format = SvmLight) ⇒ Object

Read in a problem definition from a given filename, using format SvmLight (default), Csv or Arff.



57
58
59
60
61
62
63
64
65
66
 
# File 'lib/svm_toolkit/problem.rb', line 57

def self.from_file(filename, format = SvmLight)
  case format 
  when SvmLight
    return Problem.from_file_svmlight filename
  when Csv
    return Problem.from_file_csv filename
  when Arff
    return Problem.from_file_arff filename
  end
end

.from_file_arff(filename) ⇒ Object

Read in a problem definition in arff format, from given filename. Assumes all values are numbers (non-numbers converted to 0.0), and that the class is the last field.



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
 
# File 'lib/svm_toolkit/problem.rb', line 162

def self.from_file_arff filename
  instances = []
  labels = []
  max_index = 0
  found_data = false
  IO.foreach(filename) do |line|
    unless found_data
      puts "Ignoring", line
      found_data = line.downcase.strip == "@data"
      next # repeat the loop
    end
    tokens = line.split(",")
    labels << tokens.last.to_f
    instance = []
    tokens[1...-1].each_with_index do |value, index|
      instance << Node.new(index, value.to_f)
    end
    max_index = [tokens.size, max_index].max 
    instances << instance
  end
  max_index += 1 # to allow for 0 position
  unless instances.size == labels.size
    raise ArgumentError.new "Number of labels read differs from number of instances"
  end
  # now create a Problem definition
  problem = Problem.new
  problem.l = instances.size
  # -- add in the training data
  problem.x = Node[instances.size, max_index].new
  # -- fill with blank nodes
  instances.size.times do |i|
    max_index.times do |j|
      problem.x[i][j] = Node.new(i, 0)
    end
  end
  # -- add known values
  instances.each_with_index do |instance, i|
    instance.each do |node|
      problem.x[i][node.index] = node
    end
  end
  # -- add in the labels
  problem.y = Java::double[labels.size].new
  labels.each_with_index do |v, i| 
    problem.y[i] = v
  end

  return problem
end

.from_file_csv(filename) ⇒ Object

Read in a problem definition in csv format from given filename.



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
 
# File 'lib/svm_toolkit/problem.rb', line 115

def self.from_file_csv filename
  instances = []
  labels = []
  max_index = 0
  csv_data = CSV.parse(File.read(filename), headers: false)
  csv_data.each do |tokens|
    labels << tokens[0].to_f
    instance = []
    tokens[1..-1].each_with_index do |value, index|
      instance << Node.new(index, value.to_f)
    end
    max_index = [tokens.size, max_index].max 
    instances << instance
  end
  max_index += 1 # to allow for 0 position
  unless instances.size == labels.size
    raise ArgumentError.new "Number of labels read differs from number of instances"
  end
  # now create a Problem definition
  problem = Problem.new
  problem.l = instances.size
  # -- add in the training data
  problem.x = Node[instances.size, max_index].new
  # -- fill with blank nodes
  instances.size.times do |i|
    max_index.times do |j|
      problem.x[i][j] = Node.new(i, 0)
    end
  end
  # -- add known values
  instances.each_with_index do |instance, i|
    instance.each do |node|
      problem.x[i][node.index] = node
    end
  end
  # -- add in the labels
  problem.y = Java::double[labels.size].new
  labels.each_with_index do |v, i| 
    problem.y[i] = v
  end

  return problem
end

.from_file_svmlight(filename) ⇒ Object

Read in a problem definition in svmlight format.



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
 
# File 'lib/svm_toolkit/problem.rb', line 69

def self.from_file_svmlight filename
  instances = []
  labels = []
  max_index = 0
  IO.foreach(filename) do |line|
    tokens = line.split(" ")
    labels << tokens[0].to_f
    instance = []
    tokens[1..-1].each do |feature|
      index, value = feature.split(":")
      instance << Node.new(index.to_i, value.to_f)
      max_index = [index.to_i, max_index].max 
    end
    instances << instance
  end
  max_index += 1 # to allow for 0 position
  unless instances.size == labels.size
    raise ArgumentError.new "Number of labels read differs from number of instances"
  end
  # now create a Problem definition
  problem = Problem.new
  problem.l = instances.size
  # -- add in the training data
  problem.x = Node[instances.size, max_index].new
  # -- fill with blank nodes
  instances.size.times do |i|
    max_index.times do |j|
      problem.x[i][j] = Node.new(i, 0)
    end
  end
  # -- add known values
  instances.each_with_index do |instance, i|
    instance.each do |node|
      problem.x[i][node.index] = node
    end
  end
  # -- add in the labels
  problem.y = Java::double[labels.size].new
  labels.each_with_index do |v, i| 
    problem.y[i] = v
  end

  return problem
end

Instance Method Details

#label(n) ⇒ Object

Return label of nth instance



218
219
220
 
# File 'lib/svm_toolkit/problem.rb', line 218

def label(n)
  self.y[n]
end

#merge(problem) ⇒ Object

Create a new problem by combining the instances in this problem with those in the given problem.



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
 
# File 'lib/svm_toolkit/problem.rb', line 239

def merge problem
  unless self.x[0].size == problem.x[0].size
    raise ArgumentError.new "Cannot merge two problems with different numbers of features"
  end
  num_features = self.x[0].size
  num_instances = size + problem.size

  new_problem = Problem.new
  new_problem.l = num_instances
  new_problem.x = Node[num_instances, num_features].new
  new_problem.y = Java::double[num_instances].new
  # fill out the features
  num_instances.times do |i|
    num_features.times do |j|
      if i < size
        new_problem.x[i][j] = self.x[i][j]
      else
        new_problem.x[i][j] = problem.x[i-size][j]
      end
    end
  end
  # fill out the labels
  num_instances.times do |i|
    if i < size
      new_problem.y[i] = self.y[i]
    else
      new_problem.y[i] = problem.y[i-size]
    end
  end

  return new_problem
end

#rescale(min_value = 0.0, max_value = 1.0) ⇒ Object

Rescale values within problem to be in range min_value to max_value

For SVM models, it is recommended all features be in range [0,1] or [-1,1]



230
231
232
233
234
235
 
# File 'lib/svm_toolkit/problem.rb', line 230

def rescale(min_value = 0.0, max_value = 1.0)
  return if self.l.zero?
  x[0].size.times do |i|
    rescale_column(i, min_value, max_value)
  end
end

#size ⇒ Object

Returns the number of instances



213
214
215
 
# File 'lib/svm_toolkit/problem.rb', line 213

def size
  self.l
end

#values(n) ⇒ Object

Return array of values for nth instance



223
224
225
 
# File 'lib/svm_toolkit/problem.rb', line 223

def values(n)
  self.x[n].collect { it.value }
end