fselector 1.0.1 → 1.1.0

data/lib/fselector/entropy.rb

@@ -36,8 +36,8 @@ module Entropy
   # @return [Float] H(X|Y)
   # @note vecX and vecY must be of same length
    def get_conditional_entropy(vecX, vecY)
-    abort "[#{__FILE__}@#{__LINE__}]: "+
-          "vector must be of same length" if not vecX.size == vecY.size
+    abort "[#{__FILE__}@#{__LINE__}]: \n"+
+          "  two vectors must be of same length" if not vecX.size == vecY.size
 
     hxy = 0.0
     n = vecX.size.to_f

data/lib/fselector/fileio.rb

@@ -27,9 +27,9 @@ module FileIO
   # @param [Integer] nclass number of classes
   # @param [Integer] nfeature number of features
   # @param [Integer] ncategory number of categories for each feature  
-  #  1 => binary feature with only on bit  
-  #  >1 => discrete feature with multiple values  
-  #  otherwise => continuous feature with vaule in the range of [0, 1)
+  #   1      # binary feature with only on bit  
+  #   >1     # discrete feature with multiple values  
+  #   other  # continuous feature with vaule in the range of [0, 1)
   # @param [true, false] allow_mv whether missing value of feature is alowed or not
   #
   def data_from_random(nsample=100, nclass=2, nfeature=10, ncategory=2, allow_mv=true)
@@ -38,7 +38,7 @@ module FileIO
     nsample.times do
       k = "c#{rand(nclass)+1}".to_sym
       
-      data[k] = [] if not data.has_key? k
+      data[k] ||= []
       
       feats = {}
       fs = (1..nfeature).to_a
@@ -57,7 +57,7 @@ module FileIO
         elsif ncategory > 1
           feats[f] = rand(ncategory)+1
         else
-          feats[f] = rand
+          feats[f] = rand.round(3) # round to 3-digit precision
         end
       end
       
@@ -77,7 +77,7 @@ module FileIO
   # ....
   #
   # @param [String] fname file to read from  
-  #   :stdin => read from standard input instead of file
+  #   :stdin  # read from standard input instead of file
   #
   def data_from_libsvm(fname=:stdin)
     data = {}
@@ -85,8 +85,8 @@ module FileIO
     if fname == :stdin
       ifs = $stdin
     elsif not File.exists? fname
-      abort "[#{__FILE__}@#{__LINE__}]: "+
-            "File '#{fname}' does not exist!"
+      abort "[#{__FILE__}@#{__LINE__}]: \n"+
+            "  File '#{fname}' does not exist!"
     else
       ifs = File.open(fname)
     end
@@ -94,7 +94,7 @@ module FileIO
     ifs.each_line do |ln|
       label, *features = ln.chomp.split(/\s+/)
       label = label.to_sym
-      data[label] = [] if not data.has_key? label
+      data[label] ||= []
       
       feats = {}
       features.each do |fv|
@@ -116,7 +116,7 @@ module FileIO
   # write to libsvm
   #
   # @param [String] fname file to write  
-  #   :stdout => write to standard ouput instead of file
+  #   :stdout  # write to standard ouput instead of file
   #
   def data_to_libsvm(fname=:stdout)
     if fname == :stdout
@@ -139,8 +139,8 @@ module FileIO
     
     each_sample do |k, s|
       ofs.print "#{k2idx[k]} "
-      s.keys.sort { |x, y| x.to_s.to_i <=> y.to_s.to_i }.each do |i|
-        ofs.print " #{f2idx[i]}:#{s[i]}" if not s[i].zero? # implicit mode
+      s.keys.sort { |x, y| f2idx[x] <=> f2idx[y] }.each do |f|
+        ofs.print " #{f2idx[f]}:#{s[f]}" if not s[f].zero? # implicit mode
       end
       ofs.puts
     end
@@ -155,20 +155,20 @@ module FileIO
   #
   # file should have the format with the first two rows
   # specifying features and their data types e.g.  
-  # feat1,feat2,...,featn  
-  # data\_type1,data\_type2,...,data\_typen  
+  # feat\_name1,feat\_name2,...,feat\_namen  
+  # feat\_type1,feat\_type2,...,feat\_typen  
   # 
   # and the remaing rows showing data e.g.  
   # class\_label,feat\_value1,feat\_value2,...,feat\_value3  
   # ...  
   # 
-  # allowed data types are:  
+  # allowed feature types (case-insensitive) are:  
   # INTEGER, REAL, NUMERIC, CONTINUOUS, STRING, NOMINAL, CATEGORICAL
   #
   # @param [String] fname file to read from  
-  #   :stdin => read from standard input instead of file
+  #   :stdin  # read from standard input instead of file
   #
-  # @note missing values allowed
+  # @note missing values are allowed, and feature types are stored as lower-case symbols
   #
   def data_from_csv(fname=:stdin)
     data = {}
@@ -176,29 +176,26 @@ module FileIO
     if fname == :stdin
       ifs = $stdin
     elsif not File.exists? fname
-      abort "[#{__FILE__}@#{__LINE__}]: "+
-            "File '#{fname}' does not exist!"
+      abort "[#{__FILE__}@#{__LINE__}]: \n"+
+            "  File '#{fname}' does not exist!"
     else
       ifs = File.open(fname)
     end
     
     first_row, second_row = true, true
-    feats, types = [], []
+    features, types = [], []
     
     ifs.each_line do |ln|
       if first_row # first row
         first_row = false
-        *feats = ln.chomp.split(/,/).to_sym
+        features = ln.chomp.split(/,/).to_sym
       elsif second_row # second row
         second_row = false
-        *types = ln.chomp.split(/,/)
-        if types.size == feats.size
-          types.each_with_index do |t, i|
-            set_opt(feats[i], t.upcase) # record data type
-          end
-        else
-          abort "[#{__FILE__}@#{__LINE__}]: "+
-                "the first two rows must have same number of fields"
+        # store feature type as lower-case symbol
+        types = ln.chomp.split(/,/).collect { |t| t.downcase.to_sym }
+        if not types.size == features.size
+          abort "[#{__FILE__}@#{__LINE__}]: \n"+
+                "  the first two rows must have same number of fields"
         end
       else # data rows
         label, *fvs = ln.chomp.split(/,/)
@@ -208,20 +205,20 @@ module FileIO
         fs = {}
         fvs.each_with_index do |v, i|
           next if v.empty? # missing value
-          data_type = get_opt(feats[i])
-          if data_type == 'INTEGER'
+          feat_type = types[i]
+          if feat_type == :integer
             v = v.to_i
-          elsif ['REAL', 'NUMERIC', 'CONTINUOUS'].include? data_type
+          elsif [:real, :numeric, :continuous].include? feat_type
             v = v.to_f
-          elsif ['STRING', 'NOMINAL', 'CATEGORICAL'].include? data_type
+          elsif [:string, :nominal, :categorical].include? feat_type
             #
           else
-            abort "[#{__FILE__}@#{__LINE__}]: "+
-                  "please specify correct data type "+
+            abort "[#{__FILE__}@#{__LINE__}]: \n"+
+                  "  please specify correct type "+
                   "for each feature in the 2nd row"
           end
           
-          fs[feats[i]] = v
+          fs[features[i]] = v
         end
         
         data[label] << fs
@@ -232,6 +229,11 @@ module FileIO
     ifs.close if not ifs == $stdin
     
     set_data(data)
+    set_features(features)
+    # set feature type
+    features.each_with_index do |f, i|
+      set_opt(f, types[i])
+    end
   end # data_from_csv
   
   
@@ -243,7 +245,7 @@ module FileIO
   # and the remaing rows showing data
   #
   # @param [String] fname file to write  
-  #   :stdout => write to standard ouput instead of file
+  #   :stdout  # write to standard ouput instead of file
   #
   def data_to_csv(fname=:stdout)
     if fname == :stdout
@@ -254,7 +256,7 @@ module FileIO
      
     ofs.puts get_features.join(',')
     ofs.puts get_features.collect { |f| 
-      get_opt(f) || 'STRING'
+      get_opt(f) || :string
     }.join(',')
     
     each_sample do |k, s|
@@ -270,7 +272,7 @@ module FileIO
     end
     
     # close file
-    ofs.close if not ofs == $stdout    
+    ofs.close if not ofs == $stdout
   end # data_to_csv
   
   
@@ -278,7 +280,7 @@ module FileIO
   # read from WEKA ARFF file
   #
   # @param [String] fname file to read from  
-  #   :stdin => read from standard input instead of file
+  #   :stdin  # read from standard input instead of file
   # @note it's ok if string containes spaces quoted by quote_char
   #
   def data_from_weka(fname=:stdin, quote_char='"')
@@ -287,13 +289,13 @@ module FileIO
     if fname == :stdin
       ifs = $stdin
     elsif not File.exists? fname
-      abort "[#{__FILE__}@#{__LINE__}]: "+
-            "File '#{fname}' does not exist!"
+      abort "[#{__FILE__}@#{__LINE__}]: \n"+
+            "  File '#{fname}' does not exist!"
     else
       ifs = File.open(fname)
     end
     
-    features, classes, comments = [], [], []
+    relation, features, classes, types, comments = '', [], [], [], []
     has_class, has_data = false, false
     
     ifs.each_line do |ln|
@@ -307,7 +309,6 @@ module FileIO
       # relation
       elsif ln =~ /^@RELATION/i
         tmp, relation = ln.split_me(/\s+/, quote_char)
-        set_opt('@RELATION', relation)
       # class attribute
       elsif ln =~ /^@ATTRIBUTE\s+class\s+{(.+)}/i
         has_class = true
@@ -318,13 +319,14 @@ module FileIO
         f = $1.to_sym
         features << f
         #$2.split_me(/,\s*/, quote_char) # feature nominal values
-        set_opt(f, 'NOMINAL')
+        types << :nominal
       # feature attribute (integer, real, numeric, string, date)
       elsif ln =~ /^@ATTRIBUTE/i
         tmp, v1, v2 = ln.split_me(/\s+/, quote_char)
         f = v1.to_sym
         features << f
-        set_opt(f, v2.upcase) # record feature data type
+        # store feture type as lower-case symbol
+        types << v2.downcase.to_sym
       # data header
       elsif ln =~ /^@DATA/i
         has_data = true
@@ -337,29 +339,30 @@ module FileIO
           label = label.to_sym
           
           fs = {}
-          nonzero_fi = []
+          # indices of feature with zero value
+          zero_fi = (0...features.size).to_a
+          
           feats.each do |fi_fv|
             fi, fv = fi_fv.split_me(/\s+/, quote_char)
             fi = fi.to_i             
-            add_feature_weka(fs, features[fi], fv)
-            nonzero_fi << fi
+            add_feature_weka(fs, features[fi], fv, types[fi])
+            zero_fi.delete(fi)
           end
           
           # feature with zero value
-          features.each_with_index do |f0, i|
-            add_feature_weka(fs, f0, 0) if not nonzero_fi.include?(i)
+          zero_fi.each do |zi|
+            add_feature_weka(fs, features[zi], 0, types[zi])
           end
           
           data[label] << fs
         else # regular ARFF
           feats = ln.split_me(/,\s*/, quote_char)
-          label = feats.pop.to_sym
+          label = feats.pop.to_sym          
           
           fs = {}
           feats.each_with_index do |fv, i|
-            add_feature_weka(fs, features[i], fv)
+            add_feature_weka(fs, features[i], fv, types[i])
           end
-          
           data[label] << fs if label
         end
       else
@@ -373,7 +376,11 @@ module FileIO
     set_data(data)
     set_classes(classes)
     set_features(features)
-    set_opt('COMMENTS', comments) if not comments.empty?
+    set_opt(:relation, relation)
+    features.each_with_index do |f, i|
+      set_opt(f, types[i])
+    end
+    set_opt(:comments, comments) if not comments.empty?
   end # data_from_weak
   
   
@@ -381,11 +388,11 @@ module FileIO
   # write to WEKA ARFF file
   #
   # @param [String] fname file to write  
-  #   :stdout => write to standard ouput instead of file
+  #   :stdout  # write to standard ouput instead of file
   # @param [Symbol] format sparse or regular ARFF  
-  #   :sparse => sparse ARFF, otherwise regular ARFF
+  #   :sparse  # sparse ARFF, otherwise regular ARFF
   #
-  def data_to_weka(fname=:stdout, format=:sparse)
+  def data_to_weka(fname=:stdout, format=nil)
     if fname == :stdout
       ofs = $stdout
     else
@@ -393,14 +400,14 @@ module FileIO
     end
     
     # comments
-    comments = get_opt('COMMENTS')
+    comments = get_opt(:comments)
     if comments
       ofs.puts comments.join("\n")
       ofs.puts
     end         
     
     # relation
-    relation = get_opt('@RELATION')
+    relation = get_opt(:relation)
     if relation
       ofs.puts "@RELATION #{relation}"
     else
@@ -412,15 +419,15 @@ module FileIO
     # feature attribute
     each_feature do |f|
       ofs.print "@ATTRIBUTE #{f} "
-      type = get_opt(f)
+      type = get_opt(f) # feature type
       if type
-        if type == 'NOMINAL'
+        if type == :nominal
           ofs.puts "{#{get_feature_values(f).uniq.sort.join(',')}}"
         else
           ofs.puts type
         end
-      else # treat all other data types as string
-        ofs.puts "STRING"
+      else # treat all other feature types as string
+        ofs.puts :string
       end
     end
     
@@ -462,21 +469,27 @@ module FileIO
   private
   
   # handle and add each feature for WEKA format
-  def add_feature_weka(fs, f, v)
+  #
+  # @param [Hash] fs sample that stores feature and its value
+  # @param [Symbol] f feature
+  # @param [String] v feature value
+  # @param [Symbol] type feature type
+  #
+  def add_feature_weka(fs, f, v, type)
     if v == '?' # missing value
       return
-    elsif get_opt(f) == 'INTEGER'
+    elsif type == :integer
       fs[f] = v.to_i
-    elsif get_opt(f) == 'REAL' or get_opt(f) == 'NUMERIC'
+    elsif type == :real or type == :numeric
       fs[f] = v.to_f
-    elsif get_opt(f) == 'STRING' or get_opt(f) == 'NOMINAL'
+    elsif type == :string or type == :nominal
       fs[f] = v
-    elsif get_opt(f) == 'DATE' # convert into integer
+    elsif type == :date # convert into integer
       fs[f] = (DateTime.parse(v)-DateTime.new(1970,1,1)).to_i
     else
        return
     end
-  end # add_feature
+  end # add_feature_weka
      
  
 end # module

data/lib/fselector/normalizer.rb

@@ -13,8 +13,8 @@ module Normalizer
         if s[f] > 0.0
           s[f] = Math.log(s[f], base) 
         else
-          abort "[#{__FILE__}@#{__LINE__}]: "+
-             "feature value must be positive"
+          abort "[#{__FILE__}@#{__LINE__}]: \n"+
+                "feature values must be positive!"
         end
       end
     end

data/lib/fselector/replace_missing_values.rb

@@ -19,22 +19,28 @@ module ReplaceMissingValues
     
     # clear variables
     clear_vars
-  end # replace_by_fixed_value
+  end # replace_by_fixed_value!
   
   
   #
   # replace missing feature value by mean feature value, 
   # applicable only to continuous feature
   #
+  # @param [Symbol] mode column or row mode  
+  #   - :by\_column # use the mean value of the same feature among all instances  
+  #   - :by\_row    # use the mean value of different features in current instance
+  #
   # @note data structure will be altered
   #
-  def replace_by_mean_value!
+  def replace_by_mean_value!(mode = :by_column)
     each_sample do |k, s|
+      mean = s.values.mean if mode == :by_row
+      
       each_feature do |f|
         fv = get_feature_values(f)
         next if fv.size == get_sample_size # no missing values
         
-        mean = fv.ave
+        mean = fv.ave if mode == :by_column
         if not s.has_key? f
           s[f] = mean
         end
@@ -46,6 +52,36 @@ module ReplaceMissingValues
   end # replace_by_mean_value!
   
   
+  #
+  # replace missing feature value by median feature value, 
+  # applicable only to continuous feature
+  #
+  # @param [Symbol] mode column or row mode  
+  #   - :by\_column # use the mean value of the same feature among all instances  
+  #   - :by\_row    # use the mean value of different features in current instance
+  #
+  # @note data structure will be altered
+  #
+  def replace_by_median_value!(mode = :by_column)
+    each_sample do |k, s|
+      median = s.values.median if mode == :by_row
+      
+      each_feature do |f|
+        fv = get_feature_values(f)
+        next if fv.size == get_sample_size # no missing values
+        
+        median = fv.median if mode == :by_column
+        if not s.has_key? f
+          s[f] = median
+        end
+      end
+    end
+    
+    # clear variables
+    clear_vars
+  end # replace_by_median_value!
+  
+  
   #
   # replace missing feature value by most seen feature value, 
   # applicable only to discrete feature
@@ -78,4 +114,102 @@ module ReplaceMissingValues
   end # replace_by_mean_value!
   
   
+  #
+  # replace missing feature value by weighted k-nearest neighbors' value, 
+  # applicable only to continuous feature
+  #
+  #     val = sigma_k (val_k * w_k)
+  #     
+  #     where w_k = (sum_d - d_k) / ((K-1) * sum_d)
+  #           sum_d = sigma_k (d_k)
+  #           K: number of d_k
+  #           sigma_k (w_k) = 1, normalized to 1
+  #
+  # @param [Integer] k number of nearest neighbors
+  # @note data structure will be altered, and the nearest neighbors 
+  #   are determined by Euclidean distance
+  #
+  # ref: [Microarray missing data imputation based on a set theoretic framework and biological knowledge](http://nar.oxfordjournals.org/content/34/5/1608)
+  #
+  def replace_by_knn_value!(k=1)
+    each_sample do |ki, si|
+      # potential features having missing value
+      mv_fs = get_features - si.keys
+      next if mv_fs.empty? # sample si has no missing value
+      
+      # record object value for each feature missing value
+      f2val = {}
+      mv_fs.each do |mv_f|
+        knn_s, knn_d = [], []
+        
+        each_sample do |kj, sj|
+          # sample sj also has missing value of mv_f
+          next if not sj.has_key? mv_f
+          
+          d = euclidean_distance(si, sj)
+          idx = knn_d.index { |di| d<di }
+          
+          if idx
+            knn_s.insert(idx, sj)
+            knn_d.insert(idx, d)
+            
+            if knn_s.size > k
+              knn_s = knn_s[0...k]
+              knn_d = knn_d[0...k]
+            end
+          else
+            if knn_s.size < k
+              knn_s << sj
+              knn_d << d
+            end
+          end
+        end
+        
+        # distance-weighted value from knn
+        knn_d_sum = knn_d.sum
+        sz = knn_d.size
+        val = 0.0
+        knn_s.each_with_index do |s, i|
+          if sz > 1
+            if not knn_d_sum.zero?
+              val += s[mv_f] * (knn_d_sum-knn_d[i]) / ((sz-1)*knn_d_sum)
+            else
+              val += s[mv_f] * 1.0 / sz
+            end
+          else # only one nearest neighbor
+            val = s[mv_f]
+          end
+        end
+        
+        f2val[mv_f] = val
+       #pp [si, mv_f, knn_s, knn_d, val]
+      end
+      
+      # set value
+      f2val.each do |f, v|
+        si[f] = v
+      end
+    end
+    
+    # clear variables
+    clear_vars
+  end # replace_by_knn_value!
+  
+  private
+  
+  # Euclidean distance of two samples
+  #
+  # @note features with missing value are ignored
+  def euclidean_distance(s1, s2)
+    d2 = 0.0
+    get_features.each do |f|
+      if s1.has_key? f and s2.has_key? f
+        d2 += (s1[f]-s2[f])**2
+      end
+    end
+    
+    Math.sqrt(d2)
+  end # euclidean_distance
+  
+  
 end # ReplaceMissingValues