Added balanced model generation

Need wrapper update for utils!

1 files changed, 98 insertions, 0 deletions

diff --git a/balancer.rb b/balancer.rb
new file mode 100644
index 0000000..4ed2fd7
--- /dev/null
+++ b/balancer.rb
@@ -0,0 +1,98 @@
+# cuts a classification dataset into balanced pieces
+# let inact_act_ratio := majority_class.size/minority_class.size 
+# then: nr pieces = ceil(inact_act_ratio) if inact_act_ratio > 1.5
+# each piece contains the complete minority class and ceil(inact_act_ratio) majority class compounds.
+
+class Balancer
+
+  attr_accessor :inact_act_ratio, :act_hash, :inact_hash, :majority_splits, :nr_majority_splits, :errors, :datasets
+
+  # Supply a OpenTox::Dataset here
+  # Calculates inact_act_ratio, iff inact_act_ratio != +/-Infinity and no regression dataset is given
+  def initialize(dataset, feature_uri, creator_url)
+    @act_arr = []
+    @inact_arr = []
+    @inact_act_ratio = 1.0/0  # trick to define +infinity
+    @nr_majority_splits = 1   # +/-1 means: no split
+    @split = []               # splitted arrays with ids
+    @datasets = []            # result datasets
+    @errors = []
+
+    classification = true
+    if dataset.features.include?(feature_uri)
+      dataset.data.each do |i,a|
+        inchi = i
+        acts = a
+        acts.each do |act|
+          value = act[feature_uri]
+          if OpenTox::Utils.is_true?(value)
+            @act_arr << inchi
+          elsif OpenTox::Utils.classification?(value)
+            @inact_arr << inchi
+          else
+            classification = false
+            break;
+          end
+        end
+      end
+      @inact_act_ratio = @inact_arr.size.to_f / @act_arr.size.to_f unless (@act_arr.size == 0 or !classification) # leave alone for regression
+      set_nr_majority_splits
+      # perform majority split
+      @split = @nr_majority_splits > 0 ? shuffle_split(@inact_arr) : shuffle_split(@act_arr) unless @nr_majority_splits.abs == 1
+      @split.each do |s|
+        new_c = @nr_majority_splits > 0 ? s.concat(@act_arr) : s.concat(@inac_arr)
+        @datasets << dataset.create_new_dataset(new_c, [feature_uri], dataset.title, creator_url)
+      end
+
+    else
+      errors << "Feature not present in dataset."
+    end
+    errors << "Can not split regression dataset." unless classification
+  end
+
+
+
+  # sets nr of splits for majority class ('+', if inact_cnt > act_cnt, or '-' else), or leaves unchanged for illegal values.
+  def set_nr_majority_splits
+    @nr_majority_splits = @inact_act_ratio >= 1.5 ? @inact_act_ratio.ceil : ( @inact_act_ratio <= (2.0/3.0) ? -(1.0/@inact_act_ratio).ceil : ( @inact_act_ratio>1.0 ? 1 : -1) ) unless OpenTox::Utils.infinity?(@inact_act_ratio) # leave alone for regression
+  end
+
+  # does the actual shuffle and split
+  def shuffle_split (arr)
+    arr = arr.shuffle
+    arr.chunk(@nr_majority_splits.abs)
+  end
+
+  # turns a hash into a 2 col csv
+  def hsh2csv (hsh)
+    res=""
+    hsh.each do |k,v|
+      arr = [v,(@nr_majority_splits > 0 ? 0 : 1)]
+      res += arr.join(", ") + "\n"
+    end
+    res
+  end
+
+end
+
+class Array
+
+  # cuts an array into <num-pieces> chunks - returns a two-dimensional array
+  def chunk(pieces)
+    q, r = length.divmod(pieces)
+    (0..pieces).map { |i| i * q + [r, i].min }.enum_cons(2) \
+      .map { |a, b| slice(a...b) }
+  end
+
+  # shuffles the elements of an array
+  def shuffle( seed=nil )
+    srand seed.to_i if seed
+    sort_by { Kernel.rand }
+  end
+
+  # shuffels self
+  def shuffle!( seed=nil )
+    self.replace shuffle( seed )
+  end
+
+end