som 0.0.2 → 0.0.3

data/README.rdoc

@@ -1,10 +1,10 @@
 = SOM - Self Organising Map
 
-A pure Ruby implementation of the Self Organising Map machine learning Algorithm.
+A pure Ruby implementation of the Self Organising Map machine learning algorithm.
 
 == Install
  
-  gem sources -a -http://gemcutter.org
+  gem sources -a http://gemcutter.org
   sudo gem install som
   
 == How To Use
@@ -16,11 +16,30 @@ A pure Ruby implementation of the Self Organising Map machine learning Algorithm
   
   a = SOM.new(data, :number_of_nodes => 4, :dimensions => 3)
   a.train
+  
+  # To see which class a new piece of data fits into
+  new_data = [9,8,7]
+  
+  # An array is returned containing the index of the 
+  # training data that fits into the same class
+  # The index is the same as the index in the training data e.g:
+  # data[index_returned_by_SOM (2)] == data[2]
+  a.classify(new_data)
+    #=> [node_index, [training_data_index_1, training_data_index_2...]]
 
-  # Returns the index of the data you gave it
+  # Returns the id of a node and the 
+  # index of the data that belongs to it
   a.inspect
-    #=> [[1, 0...], [99, 84...], [11, 23...], [2, 6...]]
+    #=> [[0, [1, 0...]], [1, [99, 84...]], [2, [11, 23...]]]
+    
+== Options
+
+  SOM.new(data, :number_of_nodes => 1,   #Default: 5
+                :learning_rate => 0.7,   #Default: 0.5
+                :radius => 1,            #Default: number_of_nodes / 2
+                :max_iterations => 100,  #Default: 100
+                :verbose => true)        #Default: false
 
 == Copyright
 
-Copyright (c) 2009 Red Davis. See LICENSE for details.
+Copyright (c) 2009 Red Davis. See LICENSE for details.

data/VERSION

@@ -1 +1 @@
-0.0.2
+0.0.3

data/examples/example.rb

@@ -5,6 +5,8 @@ require 'benchmark'
 require 'rubygems'
 require 'normalizer'
 
+#SOM_DATA = Array.new(100) { Array.new(50) {rand}}
+
 min, max = Normalizer.find_min_and_max(SOM_DATA)
 
 normalizer = Normalizer.new(:min => min, :max => max)
@@ -15,9 +17,8 @@ SOM_DATA.each do |n|
   data << normalizer.normalize(n)
 end
   
-a = SOM.new(data, :nodes => 10, :dimensions => data[0].size)
-
-puts a.nodes.inspect
+a = SOM.new(data, :nodes => 8, 
+                  :radius => 0.8)
 
 puts a.global_distance_error
 
@@ -27,4 +28,6 @@ end
 
 puts a.global_distance_error
 
+#puts a.nodes.inspect
+
 puts times

data/lib/som/node.rb

@@ -1,9 +1,10 @@
 class Node
   
-  attr_reader :bucket
+  attr_reader :bucket, :id
   
-  def initialize(number_of_weights)
+  def initialize(id, number_of_weights)
     create_weights(number_of_weights)
+    @id = id
     @bucket = []
   end
   

data/lib/som.rb

@@ -4,14 +4,18 @@ class SOM
   
   def initialize(training_data, options={})
     @training_data = training_data
+    @dimensions = training_data[0].size
+    @iteration_count = 1
+    
+    # Options
     @number_of_nodes = options[:nodes] || 5
-    @dimensions = options[:dimensions]
     @learning_rate = options[:learning_rate] || 0.5
     @radius = options[:radius] || @number_of_nodes / 2
-    @iteration_count = 1
     @max_iterations = options[:max_iterations] || 100
+    
     # TODO: Allow a lambda so we can use different neighborhood functions
-    @neighborhood_function = options[:neighborhood_function] || 1
+    @neighborhood_function = 1 #options[:neighborhood_function] || 1
+    @verbose = options[:verbose]
     
     create_nodes(training_data)
   end
@@ -28,25 +32,25 @@ class SOM
     place_data_into_buckets(@training_data)
   end
   
-  # Returns an array of buckets containing the index of the data given
+  # Returns an array of buckets containing the index of the training data
   def inspect
-    nodes.map {|x| x.bucket.map {|x| x[0]}}
+    nodes.map {|x| [x.id, x.bucket] }
   end
   
-  # Return data from node that is closest to data
-  # You are returned a bucket which contains arrays that look like:
-  # [index, [data]]
-  # The index is the original index of that that was pumped into the classifier
+  # Return training data from the node that is closest to input data
+  # You are returned an array that look like:
+  # [node_id, [training_data_index_1, training_data_index_2...]]
+  # The index is the original index of that that was pumped into the SOM
   # during the training process
   def classify(data)
     closest_node = find_closest_node(data)
-    closest_node.bucket
+    [closest_node.id, closest_node.bucket]
   end
   
   # Taken from AI4R SOM library #107
   def global_distance_error
     @training_data.inject(0) do |sum, n|
-      sum + find_closest(n)[1]
+      sum + find_closest_node_with_distance(n)[1]
     end
   end
   
@@ -54,17 +58,25 @@ class SOM
   
   def train_it!(data)
     return false if @iteration_count >= @max_iterations
+    
+    print_message("Iteration: #{@iteration_count}")
         
-    data.each do |input|
+    data.each_with_index do |input, index|
+      print_message("\tLooking at data #{index+1}/#{data.size}")
+      
       # Update closest node
+      print_message("\t\tUpdating closest node")
+      
       closest_node = find_closest_node(input)
       closest_node.update_weight(@learning_rate, input)
     
       # Update nodes that closer than the radius
       other_nodes = nodes - [closest_node]
-      other_nodes.each do |node|
-        next if decayed_radius > node.distance_from(closest_node.weights)
-      
+      other_nodes.each_with_index do |node, index|
+        next if node.distance_from(closest_node.weights) > decayed_radius
+        
+        print_message("\t\tUpdating other nodes: #{index+1}/#{other_nodes.size}")
+        
         node.update_weight(@learning_rate, input, neighborhood_function)
       end
     end
@@ -72,34 +84,37 @@ class SOM
     increase_iteration_count!
   end
   
+  # This places the training data into its closest node's bucket.
   def place_data_into_buckets(data)
     data.each_with_index do |input, index|
       closest_node = find_closest_node(input)
-      closest_node << [index, input]
+      closest_node << index
     end
   end
   
   def decayed_radius
-    @radius - (0.1 * @radius * @iteration_count / @max_iterations)
+    @radius - (0.7 * @radius * @iteration_count / @max_iterations)
   end
   
   def decayed_learning_rate
-    @learning_rate - (0.5 * @learning_rate * @iteration_count / @max_iterations)
+    @learning_rate - (0.7 * @learning_rate * @iteration_count / @max_iterations)
   end
     
-  def increase_iteration_count!
-    @iteration_count += 1
-  end
-  
   def neighborhood_function
     0.5 * @neighborhood_function * @iteration_count / @max_iterations
   end
   
+  def increase_iteration_count!
+    @iteration_count += 1
+  end
+  
   def find_closest_node(data)
-    find_closest(data)[0]
+    find_closest_node_with_distance(data)[0]
   end
   
-  def find_closest(data)
+  # Finds the closest node to some data and returns the node
+  # and its distance from the data => [node, distance]
+  def find_closest_node_with_distance(data)
     closest_node = [nodes[0], nodes[0].distance_from(data)]
     
     nodes[1..-1].each do |node|
@@ -110,9 +125,15 @@ class SOM
     end
     closest_node
   end
-    
+  
   def create_nodes(data)
-    @number_of_nodes.times { nodes << Node.new(@dimensions) }
+    @number_of_nodes.times do |n|
+      nodes << Node.new(n, @dimensions)
+    end
+  end
+  
+  def print_message(message)
+    puts message if @verbose == true
   end
   
 end

data/som.gemspec

@@ -5,11 +5,11 @@
 
 Gem::Specification.new do |s|
   s.name = %q{som}
-  s.version = "0.0.2"
+  s.version = "0.0.3"
 
   s.required_rubygems_version = Gem::Requirement.new(">= 0") if s.respond_to? :required_rubygems_version=
   s.authors = ["reddavis"]
-  s.date = %q{2009-11-29}
+  s.date = %q{2009-11-30}
   s.description = %q{A Self Organising Map}
   s.email = %q{reddavis@gmail.com}
   s.extra_rdoc_files = [

data/spec/node_spec.rb

@@ -3,18 +3,22 @@ require File.expand_path(File.dirname(__FILE__) + '/spec_helper')
 describe "Node" do
   describe "Initialization" do
     before do
-      @a = Node.new(5)
+      @a = Node.new(1, 5)
     end
     
     it "should have 5 weights" do
       @a.weights.size.should == 5
     end
+    
+    it "should have an ID" do
+      @a.id.should == 1
+    end
   end
   
   describe "Distance Calculation" do
     before do
-      @a = Node.new(2)
-      @b = Node.new(2)
+      @a = Node.new(1, 2)
+      @b = Node.new(1, 2)
     end
     
     it "should return 0" do
@@ -37,7 +41,7 @@ describe "Node" do
   describe "Update Weight" do
     describe "Closest" do
       before do
-        @a = Node.new(2)
+        @a = Node.new(1, 2)
         @data = [1,2]
       end
     
@@ -50,7 +54,7 @@ describe "Node" do
     
     describe "Neighbor" do
       before do
-        @a = Node.new(2)
+        @a = Node.new(1, 2)
         @data = [1,2]
       end
     
@@ -64,7 +68,7 @@ describe "Node" do
   
   describe "Bucket" do
     before do
-      @a = Node.new(2)
+      @a = Node.new(1, 2)
     end
     
     it "should put data into the nodes bucket" do

data/spec/som_spec.rb

@@ -3,8 +3,8 @@ require File.expand_path(File.dirname(__FILE__) + '/spec_helper')
 describe "Som" do  
   describe "Training" do
     before do
-      @data = [[0,0], [0,0.5], [2,4], [6,5]]
-      @a = SOM.new(@data, :nodes => 1, :dimensions => 2)
+      @data = [[0,0]]
+      @a = SOM.new(@data, :nodes => 1)
     end
     
     it "should change the weight of the best matching node" do
@@ -22,31 +22,32 @@ describe "Som" do
     
     it "should preserve data indexes" do
       @a.train
-      
-      index_returned = @a.nodes[0].bucket[0][0]
-      data_returned = @a.nodes[0].bucket[0][1]
-      
-      @data[index_returned].should == data_returned
+            
+      index_returned = @a.nodes[0].bucket[0]
+      index_returned.should == 0
     end
   end
   
   describe "Inspect" do
     before do
       data = [[2,3]]
-      @a = SOM.new(data, :nodes => 1, :dimensions => 2)
+      @a = SOM.new(data, :nodes => 1)
+      @a.train
+    end
+    
+    it "should return the id of the nodes" do
+      @a.inspect[0][0].should == 0
     end
     
     it "should show the clusters of data indexes" do
-      @a.train
-      @a.inspect.should be_an(Array)
-      @a.inspect.size.should == 1
+      @a.inspect[0][1].should be_an(Array)
     end
   end
   
   describe "Clustering" do
     before do
       data = [[0,0], [999,999]]
-      @a = SOM.new(data, :nodes => 2, :dimensions => 2)
+      @a = SOM.new(data, :nodes => 2)
     end
     
     it "should belong to 2 seperate nodes" do
@@ -59,20 +60,29 @@ describe "Som" do
   describe "Classify" do
     before do
       data = [[0,0], [999,999]]
-      @a = SOM.new(data, :nodes => 2, :dimensions => 2)
+      a = SOM.new(data, :nodes => 1)
+      a.train
+      @a = a.classify([1,1])
     end
     
     it "should belong to 2 seperate nodes" do
-      @a.train
-      @a.classify([1,1]).should be_an(Array)
-      @a.classify([1,1]).size.should == 1
+      @a.should be_an(Array)
+      @a.size.should == 2
+    end
+    
+    it "should return a node id" do
+      @a[0].should == 0
+    end
+    
+    it "should return an array of training_data ids" do
+      @a[1].should be_an(Array)
     end
   end
   
   describe "Global Distance Error" do
     before do
       data = [[0,0], [999,999]]
-      @a = SOM.new(data, :nodes => 2, :dimensions => 2)
+      @a = SOM.new(data, :nodes => 2)
     end
     
     it "should return an integer" do

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification 
 name: som
 version: !ruby/object:Gem::Version 
-  version: 0.0.2
+  version: 0.0.3
 platform: ruby
 authors: 
 - reddavis
@@ -9,7 +9,7 @@ autorequire:
 bindir: bin
 cert_chain: []
 
-date: 2009-11-29 00:00:00 +00:00
+date: 2009-11-30 00:00:00 +00:00
 default_executable: 
 dependencies: 
 - !ruby/object:Gem::Dependency