From 1e7671020d53201d4e269e1349e71938dc609a8b Mon Sep 17 00:00:00 2001
From: generall <vasnetsov93@gmail.com>
Date: Mon, 9 Nov 2015 19:33:47 +0300
Subject: [PATCH 01/13] minor changes

---
 lib/k_means/centroid.rb | 10 +++++++---
 lib/k_means/k_means.rb  |  5 +++--
 2 files changed, 10 insertions(+), 5 deletions(-)

diff --git a/lib/k_means/centroid.rb b/lib/k_means/centroid.rb
index 7bf02e2..71bd421 100644
--- a/lib/k_means/centroid.rb
+++ b/lib/k_means/centroid.rb
@@ -1,6 +1,9 @@
 class Centroid
   
   class << self
+  
+    # initial centroid positions are randomly chosen from within
+    # a bounding box that encloses all the nodes
     def create_centroids(amount, nodes)
       ranges = create_ranges(nodes, nodes[0].position.size)
       (1..amount).map do
@@ -13,14 +16,15 @@ def create_centroids(amount, nodes)
     
     private
     
+    # find centroid ranges as a bounding box for all nodes
     def create_ranges(nodes, dimensions)
-      ranges = Array.new(dimensions) {[0.0, 0.0]}
+      ranges = Array.new(dimensions) {[Float::NAN, Float::NAN]}
       nodes.each do |node|
         node.position.each_with_index do |position, index|
           # Bottom range
-          ranges[index][0] = position if position < ranges[index][0]
+          ranges[index][0] = position if ranges[index][0].nan? || position < ranges[index][0]
           # Top range
-          ranges[index][1] = position if position > ranges[index][1]
+          ranges[index][1] = position if ranges[index][1].nan? || position > ranges[index][1]
         end
       end
       ranges
diff --git a/lib/k_means/k_means.rb b/lib/k_means/k_means.rb
index f2138f7..e5ab415 100644
--- a/lib/k_means/k_means.rb
+++ b/lib/k_means/k_means.rb
@@ -2,7 +2,7 @@
 
 class KMeans
 
-  attr_reader :centroids, :nodes
+  attr_reader :centroids, :nodes, :max_iterations
 
   def initialize(data, options={})
     distance_measure = options[:distance_measure] || :euclidean_distance
@@ -10,6 +10,7 @@ def initialize(data, options={})
     @centroids = options[:custom_centroids] ||
       Centroid.create_centroids(options[:centroids] || 4, @nodes)
     @verbose = options[:verbose]
+    @max_iterations = options[:max_iterations] || 100
 
     perform_cluster_process
   end
@@ -26,7 +27,7 @@ def view
 
   def perform_cluster_process
     iterations, updates = 0, 1
-    while updates > 0 && iterations < 100
+    while updates > 0 && iterations < max_iterations
       iterations += 1
       verbose_message("Iteration #{iterations}")
       updates = 0

From cae0710cc32e48017e6f7fd8f0977a2025ca1103 Mon Sep 17 00:00:00 2001
From: generall <vasnetsov93@gmail.com>
Date: Tue, 10 Nov 2015 11:18:05 +0300
Subject: [PATCH 02/13] =?UTF-8?q?add=20Davies=E2=80=93Bouldin=20index?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 lib/k_means/centroid.rb | 38 +++++++++++++++++++++++++++++---------
 lib/k_means/k_means.rb  | 22 ++++++++++++++++++++--
 2 files changed, 49 insertions(+), 11 deletions(-)

diff --git a/lib/k_means/centroid.rb b/lib/k_means/centroid.rb
index 71bd421..06f2fdd 100644
--- a/lib/k_means/centroid.rb
+++ b/lib/k_means/centroid.rb
@@ -1,7 +1,7 @@
 class Centroid
-  
+
   class << self
-  
+
     # initial centroid positions are randomly chosen from within
     # a bounding box that encloses all the nodes
     def create_centroids(amount, nodes)
@@ -13,10 +13,10 @@ def create_centroids(amount, nodes)
         new(position)
       end
     end
-    
+
     private
-    
-    # find centroid ranges as a bounding box for all nodes
+
+    # find centroi  d ranges as a bounding box for all nodes
     def create_ranges(nodes, dimensions)
       ranges = Array.new(dimensions) {[Float::NAN, Float::NAN]}
       nodes.each do |node|
@@ -30,13 +30,33 @@ def create_ranges(nodes, dimensions)
       ranges
     end
   end
-  
+
   attr_accessor :position
-  
+
   def initialize(position)
     @position = position
+    @mean_distance = nil
   end
-  
+
+  def mean_node_distance
+
+    return @mean_distance if @mean_distance
+
+    total_dist = 0.0
+    total_nodes = @nodes.size
+
+    total_dist.reduce(0){|sum, node| sum + node.best_distance}
+
+    if total_nodes > 0
+      @mean_distance = total_dist/total_nodes
+    else
+      # if there no nodes in cluster, so the centroid is bad
+      @mean_distance = 1.0/0.0
+    end
+
+    @mean_distance
+  end
+
   # Finds the average distance of all the nodes assigned to
   # the centroid and then moves the centroid to that position
   def reposition(nodes, centroids)
@@ -49,5 +69,5 @@ def reposition(nodes, centroids)
     end
     @position = averages.map {|x| x / nodes.size}
   end
-  
+
 end
diff --git a/lib/k_means/k_means.rb b/lib/k_means/k_means.rb
index e5ab415..6c92ad0 100644
--- a/lib/k_means/k_means.rb
+++ b/lib/k_means/k_means.rb
@@ -5,8 +5,8 @@ class KMeans
   attr_reader :centroids, :nodes, :max_iterations
 
   def initialize(data, options={})
-    distance_measure = options[:distance_measure] || :euclidean_distance
-    @nodes = Node.create_nodes(data, distance_measure)
+    @distance_measure = options[:distance_measure] || :euclidean_distance
+    @nodes = Node.create_nodes(data, @distance_measure)
     @centroids = options[:custom_centroids] ||
       Centroid.create_centroids(options[:centroids] || 4, @nodes)
     @verbose = options[:verbose]
@@ -60,6 +60,24 @@ def update_nodes
     sum
   end
 
+  # Davies–Bouldin index: http://en.wikipedia.org/wiki/Cluster_analysis#Evaluation_of_clustering
+  def davies_bouldin_index(centroids = @centroids)
+    c_sz = centroids.size
+    db_index = 0
+    0..csz.each do |i|
+      max_db_index = -1.0/0
+      0..csz.each do |j|
+        if i != j
+          centroid_dist = centroids[i].position.send(@distance_measure, centroids[j].position)
+          sum_mean_nodes = centroids[i].mean_node_distance + centroids[j].mean_node_distance
+          max_db_index = [max_db_index, sum_mean_nodes / centroid_dist].max
+        end
+      end
+      db_index += max_db_index
+    end
+    return db_index/c_sz
+  end
+
   def reposition_centroids
     centroid_positions = @centroids.map(&:position)
     @centroids.each do |centroid|

From 703f583e4d7356060346470dfb2f0cda6fb4feca Mon Sep 17 00:00:00 2001
From: generall <vasnetsov93@gmail.com>
Date: Tue, 10 Nov 2015 11:26:27 +0300
Subject: [PATCH 03/13] =?UTF-8?q?make=20Davies=E2=80=93Bouldin=20public?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 lib/k_means/k_means.rb | 38 ++++++++++++++++++++------------------
 1 file changed, 20 insertions(+), 18 deletions(-)

diff --git a/lib/k_means/k_means.rb b/lib/k_means/k_means.rb
index 6c92ad0..fd91edf 100644
--- a/lib/k_means/k_means.rb
+++ b/lib/k_means/k_means.rb
@@ -23,6 +23,26 @@ def view
     @centroid_pockets
   end
 
+  # Davies–Bouldin index: http://en.wikipedia.org/wiki/Cluster_analysis#Evaluation_of_clustering
+  def davies_bouldin_index(centroids = @centroids)
+    c_sz = centroids.size
+    db_index = 0
+    0..csz.each do |i|
+      max_db_index = -1.0/0
+      0..csz.each do |j|
+        if i != j
+          centroid_dist = centroids[i].position.send(@distance_measure, centroids[j].position)
+          sum_mean_nodes = centroids[i].mean_node_distance + centroids[j].mean_node_distance
+          max_db_index = [max_db_index, sum_mean_nodes / centroid_dist].max
+        end
+      end
+      db_index += max_db_index
+    end
+    return db_index/c_sz
+  end
+
+
+
   private
 
   def perform_cluster_process
@@ -60,24 +80,6 @@ def update_nodes
     sum
   end
 
-  # Davies–Bouldin index: http://en.wikipedia.org/wiki/Cluster_analysis#Evaluation_of_clustering
-  def davies_bouldin_index(centroids = @centroids)
-    c_sz = centroids.size
-    db_index = 0
-    0..csz.each do |i|
-      max_db_index = -1.0/0
-      0..csz.each do |j|
-        if i != j
-          centroid_dist = centroids[i].position.send(@distance_measure, centroids[j].position)
-          sum_mean_nodes = centroids[i].mean_node_distance + centroids[j].mean_node_distance
-          max_db_index = [max_db_index, sum_mean_nodes / centroid_dist].max
-        end
-      end
-      db_index += max_db_index
-    end
-    return db_index/c_sz
-  end
-
   def reposition_centroids
     centroid_positions = @centroids.map(&:position)
     @centroids.each do |centroid|

From 0d3995cc473de939310522fda3b21ef3604f3a29 Mon Sep 17 00:00:00 2001
From: generall <vasnetsov93@gmail.com>
Date: Tue, 10 Nov 2015 11:39:05 +0300
Subject: [PATCH 04/13] fix NaN check

---
 lib/k_means/centroid.rb | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/lib/k_means/centroid.rb b/lib/k_means/centroid.rb
index 06f2fdd..6945afa 100644
--- a/lib/k_means/centroid.rb
+++ b/lib/k_means/centroid.rb
@@ -22,9 +22,9 @@ def create_ranges(nodes, dimensions)
       nodes.each do |node|
         node.position.each_with_index do |position, index|
           # Bottom range
-          ranges[index][0] = position if ranges[index][0].nan? || position < ranges[index][0]
+          ranges[index][0] = position if ranges[index][0].to_f.nan? || position < ranges[index][0]
           # Top range
-          ranges[index][1] = position if ranges[index][1].nan? || position > ranges[index][1]
+          ranges[index][1] = position if ranges[index][1].to_f.nan? || position > ranges[index][1]
         end
       end
       ranges

From 7299ca0e1bb19e8bd97afb376b4152074c40da17 Mon Sep 17 00:00:00 2001
From: generall <vasnetsov93@gmail.com>
Date: Tue, 10 Nov 2015 11:40:09 +0300
Subject: [PATCH 05/13] Fix errors

---
 lib/k_means/k_means.rb | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/lib/k_means/k_means.rb b/lib/k_means/k_means.rb
index fd91edf..201c544 100644
--- a/lib/k_means/k_means.rb
+++ b/lib/k_means/k_means.rb
@@ -27,9 +27,9 @@ def view
   def davies_bouldin_index(centroids = @centroids)
     c_sz = centroids.size
     db_index = 0
-    0..csz.each do |i|
+    0..c_sz.each do |i|
       max_db_index = -1.0/0
-      0..csz.each do |j|
+      0..c_sz.each do |j|
         if i != j
           centroid_dist = centroids[i].position.send(@distance_measure, centroids[j].position)
           sum_mean_nodes = centroids[i].mean_node_distance + centroids[j].mean_node_distance

From 35ef0a831fb22a36bd7a38125977cf0f73c78ce3 Mon Sep 17 00:00:00 2001
From: generall <vasnetsov93@gmail.com>
Date: Tue, 10 Nov 2015 11:41:51 +0300
Subject: [PATCH 06/13] Fix errors

---
 lib/k_means/k_means.rb | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/lib/k_means/k_means.rb b/lib/k_means/k_means.rb
index 201c544..a4931e3 100644
--- a/lib/k_means/k_means.rb
+++ b/lib/k_means/k_means.rb
@@ -27,9 +27,9 @@ def view
   def davies_bouldin_index(centroids = @centroids)
     c_sz = centroids.size
     db_index = 0
-    0..c_sz.each do |i|
+    (0..c_sz).each do |i|
       max_db_index = -1.0/0
-      0..c_sz.each do |j|
+      (0..c_sz).each do |j|
         if i != j
           centroid_dist = centroids[i].position.send(@distance_measure, centroids[j].position)
           sum_mean_nodes = centroids[i].mean_node_distance + centroids[j].mean_node_distance

From c06f63ae0482736e788e862a5dde1296a0af51bd Mon Sep 17 00:00:00 2001
From: generall <vasnetsov93@gmail.com>
Date: Tue, 10 Nov 2015 11:57:19 +0300
Subject: [PATCH 07/13] Fix errors

---
 lib/k_means/centroid.rb | 5 ++++-
 lib/k_means/k_means.rb  | 4 +++-
 2 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/lib/k_means/centroid.rb b/lib/k_means/centroid.rb
index 6945afa..fdf3b4a 100644
--- a/lib/k_means/centroid.rb
+++ b/lib/k_means/centroid.rb
@@ -31,11 +31,12 @@ def create_ranges(nodes, dimensions)
     end
   end
 
-  attr_accessor :position
+  attr_accessor :position, :nodes
 
   def initialize(position)
     @position = position
     @mean_distance = nil
+    @nodes = []
   end
 
   def mean_node_distance
@@ -65,6 +66,8 @@ def reposition(nodes, centroids)
     nodes.each do |node|
       node.position.each_with_index do |position, index|
         averages[index] += position
+        #Store closest nodes in the centroid object
+        centroid.nodes << node
       end
     end
     @position = averages.map {|x| x / nodes.size}
diff --git a/lib/k_means/k_means.rb b/lib/k_means/k_means.rb
index a4931e3..306159c 100644
--- a/lib/k_means/k_means.rb
+++ b/lib/k_means/k_means.rb
@@ -41,7 +41,9 @@ def davies_bouldin_index(centroids = @centroids)
     return db_index/c_sz
   end
 
-
+  def reset_nodes!
+    @nodes.each{|n| n.reset!}
+  end
 
   private
 

From d7ba88bea906c744d79d72107c7d4c918bd412ae Mon Sep 17 00:00:00 2001
From: generall <vasnetsov93@gmail.com>
Date: Tue, 10 Nov 2015 11:59:43 +0300
Subject: [PATCH 08/13] Fix errors

---
 lib/k_means/centroid.rb | 2 --
 lib/k_means/k_means.rb  | 2 ++
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/lib/k_means/centroid.rb b/lib/k_means/centroid.rb
index fdf3b4a..dc17703 100644
--- a/lib/k_means/centroid.rb
+++ b/lib/k_means/centroid.rb
@@ -66,8 +66,6 @@ def reposition(nodes, centroids)
     nodes.each do |node|
       node.position.each_with_index do |position, index|
         averages[index] += position
-        #Store closest nodes in the centroid object
-        centroid.nodes << node
       end
     end
     @position = averages.map {|x| x / nodes.size}
diff --git a/lib/k_means/k_means.rb b/lib/k_means/k_means.rb
index 306159c..76c10c0 100644
--- a/lib/k_means/k_means.rb
+++ b/lib/k_means/k_means.rb
@@ -68,6 +68,8 @@ def place_nodes_into_pockets
       @nodes.each_with_index do |node, node_index|
         if node.closest_centroid == centroid
           centroid_pockets[centroid_index] << node_index
+          #Store closest nodes in the centroid object
+          centroid.nodes << node
         end
       end
     end

From 8415bd0d30ebe528330f1d083940fc54758e8c43 Mon Sep 17 00:00:00 2001
From: generall <vasnetsov93@gmail.com>
Date: Tue, 10 Nov 2015 12:05:44 +0300
Subject: [PATCH 09/13] Fix errors

---
 lib/k_means/centroid.rb | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/lib/k_means/centroid.rb b/lib/k_means/centroid.rb
index dc17703..5b912aa 100644
--- a/lib/k_means/centroid.rb
+++ b/lib/k_means/centroid.rb
@@ -46,7 +46,7 @@ def mean_node_distance
     total_dist = 0.0
     total_nodes = @nodes.size
 
-    total_dist.reduce(0){|sum, node| sum + node.best_distance}
+    @nodes.reduce(0){|sum, node| sum + node.best_distance}
 
     if total_nodes > 0
       @mean_distance = total_dist/total_nodes

From e7e70046be1bbc0687528bcf501707d4e8b052f1 Mon Sep 17 00:00:00 2001
From: generall <vasnetsov93@gmail.com>
Date: Tue, 10 Nov 2015 12:07:13 +0300
Subject: [PATCH 10/13] Fix errors

---
 lib/k_means/k_means.rb | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/lib/k_means/k_means.rb b/lib/k_means/k_means.rb
index 76c10c0..6eb8f28 100644
--- a/lib/k_means/k_means.rb
+++ b/lib/k_means/k_means.rb
@@ -27,9 +27,9 @@ def view
   def davies_bouldin_index(centroids = @centroids)
     c_sz = centroids.size
     db_index = 0
-    (0..c_sz).each do |i|
+    (0..(c_sz - 1)).each do |i|
       max_db_index = -1.0/0
-      (0..c_sz).each do |j|
+      (0..(c_sz - 1)).each do |j|
         if i != j
           centroid_dist = centroids[i].position.send(@distance_measure, centroids[j].position)
           sum_mean_nodes = centroids[i].mean_node_distance + centroids[j].mean_node_distance

From 793ff9351032f81fbb520c1c1535193dc44deec6 Mon Sep 17 00:00:00 2001
From: generall <vasnetsov93@gmail.com>
Date: Tue, 10 Nov 2015 12:13:15 +0300
Subject: [PATCH 11/13] Fix errors

---
 lib/k_means/centroid.rb | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/lib/k_means/centroid.rb b/lib/k_means/centroid.rb
index 5b912aa..57ce7a1 100644
--- a/lib/k_means/centroid.rb
+++ b/lib/k_means/centroid.rb
@@ -46,7 +46,7 @@ def mean_node_distance
     total_dist = 0.0
     total_nodes = @nodes.size
 
-    @nodes.reduce(0){|sum, node| sum + node.best_distance}
+    total_dist = @nodes.reduce(0){|sum, node| sum + node.best_distance}
 
     if total_nodes > 0
       @mean_distance = total_dist/total_nodes

From bfc53d838ec33947df24a980d16b45bf5750a1aa Mon Sep 17 00:00:00 2001
From: generall <vasnetsov93@gmail.com>
Date: Wed, 11 Nov 2015 12:33:52 +0300
Subject: [PATCH 12/13] Add iterative clustering

---
 Rakefile                     |  2 +-
 lib/k_means/k_means.rb       | 34 ++++++++++++++------
 test/k_means/test_k_means.rb | 62 +++++++++++++++++++++++++++++-------
 3 files changed, 77 insertions(+), 21 deletions(-)

diff --git a/Rakefile b/Rakefile
index 1e817d3..502a68b 100644
--- a/Rakefile
+++ b/Rakefile
@@ -43,7 +43,7 @@ end
 
 task :default => :test
 
-require 'rake/rdoctask'
+require 'rdoc/task'
 Rake::RDocTask.new do |rdoc|
   if File.exist?('VERSION')
     version = File.read('VERSION')
diff --git a/lib/k_means/k_means.rb b/lib/k_means/k_means.rb
index 6eb8f28..1ecfd6b 100644
--- a/lib/k_means/k_means.rb
+++ b/lib/k_means/k_means.rb
@@ -2,17 +2,33 @@
 
 class KMeans
 
-  attr_reader :centroids, :nodes, :max_iterations
+  attr_reader :centroids, :nodes, :max_iterations, :max_tries
 
   def initialize(data, options={})
-    @distance_measure = options[:distance_measure] || :euclidean_distance
-    @nodes = Node.create_nodes(data, @distance_measure)
-    @centroids = options[:custom_centroids] ||
-      Centroid.create_centroids(options[:centroids] || 4, @nodes)
-    @verbose = options[:verbose]
-    @max_iterations = options[:max_iterations] || 100
-
-    perform_cluster_process
+
+    @max_tries = options[:max_tries] || 10
+
+    @best_DBI = Float::INFINITY
+    @best_centriods = nil
+
+    @max_tries.times do ||    
+      @distance_measure = options[:distance_measure] || :euclidean_distance
+      @nodes = Node.create_nodes(data, @distance_measure)
+      @centroids = options[:custom_centroids] ||
+        Centroid.create_centroids(options[:centroids] || 4, @nodes)
+      @verbose = options[:verbose]
+      @max_iterations = options[:max_iterations] || 100
+      perform_cluster_process
+      dbi = davies_bouldin_index
+      if dbi < @best_DBI
+        @best_DBI = dbi
+        @best_centriods = @centroids
+      end
+    end
+
+    @centroid = @best_centriods
+    place_nodes_into_pockets
+
   end
 
   def inspect
diff --git a/test/k_means/test_k_means.rb b/test/k_means/test_k_means.rb
index 82bed8b..6f9a5d3 100644
--- a/test/k_means/test_k_means.rb
+++ b/test/k_means/test_k_means.rb
@@ -1,11 +1,15 @@
 require 'helper'
 
+
 class TestKMeans < Test::Unit::TestCase
+
   context "A KMeans Instance" do
 
     setup do
       @data = Array.new(3) {Array.new(2) {rand}}
       @kmeans = KMeans.new(@data, :centroids => 2, :distance_measure => :cosine_similarity)
+      @kmeans2 = KMeans.new(data, :centroids => 2, :distance_measure => :euclidean_distance)
+      @kmeans3 = KMeans.new(data, :centroids => 3, :distance_measure => :euclidean_distance)
     end
 
     should "return an array" do
@@ -20,31 +24,67 @@ class TestKMeans < Test::Unit::TestCase
       assert_equal 3, @kmeans.nodes.size
     end
 
+    should "DBI of 3 clusters less than 2 clusters" do
+      assert @kmeans2.davies_bouldin_index > @kmeans3.davies_bouldin_index
+    end
+
   end
 
+
   context "A KMeans Instance with specified initial centroids" do
     setup do
       @data = Array.new(3) {Array.new(2) {rand}}
-      class CustomCentroid
-        attr_accessor :position
-        def initialize(position); @position = position; end
-        def reposition(nodes, centroid_positions); end
-      end
 
-      @specified_centroids = @data[0..2].map { |d| CustomCentroid.new(d) }
-      @kmeans = KMeans.new(@data, :custom_centroids => @specified_centroids, :distance_measure => :cosine_similarity)
+      #@specified_centroids = @data[0..2].map { |d| CustomCentroid.new(d) }
+      @kmeans = KMeans.new(@data, :centroids => 3, :distance_measure => :cosine_similarity)
     end
 
     should "return an inspected array" do
-      assert_kind_of String, @kmeans.inspect
+      assert_kind_of(String, @kmeans.inspect)
     end
 
     should "have 3 centroids" do
-      assert_equal 3, @kmeans.centroids.size
-   end
+      assert_equal(3, @kmeans.centroids.size)
+    end
 
     should "have 3 nodes" do
-      assert_equal 3, @kmeans.nodes.size
+      assert_equal(3, @kmeans.nodes.size)
     end
   end
+
+  def data
+    [
+      [ 1.0, 1.0 ],
+      [ 2.0, 1.0 ],
+      [ 2.0, 1.0 ],
+      [ 1.5, 1.0 ],
+      [ 1.8, 1.5 ],
+      [ 1.2, 1.4 ],
+      [ 1.9, 2.0 ],
+      [ 1.9, 1.0 ],
+      [ 1.0, 1.9 ],
+      [ 2.0, 2.0 ],
+      [ 4.3, 1.1 ],
+      [ 4.8, 1.2 ],
+      [ 4.3, 1.5 ],
+      [ 4.4, 1.2 ],
+      [ 4.1, 1.9 ],
+      [ 4.8, 1.7 ],
+      [ 4.2, 1.1 ],
+      [ 4.4, 1.7 ],
+      [ 5.8, 5.2 ],
+      [ 5.2, 6.2 ],
+      [ 5.5, 6.2 ],
+      [ 5.1, 6.2 ],
+      [ 5.2, 6.3 ],
+      [ 5.8, 6.1 ],
+      [ 5.6, 6.3 ],
+      [ 5.3, 6.4 ],
+      [ 5.5, 6.1 ],
+      [ 5.1, 6.2 ],
+      [ 5.9, 6.2 ],
+      [ 6.2, 5.1 ]
+    ]
+  end
+
 end

From f45a46efa095eadcabaa534ed6b325277d8bff28 Mon Sep 17 00:00:00 2001
From: generall <vasnetsov93@gmail.com>
Date: Thu, 12 Nov 2015 09:48:53 +0300
Subject: [PATCH 13/13] Add iterative centroid initialization

---
 lib/k_means/k_means.rb       | 56 ++++++++++++++++++++++++++----------
 lib/k_means/node.rb          |  6 ++--
 test/k_means/test_k_means.rb | 11 +++++--
 3 files changed, 53 insertions(+), 20 deletions(-)

diff --git a/lib/k_means/k_means.rb b/lib/k_means/k_means.rb
index 1ecfd6b..77a819b 100644
--- a/lib/k_means/k_means.rb
+++ b/lib/k_means/k_means.rb
@@ -1,4 +1,5 @@
 require 'ext/object'
+require 'pry'
 
 class KMeans
 
@@ -9,26 +10,44 @@ def initialize(data, options={})
     @max_tries = options[:max_tries] || 10
 
     @best_DBI = Float::INFINITY
-    @best_centriods = nil
+    @best_centroids = nil
+    @distance_measure = options[:distance_measure] || :euclidean_distance
+    @nodes = Node.create_nodes(data, @distance_measure)
+    @max_iterations = options[:max_iterations] || 100
+    @verbose = options[:verbose]
 
-    @max_tries.times do ||    
-      @distance_measure = options[:distance_measure] || :euclidean_distance
-      @nodes = Node.create_nodes(data, @distance_measure)
+    @centroid_count = options[:centroids] || 4
+
+    raise "Too many centroids(#{@centroid_count}) for #{@nodes.size} nodes" if @centroid_count > @nodes.size
+
+    @max_tries.times do |n|
+      reset_nodes!
       @centroids = options[:custom_centroids] ||
-        Centroid.create_centroids(options[:centroids] || 4, @nodes)
-      @verbose = options[:verbose]
-      @max_iterations = options[:max_iterations] || 100
+        Centroid.create_centroids(@centroid_count, @nodes)
+
       perform_cluster_process
       dbi = davies_bouldin_index
       if dbi < @best_DBI
+        p "#{n} #{dbi.round(2)} vs #{@best_DBI.round(2)}" if options[:debug]
         @best_DBI = dbi
-        @best_centriods = @centroids
+        @best_centroids = @centroids
       end
     end
 
-    @centroid = @best_centriods
+
+    if @best_centroids != nil
+      # can not fit nodes to all clusters
+      @centroids = @best_centroids
+    end
+
+
+
+    update_nodes(true) # force update
+    assign_nodes_to_centriods
     place_nodes_into_pockets
 
+    #binding.pry if options[:debug]
+
   end
 
   def inspect
@@ -58,7 +77,9 @@ def davies_bouldin_index(centroids = @centroids)
   end
 
   def reset_nodes!
-    @nodes.each{|n| n.reset!}
+    if @nodes
+      @nodes.each{|n| n.reset!}
+    end
   end
 
   private
@@ -72,7 +93,7 @@ def perform_cluster_process
       updates += update_nodes
       reposition_centroids
     end
-    place_nodes_into_pockets
+    assign_nodes_to_centriods
   end
 
   # This creates an array of arrays
@@ -84,18 +105,23 @@ def place_nodes_into_pockets
       @nodes.each_with_index do |node, node_index|
         if node.closest_centroid == centroid
           centroid_pockets[centroid_index] << node_index
-          #Store closest nodes in the centroid object
-          centroid.nodes << node
         end
       end
     end
     @centroid_pockets = centroid_pockets
   end
 
-  def update_nodes
+  def assign_nodes_to_centriods
+    @nodes.each_with_index do |node, node_index|
+      #Store closest nodes in the centroid object
+      node.closest_centroid.nodes << node
+    end
+  end
+
+  def update_nodes(force = false)
     sum = 0
     @nodes.each do |node|
-      sum += node.update_closest_centroid(@centroids)
+      sum += node.update_closest_centroid(@centroids, force)
     end
     sum
   end
diff --git a/lib/k_means/node.rb b/lib/k_means/node.rb
index 64f2ed9..cc09fc2 100644
--- a/lib/k_means/node.rb
+++ b/lib/k_means/node.rb
@@ -17,10 +17,10 @@ def initialize(position, similarity_measure)
     @similarity_measure = similarity_measure
   end
 
-  def update_closest_centroid(centroids)
+  def update_closest_centroid(centroids, force = false)
     # If we haven't processed this node we need to give it an initial centroid
     # so that we have something to compare distances against
-    calculate_initial_centroid(centroids.first) unless @closest_centroid
+    calculate_initial_centroid(centroids.first) if (!@closest_centroid || force)  
 
     updated = false
     centroids.each do |centroid|
@@ -60,7 +60,7 @@ def calculate_distance(centroid)
     begin
       @position.send(@similarity_measure, centroid.position)
     rescue NoMethodError
-      raise "Hey, '#{@similarity_measure}' is not a measurement. Read the REAdME for available measurements"
+      raise "Hey, '#{@similarity_measure}' is not a measurement. Read the README for available measurements"
     end
   end
 
diff --git a/test/k_means/test_k_means.rb b/test/k_means/test_k_means.rb
index 6f9a5d3..469a839 100644
--- a/test/k_means/test_k_means.rb
+++ b/test/k_means/test_k_means.rb
@@ -8,8 +8,6 @@ class TestKMeans < Test::Unit::TestCase
     setup do
       @data = Array.new(3) {Array.new(2) {rand}}
       @kmeans = KMeans.new(@data, :centroids => 2, :distance_measure => :cosine_similarity)
-      @kmeans2 = KMeans.new(data, :centroids => 2, :distance_measure => :euclidean_distance)
-      @kmeans3 = KMeans.new(data, :centroids => 3, :distance_measure => :euclidean_distance)
     end
 
     should "return an array" do
@@ -24,6 +22,15 @@ class TestKMeans < Test::Unit::TestCase
       assert_equal 3, @kmeans.nodes.size
     end
 
+
+  end
+
+  context "A Many tries" do
+    setup do
+      @kmeans2 = KMeans.new(data, :centroids => 2, :distance_measure => :euclidean_distance, :max_tries => 50)
+      @kmeans3 = KMeans.new(data, :centroids => 3, :distance_measure => :euclidean_distance, :max_tries => 50)
+    end
+
     should "DBI of 3 clusters less than 2 clusters" do
       assert @kmeans2.davies_bouldin_index > @kmeans3.davies_bouldin_index
     end