flock 0.4.1 → 0.5.0

data/ext/kmeanspp.c

@@ -0,0 +1,129 @@
+#include <stdlib.h>
+
+extern double uniform();
+typedef struct clusterpoint {
+    double dist;
+    int n, chosen, closest;
+}   clusterpoint;
+
+int compare(const void *ptr1, const void *ptr2) {
+    clusterpoint *p1 = (clusterpoint *)ptr1, *p2 = (clusterpoint *)ptr2;
+    return p1->dist == p2->dist ? 0 : p1->dist < p2->dist ? -1 : 1;
+}
+
+double compute_distances(int ndata, int npoints,
+                         double **data, int **mask, double weight[], int transpose, clusterpoint dists[],
+                         double (*metric)(int, double**, double**, int**, int**, const double[], int, int, int)) {
+
+    int i, j, closest = 0;
+    double min, dist, total = 0;
+
+    // compute distances to chosen point
+    for (i = 0; i < npoints; i++) {
+        if (dists[i].chosen) continue;
+
+        min = -1;
+        for (j = 0; j < npoints; j++) {
+            if (!dists[j].chosen) continue;
+
+            dist = metric(ndata, data, data, mask, mask, weight, dists[i].n, dists[j].n, transpose);
+            if (min < 0 || min > dist) {
+                min     = dist;
+                closest = j;
+            }
+        }
+
+        dists[i].dist    = min * min;
+        dists[i].closest = closest;
+        total           += dists[i].dist;
+    }
+
+    return total;
+}
+
+void weightedassign(int nclusters, int nrows, int ncolumns,
+                    double** data, int** mask, double weight[], int transpose,
+                    double (*metric)(int, double**, double**, int**, int**, const double[], int, int, int),
+                    int clusterid[]) {
+
+    int i, n, chosen = (int)((double)nrows*uniform());
+    int ndata = (transpose == 0 ? ncolumns : nrows), npoints = (transpose == 0 ? nrows : ncolumns);
+    double total = 0, cutoff, curr;
+    clusterpoint dists[npoints];
+
+    for (i = 0; i < npoints; i++) {
+        dists[i].n      = i;
+        dists[i].chosen = 0;
+        dists[i].dist   = 0;
+    }
+
+    // setup 1st centroid
+    n                    = 1;
+    clusterid[chosen]    = 0;
+    dists[chosen].chosen = 1;
+
+    // pick k-points for k-clusters with a probability weighted by square of distance from closest centroid.
+    while (n < nclusters) {
+        total = compute_distances(ndata, npoints, data, mask, weight, transpose, dists, metric);
+        qsort((void*)dists, npoints, sizeof(clusterpoint), compare);
+
+        curr   = 0;
+        cutoff = total * uniform();
+        for (i = 0; i < npoints; i++) {
+            if (dists[i].chosen) continue;
+            curr += dists[i].dist;
+            if (curr >= cutoff || i == (npoints - 1)) {
+                clusterid[dists[i].n] = n++;
+                dists[i].chosen       = 1;
+                dists[i].dist         = 0;
+                break;
+            }
+        }
+    }
+
+    // assign remaining points to closest cluster
+    compute_distances(ndata, npoints, data, mask, weight, transpose, dists, metric);
+    for (n = 0; n < npoints; n++) {
+        if (dists[n].chosen) continue;
+        clusterid[dists[n].n] = clusterid[dists[dists[n].closest].n];
+    }
+}
+
+void spreadoutassign(int nclusters, int nrows, int ncolumns,
+                     double** data, int** mask, double weight[], int transpose,
+                     double (*metric)(int, double**, double**, int**, int**, const double[], int, int, int),
+                     int clusterid[]) {
+
+    int i, n, chosen = 0;
+    int ndata = (transpose == 0 ? ncolumns : nrows), npoints = (transpose == 0 ? nrows : ncolumns);
+    clusterpoint dists[npoints];
+
+    for (i = 0; i < npoints; i++) {
+        dists[i].n      = i;
+        dists[i].chosen = 0;
+        dists[i].dist   = 0;
+    }
+
+    // setup 1st centroid
+    n                    = 1;
+    clusterid[chosen]    = 0;
+    dists[chosen].chosen = 1;
+
+    // pick k-points for k-clusters with max distance from all centers.
+    chosen = npoints - 1;
+    while (n < nclusters) {
+        compute_distances(ndata, npoints, data, mask, weight, transpose, dists, metric);
+        qsort((void*)dists, npoints, sizeof(clusterpoint), compare);
+
+        clusterid[dists[chosen].n] = n++;
+        dists[chosen].chosen       = 1;
+        dists[chosen].dist         = 0;
+    }
+
+    // assign remaining points to closest cluster
+    compute_distances(ndata, npoints, data, mask, weight, transpose, dists, metric);
+    for (n = 0; n < npoints; n++) {
+        if (dists[n].chosen) continue;
+        clusterid[dists[n].n] = clusterid[dists[dists[n].closest].n];
+    }
+}

data/flock.gemspec

@@ -1,15 +1,15 @@
 # Generated by jeweler
 # DO NOT EDIT THIS FILE DIRECTLY
-# Instead, edit Jeweler::Tasks in Rakefile, and run the gemspec command
+# Instead, edit Jeweler::Tasks in Rakefile, and run 'rake gemspec'
 # -*- encoding: utf-8 -*-
 
 Gem::Specification.new do |s|
   s.name = %q{flock}
-  s.version = "0.4.1"
+  s.version = "0.5.0"
 
   s.required_rubygems_version = Gem::Requirement.new(">= 0") if s.respond_to? :required_rubygems_version=
   s.authors = ["Bharanee Rathna"]
-  s.date = %q{2011-04-28}
+  s.date = %q{2011-07-26}
   s.description = %q{A thin ruby binding to Cluster 3.0}
   s.email = ["deepfryed@gmail.com"]
   s.extensions = ["ext/extconf.rb"]
@@ -17,27 +17,22 @@ Gem::Specification.new do |s|
     "README.rdoc"
   ]
   s.files = [
+    "API.rdoc",
     "README.rdoc",
-     "Rakefile",
-     "VERSION",
-     "ext/cluster.c",
-     "ext/cluster.h",
-     "ext/extconf.rb",
-     "ext/flock.c",
-     "flock.gemspec",
-     "lib/flock.rb"
+    "Rakefile",
+    "VERSION",
+    "ext/cluster.c",
+    "ext/cluster.h",
+    "ext/extconf.rb",
+    "ext/flock.c",
+    "ext/kmeanspp.c",
+    "flock.gemspec",
+    "lib/flock.rb"
   ]
   s.homepage = %q{http://github.com/deepfryed/flock}
-  s.rdoc_options = ["--charset=UTF-8"]
   s.require_paths = ["lib"]
   s.rubygems_version = %q{1.3.7}
   s.summary = %q{Ruby bindings to Cluster 3.0.}
-  s.test_files = [
-    "examples/sparse.rb",
-     "examples/som.rb",
-     "examples/dense.rb",
-     "examples/treecluster.rb"
-  ]
 
   if s.respond_to? :specification_version then
     current_version = Gem::Specification::CURRENT_SPECIFICATION_VERSION

data/lib/flock.rb

@@ -1,50 +1,247 @@
 require_relative '../ext/flock'
+
+# Ruby bindings to data clustering algorithms provided by
+# {Cluster 3.0}[http://bonsai.hgc.jp/~mdehoon/software/cluster/software.htm]
+#
+# == Algorithms implemented
+#
+# * K-Means, K-Medians, K-Means++
+# * Self-Organizing Maps
+# * Tree Cluster or Hierarchical Clustering
+#
+# == Synopsis
+#
+#   require 'pp'
+#   require 'flock'
+#
+#   # sparse data.
+#   data = []
+#   data << %w(apple orange)
+#   data << %w(black white)
+#   data << %w(white cyan)
+#   data << %w(apple orange)
+#   data << %w(apple)
+#
+#   pp Flock.kcluster(2, data, sparse: true, seed: Flock::SEED_RANDOM)
+#   pp Flock.kcluster(2, data, sparse: true, seed: Flock::SEED_KMEANS_PLUSPLUS)
+#   pp Flock.kcluster(2, data, sparse: true, seed: Flock::SEED_SPREADOUT)
+#
+#   # dense data.
+#   data     = Array.new(13) {[]}
+#   mask     = Array.new(13) {[]}
+#   weights  = Array.new(13) {1.0}
+#
+#   data[0][0] = 0.1; data[0][1] = 0.0;
+#   data[1][0] = 1.4; data[1][1] = 1.3;
+#   data[2][0] = 1.2; data[2][1] = 2.5;
+#   data[3][0] = 2.3; data[3][1] = 1.5;
+#   data[4][0] = 1.7; data[4][1] = 0.7;
+#   data[5][0] = 0.0; data[5][1] = 3.9;
+#   data[6][0] = 6.7; data[6][1] = 3.9;
+#
+#   mask[0][0] = 1;   mask[0][1] = 1;
+#   mask[1][0] = 1;   mask[1][1] = 1;
+#   mask[2][0] = 1;   mask[2][1] = 1;
+#   mask[3][0] = 1;   mask[3][1] = 1;
+#   mask[4][0] = 1;   mask[4][1] = 1;
+#   mask[5][0] = 0;   mask[5][1] = 1;
+#   mask[6][0] = 1;   mask[6][1] = 1;
+#
+#   pp Flock.kcluster(2, data, mask: mask, weights: weights)
+#
+#
+# == See
+# * examples/* for more examples.
+# * README.rdoc for more details.
+# * API.rdoc is a public API overview.
 module Flock
 
-  def self.sparse_hash_to_data sparse_data
-    dims = Hash[sparse_data.map(&:keys).flatten.uniq.map.with_index{|k,v| [k,v]}]
-    data = sparse_data.map do |sv|
-      vector = Array.new(dims.size) {0}
-      sv.each {|k,v| vector[dims[k]] = v }
-      vector
+  # Cluster using k-means and k-medians.
+  #
+  # @example
+  #
+  #   data = []
+  #   data << %w(apple orange)
+  #   data << %w(black white)
+  #   data << %w(white cyan)
+  #   data << %w(apple orange)
+  #   data << %w(apple)
+  #   result = Flock.kcluster(2, data, sparse: true, seed: Flock::SEED_RANDOM)
+  #
+  # @param [Fixnum] size  number of clusters the data points are grouped into.
+  # @param [Array]  data  An array of arrays of sparse or dense data, or an array of hashes of sparse data. Dense data
+  #                       should always be in numeric form. Sparse data values are converted to a dense row format
+  #                       by looking at the unique values and then converting each data point into a numeric vector
+  #                       that represents the presence or absence of a value in that data point.
+  # @option options [Array]       :mask       An array of arrays of 1s and 0s denoting if an element in the datapoint is
+  #                                           to be used for computing distance (defaults to: all 1 vectors).
+  # @option options [Array]       :weights    Numeric weight for each data point (defaults to: all 1 vector).
+  # @option options [true, false] :transpose  Transpose the dense data matrix (defaults to: false).
+  # @option options [Fixnum]      :iterations Number of iterations to be run (defaults to: 100).
+  # @option options [Fixnum]      :method     Clustering method
+  #                                             - Flock::METHOD_AVERAGE (default)
+  #                                             - Flock::METHOD_MEDIAN
+  # @option options [Fixnum]      :metric     Distance measure, one of the following
+  #                                             - Flock::METRIC_EUCLIDIAN (default)
+  #                                             - Flock::METRIC_CITY_BLOCK
+  #                                             - Flock::METRIC_CORRELATION
+  #                                             - Flock::METRIC_ABSOLUTE_CORRELATION
+  #                                             - Flock::METRIC_UNCENTERED_CORRELATION
+  #                                             - Flock::METRIC_ABSOLUTE_UNCENTERED_CORRELATION
+  #                                             - Flock::METRIC_SPEARMAN
+  #                                             - Flock::METRIC_KENDALL
+  # @option options [Fixnum]      :seed       Initial seeding of clusters
+  #                                             - Flock::SEED_RANDOM (default)
+  #                                             - Flock::SEED_KMEANS_PLUSPLUS
+  #                                             - Flock::SEED_SPREADOUT
+  # @return [Hash]
+  #   {
+  #     :cluster  => [Array],
+  #     :centroid => [Array<Array>],
+  #     :error    => [Numeric],
+  #     :repeated => [Fixnum]
+  #   }
+  def self.kcluster size, data, options = {}
+    options[:sparse] = true if sparse?(data[0])
+    if options[:sparse]
+      data, options[:weights] = densify(data, options[:weights])
+      options[:mask]          = nil
     end
-    [dims, data]
+    do_kcluster(size, data, options)
   end
 
-  def self.sparse_array_to_data sparse_data
-    dims = Hash[sparse_data.flatten.uniq.map.with_index{|k,v| [k,v]}]
-    data = sparse_data.map do |sv|
-      vector = Array.new(dims.size) {0}
-      sv.each {|k| vector[dims[k]] = 1 }
-      vector
+  # Arranges data points on a 2D grid without having to specify a fixed cluster size. So in theory you could have
+  # a maximum of nxm clusters.
+  #
+  # @example
+  #
+  #   data = []
+  #   data << %w(apple orange)
+  #   data << %w(black white)
+  #   data << %w(white cyan)
+  #   data << %w(apple orange)
+  #   data << %w(apple)
+  #   result = Flock.self_organizing_map(2, 2, data, sparse: true)
+  #
+  # @param  [Fixnum]  nx          Grid size in 1st dimension (x)
+  # @param  [Fixnum]  ny          Grid size in 2nd dimension (y)
+  # @param  [Array]   data        See Flock#kcluster
+  # @option options   [Array]       :mask       See Flock#kcluster
+  # @option options   [true, false] :transpose  See Flock#kcluster
+  # @option options   [Fixnum]      :iterations See Flock#kcluster
+  # @option options   [Fixnum]      :metric     See Flock#kcluster
+  # @option options   [Numeric]     :tau        Initial tau value for distance metric.
+  # @return [Hash]
+  #   {
+  #     :cluster  => [Array<Array>],
+  #     :centroid => [Array<Array>]
+  #   }
+  def self.self_organizing_map nx, ny, data, options = {}
+    options[:sparse] = true if sparse?(data[0])
+    if options[:sparse]
+      data, options[:weights] = densify(data, options[:weights])
+      options[:mask]          = nil
     end
-    [dims, data]
+    do_self_organizing_map(nx, ny, data, options)
   end
 
-  def self.densify sparse_data, weights = nil
-    dims, data = sparse_data[0].kind_of?(Array) ? sparse_array_to_data(sparse_data) : sparse_hash_to_data(sparse_data)
-
-    if weights
-      resampled = Array.new(dims.size) {1}
-      weights.each {|k,v| resampled[dims[k]] = v }
-      weights   = resampled
+  # Clusters data into hierarchies and then returns the clusters required using cut-tree.
+  #
+  # @example
+  #
+  #   data = []
+  #   data << %w(apple orange)
+  #   data << %w(black white)
+  #   data << %w(white cyan)
+  #   data << %w(apple orange)
+  #   data << %w(apple)
+  #   result = Flock.treecluster(2, data, sparse: true)
+  #
+  # @param  [Fixnum]  size        Number of clusters required. (See Flock#kcluster)
+  # @param  [Array]   data        See Flock#kcluster
+  # @option options   [Array]       :mask       See Flock#kcluster
+  # @option options   [true, false] :transpose  See Flock#kcluster
+  # @option options   [Fixnum]      :iterations See Flock#kcluster
+  # @option options   [Fixnum]      :metric     See Flock#kcluster
+  # @option options   [Fixnum]      :method     Method to use for treecluster
+  #                                               - Flock::METHOD_SINGLE_LINKAGE
+  #                                               - Flock::METHOD_MAXIMUM_LINKAGE
+  #                                               - Flock::METHOD_AVERAGE_LINKAGE (default)
+  #                                               - Flock::METHOD_CENTROID_LINKAGE
+  # @return [Hash]
+  #   {
+  #     :cluster => [Array]
+  #   }
+  def self.treecluster size, data, options = {}
+    options[:sparse] = true if sparse?(data[0])
+    if options[:sparse]
+      data, options[:weights] = densify(data, options[:weights])
+      options[:mask]          = nil
     end
+    do_treecluster(size, data, options)
+  end
 
-    [data, weights]
+  # @deprecated use {kcluster} instead.
+  def self.kmeans size, data, options = {}
+    kcluster(size, data, options)
   end
 
-  def self.sparse_kmeans size, sparse_data, options = {}
-    data, options[:weights] = densify(sparse_data, options[:weights])
-    kmeans(size, data, data, options)
+  # @deprecated use {kcluster}(size, data, sparse: true, ...) instead.
+  def self.sparse_kmeans size, data, options = {}
+    kcluster(size, data, options.merge(sparse: true))
   end
 
-  def self.sparse_self_organizing_map nx, ny, sparse_data, options = {}
-    data, options[:weights] = densify(sparse_data, options[:weights])
-    self_organizing_map(nx, ny, data, data, options)
+  # @deprecated use {treecluster}(size, data, sparse: true, ...) instead.
+  def self.sparse_treecluster size, data, options = {}
+    treecluster(size, data, options.merge(sparse: true))
   end
 
-  def self.sparse_treecluster size, sparse_data, options = {}
-    data, options[:weights] = densify(sparse_data, options[:weights])
-    treecluster(size, data, data, options)
+  # @deprecated use {self_organizing_map}(nx, ny, data, sparse: true, ...) instead.
+  def self.sparse_self_organizing_map nx, ny, data, options = {}
+    self_organizing_map(nx, ny, data, options.merge(sparse: true))
   end
-end
+
+  private
+
+    def self.sparse? row
+      row.kind_of?(Hash) or !row[0].kind_of?(Numeric)
+    end
+
+    def self.sparse_array? row
+      !row.kind_of?(Hash)
+    end
+
+    def self.sparse_hash_to_data sparse_data
+      dims = Hash[sparse_data.map(&:keys).flatten.uniq.map.with_index{|k,v| [k,v]}]
+      data = sparse_data.map do |sv|
+        vector = Array.new(dims.size) {0}
+        sv.each {|k,v| vector[dims[k]] = v }
+        vector
+      end
+
+      [dims, data]
+    end
+
+    def self.sparse_array_to_data sparse_data
+      dims = Hash[sparse_data.flatten.uniq.map.with_index{|k,v| [k,v]}]
+      data = sparse_data.map do |sv|
+        vector = Array.new(dims.size) {0}
+        sv.each {|k| vector[dims[k]] = 1 }
+        vector
+      end
+
+      [dims, data]
+    end
+
+    def self.densify sparse_data, weights = nil
+      dims, data = sparse_array?(sparse_data[0]) ? sparse_array_to_data(sparse_data) : sparse_hash_to_data(sparse_data)
+
+      if weights
+        resampled = Array.new(dims.size) {1}
+        weights.each {|k,v| resampled[dims[k]] = v }
+        weights   = resampled
+      end
+
+      [data, weights]
+    end
+end # Flock