NAME

Algorithm::Kmeanspp - perl implementation of K-means++

SYNOPSIS

  use Algorithm::Kmeanspp;
  
  # input documents
  my %documents = (
      Alex => { 'Pop'     => 10, 'R&B'    => 6, 'Rock'   => 4 },
      Bob  => { 'Jazz'    => 8,  'Reggae' => 9                },
      Dave => { 'Classic' => 4,  'World'  => 4                },
      Ted  => { 'Jazz'    => 9,  'Metal'  => 2, 'Reggae' => 6 },
      Fred => { 'Hip-hop' => 3,  'Rock'   => 3, 'Pop'    => 3 },
      Sam  => { 'Classic' => 8,  'Rock'   => 1                },
  );
  
  my $kmp = Algorithm::Kmeanspp->new;
  
  foreach my $id (keys %documents) {
      $kmp->add_document($id, $documents{$id});
  }
  
  my $num_cluster = 3;
  my $num_iter    = 20;
  $kmp->do_clustering($num_cluster, $num_iter);             
  
  # show clustering result
  foreach my $cluster (@{ $kmp->clusters }) {
      print join "\t", @{ $cluster };
      print "\n";
  }
  # show cluster centroids
  foreach my $centroid (@{ $kmp->centroids }) {
      print join "\t", map { sprintf "%s:%.4f", $_, $centroid->{$_} }
          keys %{ $centroid };
      print "\n";
  }

DESCRIPTION

Algorithm::Kmeanspp is a perl implementation of K-means++.

METHODS

new

Create a new instance.

add_document($id, $vector)

Add an input document to the instance of Algorithm::Kmeanspp. $id parameter is the identifier of a document, and $vector parameter is the feature vector of a document. $vector parameter must be a hash reference, each key of $vector parameter is the identifier of the feature of documents and each value of $vector is the degree of the feature.

do_clustering($num_cluster, $num_iter)

Do clustering input documents. $num_cluster parameter specifies the number of output clusters, and $num_iter parameter specifies the number of clustering iterations.

clusters

This method is the accessor of clustering result. The output of the method is a array reference, and each item in the array reference includes the list of the identifiers of input documents in each cluster.

  # format of output clusters
  [
      [ document_id1, document_id2, ... ],  # cluster-1
      [ document_id3, document_id4, ... ],  # cluster-2
      ...
  ]

centroids

This method is the accessor of the vectors of cluster centroids.

AUTHOR

Mizuki Fujisawa <fujisawa@bayon.cc>

SEE ALSO

Wikipedia: K-means++

http://en.wikipedia.org/wiki/K-means%2B%2B

LICENSE

This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself.