view release on metacpan or  search on metacpan

lib/Algorithm/AdaBoost.pm  view on Meta::CPAN

  while ($num_iterations--) {
    # Construct a weak classifier which classifies data on the distribution.
    $weak_classifier = $weak_classifier_generator->(
      distribution => $distribution,
      training_set => $training_set,
    );
    $weight = $self->calculate_classifier_weight(
      classifier => $weak_classifier,
      distribution => $distribution,
    );
    push @weak_classifiers, +{
      classifier => $weak_classifier,
      weight => $weight,
    };
  } continue {
    $distribution = $self->construct_hardest_distribution(
      classifier => $weak_classifier,
      previous_distribution => $distribution,
      training_set => $training_set,
      weight => $weight,
    );
  }

  return $self->{final_classifier} = Algorithm::AdaBoost::Classifier->new(
    weak_classifiers => \@weak_classifiers,
  );
}

sub trained { exists shift->{final_classifier} }

sub training_set { shift->{training_set} }

sub weak_classifier_generator { shift->{weak_classifier_generator} }

1;
__END__

=head1 NAME

Algorithm::AdaBoost - AdaBoost learning algorithm

=head1 SYNOPSIS

  use Algorithm::AdaBoost;

  # Training phase.
  my $learner = Alogrithm::AdaBoost->new(
    training_set => [
      +{ feature => [...], label => 1, },
      +{ feature => [...], label => -1, },
      +{ feature => [...], label => -1, },
      ...
    ],
    weak_classifier_generator => \&my_poor_learning_algorithm,
  );
  $learner->train(num_iterations => 1_000);

  # Now you have a boost-ed classifier (Algorithm::AdaBoost::Classifier).
  my $classifier = $learner->final_classifier;
  given ($classifier->classify([...])) {
    when ($_ > 0) { say 'The data belongs to class 1.' }
    when ($_ < 0) { say 'The data belongs to class 2.' }
    default { warn 'The data cannot be classified.' }
  }

=head1 DESCRIPTION

AdaBoost is a machine learning algorithm proposed by Freund and Schapire.
Using an arbitrary binary classification algorithm, The algorithm can construct a more accurate classifier (i.e. it is a meta-algorithm).

=head1 METHODS

=head2 new

Constructor. You can specify 2 optional attributes:

=over 2

=item training_set

An ArrayRef which is used as a training data set.

Each item is a HashRef having 2 keys: C<feature> and C<label>. C<feature> is a arbitrary input that classifier accepts and C<label> is a expected output label (C<+1> or C<-1>).

=item weak_classifier_generator

A CodeRef which is expected to generate a binary classifier function.

When the function is called, 2 named parameters are specified like this:

  my $classifier = $generator->(
     distribution => [...],
     training_set => [...],
  );

C<distribution> is an ArrayRef which each item is a probability of corresponding item in C<training_set>. i.e. C<distribution> is P(X = t_i) where t_i is i-th item in C<training_set>.

The generated classifier is expected to be a CodeRef which takes 1 argument (value of C<feature>) and return C<+1> or C<-1> as a output label.

=back

Either of both can be overriden temporarily with parameters for C<train>.

=head2 classify

Shorthand for C<< $learner->final_classifier->classify >>.

=head2 final_classifier

Returns the last constructed classifier.

=head2 train

Constructs a stronger classifier from given training set and weak learning algorithm.

This method takes 1 mandatory parameter:

=over 2

=item num_iterations

Specifies how many training iterations to be excuted (i.e., how many weak classifiers to be generated).