package Clair::Classify
 Take in the model file generated by Learn.pm and then carry out the classification
 on the test data.

 Use the test data produced by Clair::Feature.pm (in svm_light format), and the model
 file generated by Clair::Learn.pm to carry out classification on the test data.
 The underlying algorithm can be either Naive Bayes or Perceptron. (default: perceptron)

 In case of perceptron, all you have to do is to obtain the dot product between the weight
 vector (W) and the feature vector (X), and determine whether they are positive or negative.

 Here, both the "test" and "model" parameter in the constructor is required.

  use Clair::Classify;

  my $cla = new Clair::Classify(DEBUG => $DEBUG, test => "test.dat", model => "model.file");
  $cla->classify($algo);

 The module should provide the ability to choose between different classifier
 algorithms.

 Classifies and then returns the results.
 Simple algorithm:

  if wt.f(x) > 0: classify x as positive group (+1)
  if wt.f(x) < 0: classify x as negative group (-1)

 
 A wrapper function for the underlying algorithms.

 The constructor. Initializes several container hashes for later use.

	my ($self) = @_;

	my $w0 = $self->{model_data}->{intercept};
	delete $self->{model_data}->{intercept};
	my $w = $self->{model_data};

	my $correct_count = 0;
	my $total_count = scalar @{ $self->{test_data} };

	my @return;
	for my $d (@{ $self->{test_data} })
	{

		my $y = $d->{class};
		my $x = $d->{features};
		my $com = $d->{comment};
		my ($junk, $grp, $path) = split /\s+/, $com;
		$com = "$grp:$path";
		# dot_product function is in Learn.pm - and we don't want to reimplement the sub.
		my $sum = $self->{learn}->dot_product($x, $w);
		my $s = $sum + $w0;
		# my $s = $sum;

		# $self->debugmsg("what??? \$s = $s",0) if($s == 0);

		my $I = ($s > 0) ? "+1" : "-1";
		my $yesno = ($I == $y) ? "y" : "n";
		$correct_count++ if($yesno eq "y");

		my @items = ($com, $s, $I, $y, $yesno);
		push @return,\@ items;
	}

	return (\@return, $correct_count, $total_count);

	my ($self, $algo) = @_;

	$algo = "_classify_perceptron" unless($algo);

	$self->debugmsg("running\$ self->$algo()", 2);

	return $self->$algo();

	my ($proto, %args) = @_;
	my $class = ref $proto || $proto;

	my $self = bless {}, $class;
	$DEBUG = $args{DEBUG} || $ENV{MYDEBUG};
	
	$self->{test} = "output.test";
	$self->{model} = "model";

	# overrides
	while ( my($k, $v) = each %args )
	{
		$self->{$k} = $v if(defined $v);
	}


	if(-f $self->{test})
	{
		# read in our features from the training data set
		my $fea = new Clair::Features(DEBUG => $DEBUG);
		$self->{test_data} = $fea->input($self->{test});
  }

	if( -f $self->{model})
	{
		# read in the model file
		$self->{learn} = new Clair::Learn(DEBUG => $DEBUG);
		$self->{model_data} = $self->{learn}->read_model($self->{model});
	}

  return $self;

 JB Kim
 jbremnant@gmail.com
 20070407