package SNF;
use strict;
use base 'Continuity::Application';
use YAML qw();
use Algorithm::NaiveBayes;
use Algorithm::NaiveBayes::Model::Frequency;
use LWP::UserAgent;
use XML::RSS;
use Text::ParseWords;
use Data::Dumper;


my @categories = qw{
  Good Bad
};

my $datafile = '/tmp/cat.dat';

sub main {
  my ($self) = @_;
  $self->init_once;
  $self->init_always;
  while(1) {
    my $items = $self->getNews;
    my $f = $self->dispTemplate('main.tpl',
      items => $items,
      categories => [@categories]
    );
    my $category = $f->{category} || '';
    if ($category) {
      my $item = $f->{num};
      $self->train($category, $items->[$item]);
      $self->save;
    }
  }
}

sub init_once {
  my ($self) = @_;
  if(-e $datafile) {
    use Carp;
    carp('Loading from datafile.');
    my $bayes;
    $bayes = do $datafile;
    $self->{bayes} = $bayes;
  my $VAR1;
  eval(Dumper($self->{bayes}));
  $self->{bayes} = $VAR1;
  if($self->{bayes}->isa('Algorithm::NaiveBayes')) {
    carp("is a AlgorithmNaiveBayes!");
  } else {
    carp(" is NOT a A:NB");
  }
  } else {
    $self->{bayes} = Algorithm::NaiveBayes->new(purge => 0);
  if($self->{bayes}->isa('Algorithm::NaiveBayes')) {
    carp("is a AlgorithmNaiveBayes!");
  } else {
    carp(" is NOT a A:NB");
  }
  }
}

sub init_always {
  my ($self) = @_;
  $self->nocache('init_always');
  my $VAR1;
  eval(Dumper($self->{bayes}));
  $self->{bayes} = $VAR1;
  carp("dump: " . Dumper($self->{bayes}));
}


sub save {
  my ($self) = @_;
  carp("saving...");
  $Data::Dumper::Purity = 1;
  $Data::Dumper::Terse = 1;
  my $str = '$bayes = ' . Dumper($self->{bayes}) . ';';
  open(OUT,">$datafile") || carp("Error: $!");
  print OUT $str;
  close OUT;
}

sub getNews {
  my ($self) = @_;
  my @items;
  my $ua = LWP::UserAgent->new;

  my @news = (
    'http://download.freshmeat.net/backend/fm.rdf',
    'http://linuxtoday.com/backend/biglt.rss',
    'http://slashdot.org/index.rss',
    'http://www.bbc.co.uk/syndication/feeds/news/ukfs_news/front_page/rss091.xml',
    'http://www.perl.com/pace/perlnews.rdf'
  );

  foreach my $url (@news) {
    my $request = HTTP::Request->new(GET => $url);
    my $response = $ua->request($request);
    my $p = new XML::RSS;
    $p->parse($response->content);
    @items = (@items, @{$p->{items}});
  }


  for(my $i = 0; $i <= $#items ; $i++) {
    $items[$i]{num} = $i;
    $items[$i]{predict} = $self->predict($items[$i]);
  }
  @items = map { $_->[0] }
           sort { $b->[1] <=> $a->[1] }
           map { [ $_, ($_->{predict}{Good} - $_->{predict}{Bad}) ] }
           @items;
  my $i = 0;
  @items = map { $_->{num} = $i++ ; $_ } @items;
  return [@items];
}

sub train {
  my ($self, $category, $item) = @_;
  my $text = $item->{title} . ' ' . $item->{item} . ' ' . $item->{description};
  $text =~ s/[\n\r"']//g;
  carp("text: '$text'");
  my @words = quotewords('\s+',0,$text);
  my %wordcount;
  $wordcount{$_}++ foreach @words;
  carp("Training new words: " . Dumper({%wordcount}));
  $self->{bayes}->add_instance(
    attributes => { %wordcount },
    label => $category
  );
  $self->{bayes}->train();
}

sub predict {
  my ($self, $item) = @_;
  my $text = $item->{title} . ' ' . $item->{item} . ' ' . $item->{description};
  $text =~ s/[\n\r"']//g;
  my @words = quotewords('\s+',0,$text);
  my %wordcount;
  $wordcount{$_}++ foreach @words;
  my $labels = $self->{bayes}->predict(
    attributes => { %wordcount }
  );
  return $labels;
}

1;