| line | stmt | bran | cond | sub | pod | time | code | 
| 1 |  |  |  |  |  |  | package Data::CosineSimilarity; | 
| 2 | 2 |  |  | 2 |  | 56424 | use strict; | 
|  | 2 |  |  |  |  | 4 |  | 
|  | 2 |  |  |  |  | 79 |  | 
| 3 | 2 |  |  | 2 |  | 12 | use warnings; | 
|  | 2 |  |  |  |  | 5 |  | 
|  | 2 |  |  |  |  | 1709 |  | 
| 4 |  |  |  |  |  |  |  | 
| 5 |  |  |  |  |  |  | our $VERSION = 0.02; | 
| 6 |  |  |  |  |  |  |  | 
| 7 |  |  |  |  |  |  | =head1 NAME | 
| 8 |  |  |  |  |  |  |  | 
| 9 |  |  |  |  |  |  | Data::CosineSimilarity - Compute the Cosine Similarity | 
| 10 |  |  |  |  |  |  |  | 
| 11 |  |  |  |  |  |  | =head1 SYNOPSIS | 
| 12 |  |  |  |  |  |  |  | 
| 13 |  |  |  |  |  |  | $cs = Data::CosineSimilarity->new; | 
| 14 |  |  |  |  |  |  | $cs->add( label1 => { feature1 => 3, feature2 => 1, feature3 => 10 } ); | 
| 15 |  |  |  |  |  |  | $cs->add( label2 => ... ); | 
| 16 |  |  |  |  |  |  | $cs->add( label3 => ... ); | 
| 17 |  |  |  |  |  |  |  | 
| 18 |  |  |  |  |  |  | # computes the cosine similarity | 
| 19 |  |  |  |  |  |  | my $r = $cs->similarity( 'label1', 'label2' ); | 
| 20 |  |  |  |  |  |  |  | 
| 21 |  |  |  |  |  |  | # the result object | 
| 22 |  |  |  |  |  |  | my $cosine = $r->cosine; | 
| 23 |  |  |  |  |  |  | my $radian = $r->radian; | 
| 24 |  |  |  |  |  |  | my $degree = $r->degree; | 
| 25 |  |  |  |  |  |  | my ($label1, $label2) = $r->labels; | 
| 26 |  |  |  |  |  |  |  | 
| 27 |  |  |  |  |  |  | # computes all the cosine similarity between 'label1' and the others. | 
| 28 |  |  |  |  |  |  | my @all = $cs->all_for_label('label1'); | 
| 29 |  |  |  |  |  |  |  | 
| 30 |  |  |  |  |  |  | # computes all, and returns the best | 
| 31 |  |  |  |  |  |  | my ($best_label, $r) = $cs->best_for_label('label2'); | 
| 32 |  |  |  |  |  |  |  | 
| 33 |  |  |  |  |  |  | # computes all, and returns the worst | 
| 34 |  |  |  |  |  |  | my ($worst_label, $r) = $cs->worst_for_label('label2'); | 
| 35 |  |  |  |  |  |  |  | 
| 36 |  |  |  |  |  |  | =head1 DESCRIPTION | 
| 37 |  |  |  |  |  |  |  | 
| 38 |  |  |  |  |  |  | Compute the cosine similarities between a set of vectors. | 
| 39 |  |  |  |  |  |  |  | 
| 40 |  |  |  |  |  |  | =head2 $class->new( %opts ) | 
| 41 |  |  |  |  |  |  |  | 
| 42 |  |  |  |  |  |  | If all the feature vectors are normed then the computation of the cosine | 
| 43 |  |  |  |  |  |  | becomes just the dot product of the vectors. In this case, specify the | 
| 44 |  |  |  |  |  |  | option normed => 1, the performance will be greatly improved. | 
| 45 |  |  |  |  |  |  |  | 
| 46 |  |  |  |  |  |  | =cut | 
| 47 |  |  |  |  |  |  |  | 
| 48 |  |  |  |  |  |  | sub new { | 
| 49 | 4 |  |  | 4 | 1 | 2897 | my $class = shift; | 
| 50 | 4 |  |  |  |  | 8 | my %opts = @_; | 
| 51 | 4 | 50 |  |  |  | 32 | return bless { | 
| 52 |  |  |  |  |  |  | normed => $opts{normed} ? 1 : 0, | 
| 53 |  |  |  |  |  |  | labels => {}, | 
| 54 |  |  |  |  |  |  | }, $class; | 
| 55 |  |  |  |  |  |  | } | 
| 56 |  |  |  |  |  |  |  | 
| 57 |  |  |  |  |  |  | =head2 $self->add( label => $features ) | 
| 58 |  |  |  |  |  |  |  | 
| 59 |  |  |  |  |  |  | =cut | 
| 60 |  |  |  |  |  |  |  | 
| 61 |  |  |  |  |  |  | sub add { | 
| 62 | 9 |  |  | 9 | 1 | 2196 | my $self = shift; | 
| 63 | 9 |  |  |  |  | 13 | my ($label, $features) = @_; | 
| 64 | 9 | 50 |  |  |  | 22 | die 'label required' unless $label; | 
| 65 | 9 | 50 |  |  |  | 21 | die 'features required' unless $features; | 
| 66 | 9 | 50 |  |  |  | 29 | die 'features must be a hashref' | 
| 67 |  |  |  |  |  |  | unless ref $features eq 'HASH'; | 
| 68 | 9 | 50 |  |  |  | 28 | die 'features must contain terms' | 
| 69 |  |  |  |  |  |  | unless keys %$features; | 
| 70 |  |  |  |  |  |  |  | 
| 71 | 9 | 50 |  |  |  | 31 | my $norm = $self->{normed} ? 1 : _euclidean_norm($features); | 
| 72 |  |  |  |  |  |  |  | 
| 73 | 9 | 50 |  |  |  | 24 | die 'euclidean norm is null' if $norm == 0; | 
| 74 |  |  |  |  |  |  |  | 
| 75 | 9 |  |  |  |  | 49 | $self->{labels}{$label} = { | 
| 76 |  |  |  |  |  |  | features => $features, | 
| 77 |  |  |  |  |  |  | norm => $norm, | 
| 78 |  |  |  |  |  |  | }; | 
| 79 |  |  |  |  |  |  | } | 
| 80 |  |  |  |  |  |  |  | 
| 81 |  |  |  |  |  |  | sub _euclidean_norm { | 
| 82 | 9 |  |  | 9 |  | 14 | my ($features) = @_; | 
| 83 | 9 |  |  |  |  | 12 | my $sum = 0; | 
| 84 | 9 |  |  |  |  | 41 | $sum += $_**2 for values %$features; | 
| 85 | 9 |  |  |  |  | 24 | return sqrt $sum; | 
| 86 |  |  |  |  |  |  | } | 
| 87 |  |  |  |  |  |  |  | 
| 88 |  |  |  |  |  |  | sub _scalar_product { | 
| 89 | 7 |  |  | 7 |  | 11 | my ($features1, $features2) = @_; | 
| 90 | 7 |  |  |  |  | 8 | my $product = 0; | 
| 91 | 7 |  |  |  |  | 19 | for (keys %$features1) { | 
| 92 | 14 |  |  |  |  | 20 | my $c1 = $features1->{$_}; | 
| 93 | 14 | 100 |  |  |  | 38 | my $c2 = $features2->{$_} or next; | 
| 94 | 9 |  |  |  |  | 21 | $product += $c1 * $c2; | 
| 95 |  |  |  |  |  |  | } | 
| 96 | 7 |  |  |  |  | 18 | return $product; | 
| 97 |  |  |  |  |  |  | } | 
| 98 |  |  |  |  |  |  |  | 
| 99 |  |  |  |  |  |  | =head2 $self->similarity( $label1, $label2 ) | 
| 100 |  |  |  |  |  |  |  | 
| 101 |  |  |  |  |  |  | =cut | 
| 102 |  |  |  |  |  |  |  | 
| 103 |  |  |  |  |  |  | sub similarity { | 
| 104 | 7 |  |  | 7 | 1 | 18 | my $self = shift; | 
| 105 | 7 |  |  |  |  | 11 | my ($label1, $label2) = @_; | 
| 106 |  |  |  |  |  |  |  | 
| 107 | 7 |  |  |  |  | 25 | my $product = _scalar_product( | 
| 108 |  |  |  |  |  |  | $self->{labels}{$label1}{features}, | 
| 109 |  |  |  |  |  |  | $self->{labels}{$label2}{features} | 
| 110 |  |  |  |  |  |  | ); | 
| 111 |  |  |  |  |  |  |  | 
| 112 | 7 |  |  |  |  | 9 | my $cosine; | 
| 113 | 7 | 50 |  |  |  | 16 | if ($self->{normed}) { | 
| 114 | 0 |  |  |  |  | 0 | $cosine = $product; | 
| 115 |  |  |  |  |  |  | } | 
| 116 |  |  |  |  |  |  | else { | 
| 117 | 7 |  |  |  |  | 21 | $cosine = $product / ( $self->{labels}{$label1}{norm} * $self->{labels}{$label2}{norm} ); | 
| 118 |  |  |  |  |  |  | } | 
| 119 |  |  |  |  |  |  |  | 
| 120 | 7 |  |  |  |  | 33 | return Data::CosineSimilarity::Result->_new( | 
| 121 |  |  |  |  |  |  | labels => [ $label1, $label2 ], | 
| 122 |  |  |  |  |  |  | cosine => $cosine, | 
| 123 |  |  |  |  |  |  | ); | 
| 124 |  |  |  |  |  |  | } | 
| 125 |  |  |  |  |  |  |  | 
| 126 |  |  |  |  |  |  | =head2 $self->all_for_label( $label ) | 
| 127 |  |  |  |  |  |  |  | 
| 128 |  |  |  |  |  |  | =cut | 
| 129 |  |  |  |  |  |  |  | 
| 130 |  |  |  |  |  |  | sub all_for_label { | 
| 131 | 2 |  |  | 2 | 1 | 4 | my $self = shift; | 
| 132 | 2 |  |  |  |  | 3 | my ($label) = @_; | 
| 133 | 2 |  |  |  |  | 3 | my @result; | 
| 134 | 2 |  |  |  |  | 4 | for (keys %{ $self->{labels} }) { | 
|  | 2 |  |  |  |  | 9 |  | 
| 135 | 6 | 100 |  |  |  | 19 | next if $_ eq $label; | 
| 136 | 4 |  |  |  |  | 10 | push @result, $self->similarity($label, $_); | 
| 137 |  |  |  |  |  |  | } | 
| 138 | 2 |  |  |  |  | 10 | return sort { $b->cosine <=> $a->cosine } @result; | 
|  | 2 |  |  |  |  | 6 |  | 
| 139 |  |  |  |  |  |  | } | 
| 140 |  |  |  |  |  |  |  | 
| 141 |  |  |  |  |  |  | =head2 $self->best_for_label( $label ) | 
| 142 |  |  |  |  |  |  |  | 
| 143 |  |  |  |  |  |  | =cut | 
| 144 |  |  |  |  |  |  |  | 
| 145 |  |  |  |  |  |  | sub best_for_label { | 
| 146 | 1 |  |  | 1 | 1 | 6 | my $self = shift; | 
| 147 | 1 |  |  |  |  | 3 | my ($label) = @_; | 
| 148 | 1 |  |  |  |  | 5 | my @sorted = $self->all_for_label($label); | 
| 149 | 1 |  |  |  |  | 4 | my $r = shift @sorted; | 
| 150 | 1 |  |  |  |  | 3 | my (undef, $best) = $r->labels; | 
| 151 | 1 |  |  |  |  | 6 | return ($best, $r); | 
| 152 |  |  |  |  |  |  | } | 
| 153 |  |  |  |  |  |  |  | 
| 154 |  |  |  |  |  |  | =head2 $self->worst_for_label( $label ) | 
| 155 |  |  |  |  |  |  |  | 
| 156 |  |  |  |  |  |  | =cut | 
| 157 |  |  |  |  |  |  |  | 
| 158 |  |  |  |  |  |  | sub worst_for_label { | 
| 159 | 1 |  |  | 1 | 1 | 425 | my $self = shift; | 
| 160 | 1 |  |  |  |  | 2 | my ($label) = @_; | 
| 161 | 1 |  |  |  |  | 4 | my @sorted = $self->all_for_label($label); | 
| 162 | 1 |  |  |  |  | 3 | my $r = pop @sorted; | 
| 163 | 1 |  |  |  |  | 4 | my (undef, $worst) = $r->labels; | 
| 164 | 1 |  |  |  |  | 6 | return ($worst, $r); | 
| 165 |  |  |  |  |  |  | } | 
| 166 |  |  |  |  |  |  |  | 
| 167 |  |  |  |  |  |  | package Data::CosineSimilarity::Result; | 
| 168 | 2 |  |  | 2 |  | 32 | use strict; | 
|  | 2 |  |  |  |  | 4 |  | 
|  | 2 |  |  |  |  | 62 |  | 
| 169 | 2 |  |  | 2 |  | 10 | use warnings; | 
|  | 2 |  |  |  |  | 5 |  | 
|  | 2 |  |  |  |  | 62 |  | 
| 170 |  |  |  |  |  |  |  | 
| 171 | 2 |  |  | 2 |  | 984 | use Math::Trig; | 
|  | 2 |  |  |  |  | 22825 |  | 
|  | 2 |  |  |  |  | 8356 |  | 
| 172 |  |  |  |  |  |  |  | 
| 173 |  |  |  |  |  |  | sub _new { | 
| 174 | 7 |  |  | 7 |  | 10 | my $class = shift; | 
| 175 | 7 |  |  |  |  | 24 | my %args = @_; | 
| 176 | 7 |  |  |  |  | 33 | return bless \%args, $class; | 
| 177 |  |  |  |  |  |  | } | 
| 178 |  |  |  |  |  |  |  | 
| 179 | 7 |  |  | 7 |  | 3778 | sub labels { @{ $_[0]->{labels} } } | 
|  | 7 |  |  |  |  | 47 |  | 
| 180 |  |  |  |  |  |  |  | 
| 181 | 19 |  |  | 19 |  | 85 | sub cosine { $_[0]->{cosine} } | 
| 182 |  |  |  |  |  |  |  | 
| 183 | 10 |  |  | 10 |  | 26 | sub radian { acos( $_[0]->cosine ) } | 
| 184 |  |  |  |  |  |  |  | 
| 185 | 5 |  |  | 5 |  | 2531 | sub degree { rad2deg( $_[0]->radian ) } | 
| 186 |  |  |  |  |  |  |  | 
| 187 |  |  |  |  |  |  | =head1 AUTHOR | 
| 188 |  |  |  |  |  |  |  | 
| 189 |  |  |  |  |  |  | Antoine Imbert, C<<  >> | 
| 190 |  |  |  |  |  |  |  | 
| 191 |  |  |  |  |  |  | =head1 LICENSE AND COPYRIGHT | 
| 192 |  |  |  |  |  |  |  | 
| 193 |  |  |  |  |  |  | This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself. | 
| 194 |  |  |  |  |  |  |  | 
| 195 |  |  |  |  |  |  | =cut | 
| 196 |  |  |  |  |  |  |  | 
| 197 |  |  |  |  |  |  | 1; |