| line | stmt | bran | cond | sub | pod | time | code | 
| 1 |  |  |  |  |  |  | package AI::FANN::Evolving::TrainData; | 
| 2 | 3 |  |  | 3 |  | 40993 | use strict; | 
|  | 3 |  |  |  |  | 8 |  | 
|  | 3 |  |  |  |  | 153 |  | 
| 3 | 3 |  |  | 3 |  | 19 | use List::Util 'shuffle'; | 
|  | 3 |  |  |  |  | 7 |  | 
|  | 3 |  |  |  |  | 340 |  | 
| 4 | 3 |  |  | 3 |  | 2940 | use AI::FANN ':all'; | 
|  | 0 |  |  |  |  |  |  | 
|  | 0 |  |  |  |  |  |  | 
| 5 |  |  |  |  |  |  | use Algorithm::Genetic::Diploid::Base; | 
| 6 |  |  |  |  |  |  | use base 'Algorithm::Genetic::Diploid::Base'; | 
| 7 |  |  |  |  |  |  |  | 
| 8 |  |  |  |  |  |  | our $AUTOLOAD; | 
| 9 |  |  |  |  |  |  | my $log = __PACKAGE__->logger; | 
| 10 |  |  |  |  |  |  |  | 
| 11 |  |  |  |  |  |  | =head1 NAME | 
| 12 |  |  |  |  |  |  |  | 
| 13 |  |  |  |  |  |  | AI::FANN::Evolving::TrainData - wrapper class for FANN data | 
| 14 |  |  |  |  |  |  |  | 
| 15 |  |  |  |  |  |  | =head1 METHODS | 
| 16 |  |  |  |  |  |  |  | 
| 17 |  |  |  |  |  |  | =over | 
| 18 |  |  |  |  |  |  |  | 
| 19 |  |  |  |  |  |  | =item new | 
| 20 |  |  |  |  |  |  |  | 
| 21 |  |  |  |  |  |  | Constructor takes named arguments. By default, ignores column | 
| 22 |  |  |  |  |  |  | named ID and considers column named CLASS as classifier. | 
| 23 |  |  |  |  |  |  |  | 
| 24 |  |  |  |  |  |  | =cut | 
| 25 |  |  |  |  |  |  |  | 
| 26 |  |  |  |  |  |  | sub new { | 
| 27 |  |  |  |  |  |  | my $self = shift->SUPER::new( | 
| 28 |  |  |  |  |  |  | 'ignore'    => [ 'ID'    ], | 
| 29 |  |  |  |  |  |  | 'dependent' => [ 'CLASS' ], | 
| 30 |  |  |  |  |  |  | 'header'    => {}, | 
| 31 |  |  |  |  |  |  | 'table'     => [], | 
| 32 |  |  |  |  |  |  | @_ | 
| 33 |  |  |  |  |  |  | ); | 
| 34 |  |  |  |  |  |  | my %args  = @_; | 
| 35 |  |  |  |  |  |  | $self->read_data($args{'file'}) if $args{'file'}; | 
| 36 |  |  |  |  |  |  | $self->trim_data if $args{'trim'}; | 
| 37 |  |  |  |  |  |  | return $self; | 
| 38 |  |  |  |  |  |  | } | 
| 39 |  |  |  |  |  |  |  | 
| 40 |  |  |  |  |  |  | =item ignore_columns | 
| 41 |  |  |  |  |  |  |  | 
| 42 |  |  |  |  |  |  | Getter/setter for column names to ignore in the train data structure. | 
| 43 |  |  |  |  |  |  | For example: an identifier columns named 'ID' | 
| 44 |  |  |  |  |  |  |  | 
| 45 |  |  |  |  |  |  | =cut | 
| 46 |  |  |  |  |  |  |  | 
| 47 |  |  |  |  |  |  | sub ignore_columns { | 
| 48 |  |  |  |  |  |  | my $self = shift; | 
| 49 |  |  |  |  |  |  | $self->{'ignore'} = \@_ if @_; | 
| 50 |  |  |  |  |  |  | return @{ $self->{'ignore'} }; | 
| 51 |  |  |  |  |  |  | } | 
| 52 |  |  |  |  |  |  |  | 
| 53 |  |  |  |  |  |  | =item dependent_columns | 
| 54 |  |  |  |  |  |  |  | 
| 55 |  |  |  |  |  |  | Getter/setter for column name(s) of the output value(s). | 
| 56 |  |  |  |  |  |  |  | 
| 57 |  |  |  |  |  |  | =cut | 
| 58 |  |  |  |  |  |  |  | 
| 59 |  |  |  |  |  |  | sub dependent_columns { | 
| 60 |  |  |  |  |  |  | my $self = shift; | 
| 61 |  |  |  |  |  |  | $self->{'dependent'} = \@_ if @_; | 
| 62 |  |  |  |  |  |  | return @{ $self->{'dependent'} }; | 
| 63 |  |  |  |  |  |  | } | 
| 64 |  |  |  |  |  |  |  | 
| 65 |  |  |  |  |  |  | =item predictor_columns | 
| 66 |  |  |  |  |  |  |  | 
| 67 |  |  |  |  |  |  | Getter for column name(s) of input value(s) | 
| 68 |  |  |  |  |  |  |  | 
| 69 |  |  |  |  |  |  | =cut | 
| 70 |  |  |  |  |  |  |  | 
| 71 |  |  |  |  |  |  | sub predictor_columns { | 
| 72 |  |  |  |  |  |  | my $self = shift; | 
| 73 |  |  |  |  |  |  | my @others = ( $self->ignore_columns, $self->dependent_columns ); | 
| 74 |  |  |  |  |  |  | my %skip = map { $_ => 1 } @others; | 
| 75 |  |  |  |  |  |  | return grep { ! $skip{$_} } keys %{ $self->{'header'} }; | 
| 76 |  |  |  |  |  |  | } | 
| 77 |  |  |  |  |  |  |  | 
| 78 |  |  |  |  |  |  | =item predictor_data | 
| 79 |  |  |  |  |  |  |  | 
| 80 |  |  |  |  |  |  | Getter for rows of input values | 
| 81 |  |  |  |  |  |  |  | 
| 82 |  |  |  |  |  |  | =cut | 
| 83 |  |  |  |  |  |  |  | 
| 84 |  |  |  |  |  |  | sub predictor_data { | 
| 85 |  |  |  |  |  |  | my ( $self, %args ) = @_; | 
| 86 |  |  |  |  |  |  | my $i = $args{'row'}; | 
| 87 |  |  |  |  |  |  | my @cols = $args{'cols'} ? @{ $args{'cols'} } : $self->predictor_columns; | 
| 88 |  |  |  |  |  |  |  | 
| 89 |  |  |  |  |  |  | # build hash of indices to keep | 
| 90 |  |  |  |  |  |  | my %keep = map { $self->{'header'}->{$_} => 1 } @cols; | 
| 91 |  |  |  |  |  |  |  | 
| 92 |  |  |  |  |  |  | # only return a single row | 
| 93 |  |  |  |  |  |  | if ( defined $i ) { | 
| 94 |  |  |  |  |  |  | my @pred; | 
| 95 |  |  |  |  |  |  | for my $j ( 0 .. $#{ $self->{'table'}->[$i] } ) { | 
| 96 |  |  |  |  |  |  | push @pred, $self->{'table'}->[$i]->[$j] if $keep{$j}; | 
| 97 |  |  |  |  |  |  | } | 
| 98 |  |  |  |  |  |  | return \@pred; | 
| 99 |  |  |  |  |  |  | } | 
| 100 |  |  |  |  |  |  | else { | 
| 101 |  |  |  |  |  |  | my @preds; | 
| 102 |  |  |  |  |  |  | my $max = $self->size - 1; | 
| 103 |  |  |  |  |  |  | for my $j ( 0 .. $max ) { | 
| 104 |  |  |  |  |  |  | push @preds, $self->predictor_data( 'row' => $j, 'cols' => \@cols); | 
| 105 |  |  |  |  |  |  | } | 
| 106 |  |  |  |  |  |  | return @preds; | 
| 107 |  |  |  |  |  |  | } | 
| 108 |  |  |  |  |  |  | } | 
| 109 |  |  |  |  |  |  |  | 
| 110 |  |  |  |  |  |  | =item dependent_data | 
| 111 |  |  |  |  |  |  |  | 
| 112 |  |  |  |  |  |  | Getter for dependent (classifier) data | 
| 113 |  |  |  |  |  |  |  | 
| 114 |  |  |  |  |  |  | =cut | 
| 115 |  |  |  |  |  |  |  | 
| 116 |  |  |  |  |  |  | sub dependent_data { | 
| 117 |  |  |  |  |  |  | my ( $self, $i ) = @_; | 
| 118 |  |  |  |  |  |  | my @dc = map { $self->{'header'}->{$_} } $self->dependent_columns; | 
| 119 |  |  |  |  |  |  | if ( defined $i ) { | 
| 120 |  |  |  |  |  |  | return [ map { $self->{'table'}->[$i]->[$_] } @dc ]; | 
| 121 |  |  |  |  |  |  | } | 
| 122 |  |  |  |  |  |  | else { | 
| 123 |  |  |  |  |  |  | my @dep; | 
| 124 |  |  |  |  |  |  | for my $j ( 0 .. $self->size - 1 ) { | 
| 125 |  |  |  |  |  |  | push @dep, $self->dependent_data($j); | 
| 126 |  |  |  |  |  |  | } | 
| 127 |  |  |  |  |  |  | return @dep; | 
| 128 |  |  |  |  |  |  | } | 
| 129 |  |  |  |  |  |  | } | 
| 130 |  |  |  |  |  |  |  | 
| 131 |  |  |  |  |  |  | =item read_data | 
| 132 |  |  |  |  |  |  |  | 
| 133 |  |  |  |  |  |  | Reads provided input file | 
| 134 |  |  |  |  |  |  |  | 
| 135 |  |  |  |  |  |  | =cut | 
| 136 |  |  |  |  |  |  |  | 
| 137 |  |  |  |  |  |  | sub read_data { | 
| 138 |  |  |  |  |  |  | my ( $self, $file ) = @_; # file is tab-delimited | 
| 139 |  |  |  |  |  |  | $log->debug("reading data from file $file"); | 
| 140 |  |  |  |  |  |  | open my $fh, '<', $file or die "Can't open $file: $!"; | 
| 141 |  |  |  |  |  |  | my ( %header, @table ); | 
| 142 |  |  |  |  |  |  | while(<$fh>) { | 
| 143 |  |  |  |  |  |  | chomp; | 
| 144 |  |  |  |  |  |  | next if /^\s*$/; | 
| 145 |  |  |  |  |  |  | my @fields = split /\t/, $_; | 
| 146 |  |  |  |  |  |  | if ( not %header ) { | 
| 147 |  |  |  |  |  |  | my $i = 0; | 
| 148 |  |  |  |  |  |  | %header = map { $_ => $i++ } @fields; | 
| 149 |  |  |  |  |  |  | } | 
| 150 |  |  |  |  |  |  | else { | 
| 151 |  |  |  |  |  |  | push @table, \@fields; | 
| 152 |  |  |  |  |  |  | } | 
| 153 |  |  |  |  |  |  | } | 
| 154 |  |  |  |  |  |  | $self->{'header'} = \%header; | 
| 155 |  |  |  |  |  |  | $self->{'table'}  = \@table; | 
| 156 |  |  |  |  |  |  | return $self; | 
| 157 |  |  |  |  |  |  | } | 
| 158 |  |  |  |  |  |  |  | 
| 159 |  |  |  |  |  |  | =item write_data | 
| 160 |  |  |  |  |  |  |  | 
| 161 |  |  |  |  |  |  | Writes to provided output file | 
| 162 |  |  |  |  |  |  |  | 
| 163 |  |  |  |  |  |  | =cut | 
| 164 |  |  |  |  |  |  |  | 
| 165 |  |  |  |  |  |  | sub write_data { | 
| 166 |  |  |  |  |  |  | my ( $self, $file ) = @_; | 
| 167 |  |  |  |  |  |  |  | 
| 168 |  |  |  |  |  |  | # use file or STDOUT | 
| 169 |  |  |  |  |  |  | my $fh; | 
| 170 |  |  |  |  |  |  | if ( $file ) { | 
| 171 |  |  |  |  |  |  | open $fh, '>', $file or die "Can't write to $file: $!"; | 
| 172 |  |  |  |  |  |  | $log->info("writing data to $file"); | 
| 173 |  |  |  |  |  |  | } | 
| 174 |  |  |  |  |  |  | else { | 
| 175 |  |  |  |  |  |  | $fh = \*STDOUT; | 
| 176 |  |  |  |  |  |  | $log->info("writing data to STDOUT"); | 
| 177 |  |  |  |  |  |  | } | 
| 178 |  |  |  |  |  |  |  | 
| 179 |  |  |  |  |  |  | # print header | 
| 180 |  |  |  |  |  |  | my $h = $self->{'header'}; | 
| 181 |  |  |  |  |  |  | print $fh join "\t", sort { $h->{$a} <=> $h->{$b} } keys %{ $h }; | 
| 182 |  |  |  |  |  |  | print $fh "\n"; | 
| 183 |  |  |  |  |  |  |  | 
| 184 |  |  |  |  |  |  | # print rows | 
| 185 |  |  |  |  |  |  | for my $row ( @{ $self->{'table'} } ) { | 
| 186 |  |  |  |  |  |  | print $fh join "\t", @{ $row }; | 
| 187 |  |  |  |  |  |  | print $fh "\n"; | 
| 188 |  |  |  |  |  |  | } | 
| 189 |  |  |  |  |  |  | } | 
| 190 |  |  |  |  |  |  |  | 
| 191 |  |  |  |  |  |  | =item trim_data | 
| 192 |  |  |  |  |  |  |  | 
| 193 |  |  |  |  |  |  | Trims sparse rows with missing values | 
| 194 |  |  |  |  |  |  |  | 
| 195 |  |  |  |  |  |  | =cut | 
| 196 |  |  |  |  |  |  |  | 
| 197 |  |  |  |  |  |  | sub trim_data { | 
| 198 |  |  |  |  |  |  | my $self = shift; | 
| 199 |  |  |  |  |  |  | my @trimmed; | 
| 200 |  |  |  |  |  |  | ROW: for my $row ( @{ $self->{'table'} } ) { | 
| 201 |  |  |  |  |  |  | next ROW if grep { not defined $_ } @{ $row }; | 
| 202 |  |  |  |  |  |  | push @trimmed, $row; | 
| 203 |  |  |  |  |  |  | } | 
| 204 |  |  |  |  |  |  | my $num = $self->{'size'} - scalar @trimmed; | 
| 205 |  |  |  |  |  |  | $log->info("removed $num incomplete rows"); | 
| 206 |  |  |  |  |  |  | $self->{'table'} = \@trimmed; | 
| 207 |  |  |  |  |  |  | } | 
| 208 |  |  |  |  |  |  |  | 
| 209 |  |  |  |  |  |  | =item sample_data | 
| 210 |  |  |  |  |  |  |  | 
| 211 |  |  |  |  |  |  | Sample a fraction of the data | 
| 212 |  |  |  |  |  |  |  | 
| 213 |  |  |  |  |  |  | =cut | 
| 214 |  |  |  |  |  |  |  | 
| 215 |  |  |  |  |  |  | sub sample_data { | 
| 216 |  |  |  |  |  |  | my $self   = shift; | 
| 217 |  |  |  |  |  |  | my $sample = shift || 0.5; | 
| 218 |  |  |  |  |  |  | my $clone1 = $self->clone; | 
| 219 |  |  |  |  |  |  | my $clone2 = $self->clone; | 
| 220 |  |  |  |  |  |  | my $size   = $self->size; | 
| 221 |  |  |  |  |  |  | my @sample; | 
| 222 |  |  |  |  |  |  | $clone2->{'table'} = \@sample; | 
| 223 |  |  |  |  |  |  | while( scalar(@sample) < int( $size * $sample ) ) { | 
| 224 |  |  |  |  |  |  | my @shuffled = shuffle( @{ $clone1->{'table'} } ); | 
| 225 |  |  |  |  |  |  | push @sample, shift @shuffled; | 
| 226 |  |  |  |  |  |  | $clone1->{'table'} = \@shuffled; | 
| 227 |  |  |  |  |  |  | } | 
| 228 |  |  |  |  |  |  | return $clone2, $clone1; | 
| 229 |  |  |  |  |  |  | } | 
| 230 |  |  |  |  |  |  |  | 
| 231 |  |  |  |  |  |  | =item partition_data | 
| 232 |  |  |  |  |  |  |  | 
| 233 |  |  |  |  |  |  | Creates two clones that partition the data according to the provided ratio. | 
| 234 |  |  |  |  |  |  |  | 
| 235 |  |  |  |  |  |  | =cut | 
| 236 |  |  |  |  |  |  |  | 
| 237 |  |  |  |  |  |  | sub partition_data { | 
| 238 |  |  |  |  |  |  | my $self   = shift; | 
| 239 |  |  |  |  |  |  | my $sample = shift || 0.5; | 
| 240 |  |  |  |  |  |  | my $clone1 = $self->clone; | 
| 241 |  |  |  |  |  |  | my $clone2 = $self->clone; | 
| 242 |  |  |  |  |  |  | my $remain = 1 - $sample; | 
| 243 |  |  |  |  |  |  | $log->info("going to partition into $sample : $remain"); | 
| 244 |  |  |  |  |  |  |  | 
| 245 |  |  |  |  |  |  | # compute number of different dependent patterns and ratios of each | 
| 246 |  |  |  |  |  |  | my @dependents = $self->dependent_data; | 
| 247 |  |  |  |  |  |  | my %seen; | 
| 248 |  |  |  |  |  |  | for my $dep ( @dependents ) { | 
| 249 |  |  |  |  |  |  | my $key = join '/', @{ $dep }; | 
| 250 |  |  |  |  |  |  | $seen{$key}++; | 
| 251 |  |  |  |  |  |  | } | 
| 252 |  |  |  |  |  |  |  | 
| 253 |  |  |  |  |  |  | # adjust counts to sample size | 
| 254 |  |  |  |  |  |  | for my $key ( keys %seen ) { | 
| 255 |  |  |  |  |  |  | $log->debug("counts: $key => $seen{$key}"); | 
| 256 |  |  |  |  |  |  | $seen{$key} = int( $seen{$key} * $sample ); | 
| 257 |  |  |  |  |  |  | $log->debug("rescaled: $key => $seen{$key}"); | 
| 258 |  |  |  |  |  |  | } | 
| 259 |  |  |  |  |  |  |  | 
| 260 |  |  |  |  |  |  | # start the sampling | 
| 261 |  |  |  |  |  |  | my @dc = map { $self->{'header'}->{$_} } $self->dependent_columns; | 
| 262 |  |  |  |  |  |  | my @new_table; # we will populate this | 
| 263 |  |  |  |  |  |  | my @table = @{ $clone1->{'table'} }; # work on cloned instance | 
| 264 |  |  |  |  |  |  |  | 
| 265 |  |  |  |  |  |  | # as long as there is still sampling to do | 
| 266 |  |  |  |  |  |  | SAMPLE: while( grep { !!$_ } values %seen ) { | 
| 267 |  |  |  |  |  |  | for my $i ( 0 .. $#table ) { | 
| 268 |  |  |  |  |  |  | my @r = @{ $table[$i] }; | 
| 269 |  |  |  |  |  |  | my $key = join '/', @r[@dc]; | 
| 270 |  |  |  |  |  |  | if ( $seen{$key} ) { | 
| 271 |  |  |  |  |  |  | my $rand = rand(1); | 
| 272 |  |  |  |  |  |  | if ( $rand < $sample ) { | 
| 273 |  |  |  |  |  |  | push @new_table, \@r; | 
| 274 |  |  |  |  |  |  | splice @table, $i, 1; | 
| 275 |  |  |  |  |  |  | $seen{$key}--; | 
| 276 |  |  |  |  |  |  | $log->debug("still to go for $key: $seen{$key}"); | 
| 277 |  |  |  |  |  |  | next SAMPLE; | 
| 278 |  |  |  |  |  |  | } | 
| 279 |  |  |  |  |  |  | } | 
| 280 |  |  |  |  |  |  | } | 
| 281 |  |  |  |  |  |  | } | 
| 282 |  |  |  |  |  |  | $clone2->{'table'} = \@new_table; | 
| 283 |  |  |  |  |  |  | $clone1->{'table'} = \@table; | 
| 284 |  |  |  |  |  |  | return $clone2, $clone1; | 
| 285 |  |  |  |  |  |  | } | 
| 286 |  |  |  |  |  |  |  | 
| 287 |  |  |  |  |  |  | =item size | 
| 288 |  |  |  |  |  |  |  | 
| 289 |  |  |  |  |  |  | Returns the number of data records | 
| 290 |  |  |  |  |  |  |  | 
| 291 |  |  |  |  |  |  | =cut | 
| 292 |  |  |  |  |  |  |  | 
| 293 |  |  |  |  |  |  | sub size { scalar @{ shift->{'table'} } } | 
| 294 |  |  |  |  |  |  |  | 
| 295 |  |  |  |  |  |  | =item to_fann | 
| 296 |  |  |  |  |  |  |  | 
| 297 |  |  |  |  |  |  | Packs data into an L TrainData structure | 
| 298 |  |  |  |  |  |  |  | 
| 299 |  |  |  |  |  |  | =cut | 
| 300 |  |  |  |  |  |  |  | 
| 301 |  |  |  |  |  |  | sub to_fann { | 
| 302 |  |  |  |  |  |  | $log->debug("encoding data as FANN struct"); | 
| 303 |  |  |  |  |  |  | my $self = shift; | 
| 304 |  |  |  |  |  |  | my @cols = @_ ? @_ : $self->predictor_columns; | 
| 305 |  |  |  |  |  |  | my @deps = $self->dependent_data; | 
| 306 |  |  |  |  |  |  | my @pred = $self->predictor_data( 'cols' => \@cols ); | 
| 307 |  |  |  |  |  |  | my @interdigitated; | 
| 308 |  |  |  |  |  |  | for my $i ( 0 .. $#deps ) { | 
| 309 |  |  |  |  |  |  | push @interdigitated, $pred[$i], $deps[$i]; | 
| 310 |  |  |  |  |  |  | } | 
| 311 |  |  |  |  |  |  | return AI::FANN::TrainData->new(@interdigitated); | 
| 312 |  |  |  |  |  |  | } | 
| 313 |  |  |  |  |  |  |  | 
| 314 |  |  |  |  |  |  | =back | 
| 315 |  |  |  |  |  |  |  | 
| 316 |  |  |  |  |  |  | =cut | 
| 317 |  |  |  |  |  |  |  | 
| 318 |  |  |  |  |  |  | 1; |