| 1 |
# -*- Mode: Perl -*- |
| 2 |
# $Basename: InvertedIndex.pm $ |
| 3 |
# $Revision: 1.30 $ |
| 4 |
# Author : Ulrich Pfeifer |
| 5 |
# Created On : Thu Aug 8 13:05:10 1996 |
| 6 |
# Last Modified By: Ulrich Pfeifer |
| 7 |
# Last Modified On: Sat Nov 11 16:32:38 2000 |
| 8 |
# Language : CPerl |
| 9 |
# |
| 10 |
# (C) Copyright 1996-2000, Ulrich Pfeifer |
| 11 |
# |
| 12 |
|
| 13 |
package WAIT::InvertedIndex; |
| 14 |
use strict; |
| 15 |
use DB_File; |
| 16 |
use Fcntl; |
| 17 |
use WAIT::Filter; |
| 18 |
use Carp; |
| 19 |
use vars qw(%FUNC); |
| 20 |
|
| 21 |
# The dictionary has three different key types: |
| 22 |
# 'o'.$word |
| 23 |
# |
| 24 |
# The document frequency is the number of documents a term occurs |
| 25 |
# in. The idea is that a term occuring in a significant part of the |
| 26 |
# documents is not too significant. |
| 27 |
# |
| 28 |
# 'm'.$word |
| 29 |
# |
| 30 |
# The maximum term frequency of a document is the frequency of the |
| 31 |
# most frequent term in the document. It is related to the document |
| 32 |
# length obviously. A document in which the most frequnet term occurs |
| 33 |
# 100 times is probably much longer than a document whichs most |
| 34 |
# frequent term occurs five time. |
| 35 |
# |
| 36 |
# 'p'.$word |
| 37 |
# |
| 38 |
# Under this key we store the actual posting list as pairs of |
| 39 |
# packed integers. |
| 40 |
|
| 41 |
my $no_old_index_support = 0; # do not check for old indices if set |
| 42 |
|
| 43 |
sub new { |
| 44 |
my $type = shift; |
| 45 |
my %parm = @_; |
| 46 |
my $self = {}; |
| 47 |
|
| 48 |
$self->{file} = $parm{file} or croak "No file specified"; |
| 49 |
$self->{attr} = $parm{attr} or croak "No attributes specified"; |
| 50 |
$self->{filter} = $parm{filter}; |
| 51 |
$self->{'name'} = $parm{'name'}; |
| 52 |
$self->{records} = 0; |
| 53 |
for (qw(intervall prefix)) { |
| 54 |
if (exists $parm{$_}) { |
| 55 |
if (ref $parm{$_}) { |
| 56 |
$self->{$_} = [@{$parm{$_}}] # clone |
| 57 |
} else { |
| 58 |
$self->{$_} = $parm{$_} |
| 59 |
} |
| 60 |
} |
| 61 |
} |
| 62 |
bless $self, ref($type) || $type; |
| 63 |
} |
| 64 |
|
| 65 |
sub name {$_[0]->{'name'}} |
| 66 |
|
| 67 |
sub _split_pos { |
| 68 |
my ($text, $pos) = @{$_[0]}; |
| 69 |
my @result; |
| 70 |
|
| 71 |
$text =~ s/(^\s+)// and $pos += length($1); |
| 72 |
while ($text =~ s/(^\S+)//) { |
| 73 |
my $word = $1; |
| 74 |
push @result, [$word, $pos]; |
| 75 |
$pos += length($word); |
| 76 |
$text =~ s/(^\s+)// and $pos += length($1); |
| 77 |
} |
| 78 |
@result; |
| 79 |
} |
| 80 |
|
| 81 |
sub _xfiltergen { |
| 82 |
my $filter = pop @_; |
| 83 |
|
| 84 |
# Oops, we cannot overrule the user's choice. Other filters may kill |
| 85 |
# stopwords, such as isotr clobbers "isn't" to "isnt". |
| 86 |
|
| 87 |
# if ($filter eq 'stop') { # avoid the slow stopword elimination |
| 88 |
# return _xfiltergen(@_); # it's cheaper to look them up afterwards |
| 89 |
# } |
| 90 |
if (@_) { |
| 91 |
if ($filter =~ /^split(\d*)/) { |
| 92 |
if ($1) { |
| 93 |
"grep(length(\$_->[0])>=$1, map(&WAIT::Filter::split_pos(\$_), " . _xfiltergen(@_) .'))' ; |
| 94 |
} else { |
| 95 |
"map(&WAIT::Filter::split_pos(\$_), " . _xfiltergen(@_) .')' ; |
| 96 |
} |
| 97 |
} else { |
| 98 |
"map ([&WAIT::Filter::$filter(\$_->[0]), \$_->[1]]," ._xfiltergen(@_) .')'; |
| 99 |
} |
| 100 |
} else { |
| 101 |
if ($filter =~ /^split(\d*)/) { |
| 102 |
if ($1) { |
| 103 |
"grep(length(\$_->[0])>=$1, map(&WAIT::Filter::split_pos(\$_), [\$_[0], 0]))" ; |
| 104 |
} else { |
| 105 |
"map(&WAIT::Filter::split_pos(\$_), [\$_[0], 0])" ; |
| 106 |
} |
| 107 |
} else { |
| 108 |
"map ([&WAIT::Filter::$filter(\$_->[0]), \$_->[1]], [\$_[0], 0])"; |
| 109 |
} |
| 110 |
} |
| 111 |
} |
| 112 |
|
| 113 |
sub parse_pos { |
| 114 |
my $self = shift; |
| 115 |
|
| 116 |
unless (exists $self->{xfunc}) { |
| 117 |
$self->{xfunc} = |
| 118 |
eval sprintf("sub {%s}", _xfiltergen(@{$self->{filter}})); |
| 119 |
#printf "\nsub{%s}$@\n", _xfiltergen(@{$self->{filter}}); |
| 120 |
} |
| 121 |
&{$self->{xfunc}}($_[0]); |
| 122 |
} |
| 123 |
|
| 124 |
sub _filtergen { |
| 125 |
my $filter = pop @_; |
| 126 |
|
| 127 |
if (@_) { |
| 128 |
"map(&WAIT::Filter::$filter(\$_), " . _filtergen(@_) . ')'; |
| 129 |
} else { |
| 130 |
"map(&WAIT::Filter::$filter(\$_), \@_)"; |
| 131 |
} |
| 132 |
} |
| 133 |
|
| 134 |
sub drop { |
| 135 |
my $self = shift; |
| 136 |
if ((caller)[0] eq 'WAIT::Table') { # Table knows about this |
| 137 |
my $file = $self->{file}; |
| 138 |
|
| 139 |
! (!-e $file or unlink $file); |
| 140 |
} else { # notify our database |
| 141 |
croak ref($self)."::drop called directly"; |
| 142 |
} |
| 143 |
} |
| 144 |
|
| 145 |
sub is_an_old_index { |
| 146 |
my $self = shift; |
| 147 |
|
| 148 |
return 0 if $no_old_index_support; |
| 149 |
return $self->{old_index} if exists $self->{old_index}; |
| 150 |
|
| 151 |
# We can only guess if this is an old index. We lookup the first 10 |
| 152 |
# $O entries. If all values are integers, we assume that the index |
| 153 |
# is an old one. |
| 154 |
|
| 155 |
defined $self->{db} or $self->open; |
| 156 |
$self->sync; |
| 157 |
my $dbh = $self->{dbh}; # for convenience |
| 158 |
|
| 159 |
my $O = pack('C', 0xff)."o"; |
| 160 |
my ($word, $value) = ($O.$;); |
| 161 |
$dbh->seq($word, $value, R_CURSOR); |
| 162 |
for (my $i=0; $i<10;$i++) { |
| 163 |
if ($value !~ /^\d+$/) { |
| 164 |
return $self->{old_index} = 0; |
| 165 |
} |
| 166 |
if ($dbh->seq($word, $value, R_NEXT) or # no values left |
| 167 |
$word !~ /^$O/o # no $O values left |
| 168 |
) { |
| 169 |
# we are not sure enough that this is an old index |
| 170 |
return $self->{old_index} = 0; |
| 171 |
} |
| 172 |
} |
| 173 |
return $self->{old_index} = 1; |
| 174 |
} |
| 175 |
|
| 176 |
sub open { |
| 177 |
my $self = shift; |
| 178 |
my $file = $self->{file}; |
| 179 |
|
| 180 |
if (defined $self->{dbh}) { |
| 181 |
$self->{dbh}; |
| 182 |
} else { |
| 183 |
$self->{func} = |
| 184 |
eval sprintf("sub {grep /./, %s}", _filtergen(@{$self->{filter}})); |
| 185 |
$self->{dbh} = tie(%{$self->{db}}, 'DB_File', $file, |
| 186 |
$self->{mode}, 0664, $DB_BTREE); |
| 187 |
$self->{cache} = {} |
| 188 |
if $self->{mode} & O_RDWR; |
| 189 |
$self->{cdict} = {} |
| 190 |
if $self->{mode} & O_RDWR; |
| 191 |
$self->{cached} = 0; |
| 192 |
if (!$no_old_index_support and $self->is_an_old_index()) { |
| 193 |
warn "This is an old index, upgrade you database"; |
| 194 |
require WAIT::InvertedIndexOld; |
| 195 |
bless $self, 'WAIT::InvertedIndexOld'; |
| 196 |
} |
| 197 |
} |
| 198 |
} |
| 199 |
|
| 200 |
sub insert { |
| 201 |
my $self = shift; |
| 202 |
my $key = shift; |
| 203 |
my %occ; |
| 204 |
|
| 205 |
defined $self->{db} or $self->open; |
| 206 |
grep $occ{$_}++, &{$self->{func}}(@_); |
| 207 |
my ($word, $noc); |
| 208 |
$self->{records}++; |
| 209 |
while (($word, $noc) = each %occ) { |
| 210 |
if (defined $self->{cache}->{$word}) { |
| 211 |
$self->{cdict}->{$word}++; |
| 212 |
$self->{cache}->{$word} .= pack 'w2', $key, $noc; |
| 213 |
} else { |
| 214 |
$self->{cdict}->{$word} = 1; |
| 215 |
$self->{cache}->{$word} = pack 'w2', $key, $noc; |
| 216 |
} |
| 217 |
$self->{cached}++; |
| 218 |
} |
| 219 |
# This cache limit should be configurable |
| 220 |
$self->sync if $self->{cached} > 100_000; |
| 221 |
my $maxtf = 0; |
| 222 |
for (values %occ) { |
| 223 |
$maxtf = $_ if $_ > $maxtf; |
| 224 |
} |
| 225 |
$self->{db}->{'m'. $key} = $maxtf; |
| 226 |
} |
| 227 |
|
| 228 |
# We sort postings by increasing max term frequency (~ by increasing |
| 229 |
# document length. This reduces the quality degradation if we process |
| 230 |
# only the first part of a posting list. |
| 231 |
|
| 232 |
sub sort_postings { |
| 233 |
my $self = shift; |
| 234 |
my $post = shift; # reference to a hash or packed string |
| 235 |
|
| 236 |
if (ref $post) { |
| 237 |
# we skip the sort part, if the index is not sorted |
| 238 |
return pack('w*', %$post) unless $self->{reorg}; |
| 239 |
} else { |
| 240 |
$post = { unpack 'w*', $post }; |
| 241 |
} |
| 242 |
|
| 243 |
my $r = ''; |
| 244 |
|
| 245 |
# Sort posting list by increasing ratio of maximum term frequency (~ |
| 246 |
# "document length") and term frequency. This rati multipied by the |
| 247 |
# inverse document frequence gives the score for a term. This sort |
| 248 |
# order can be exploited for tuning of single term queries. |
| 249 |
|
| 250 |
for my $did (sort { $post->{$b} / $self->{db}->{'m'. $b} |
| 251 |
<=> |
| 252 |
$post->{$a} / $self->{db}->{'m'. $a} |
| 253 |
} keys %$post) { |
| 254 |
$r .= pack 'w2', $did, $post->{$did}; |
| 255 |
} |
| 256 |
#warn sprintf "reorg %d %s\n", scalar keys %$post, join ' ', unpack 'w*', $r; |
| 257 |
$r; |
| 258 |
} |
| 259 |
|
| 260 |
sub delete { |
| 261 |
my $self = shift; |
| 262 |
my $key = shift; |
| 263 |
my %occ; |
| 264 |
|
| 265 |
my $db; |
| 266 |
defined $self->{db} or $self->open; |
| 267 |
$db = $self->{db}; |
| 268 |
$self->sync; |
| 269 |
$self->{records}--; |
| 270 |
|
| 271 |
# less than zero documents in database? |
| 272 |
_complain('delete of document', $key) and $self->{records} = 0 |
| 273 |
if $self->{records} < 0; |
| 274 |
|
| 275 |
grep $occ{$_}++, &{$self->{func}}(@_); |
| 276 |
|
| 277 |
for (keys %occ) {# may reorder posting list |
| 278 |
my %post = unpack 'w*', $db->{'p'.$_}; |
| 279 |
delete $post{$key}; |
| 280 |
$db->{'p'.$_} = $self->sort_postings(\%post); |
| 281 |
_complain('delete of term', $_) if $db->{'o'.$_}-1 != keys %post; |
| 282 |
$db->{'o'.$_} = scalar keys %post; |
| 283 |
} |
| 284 |
delete $db->{'m'. $key}; |
| 285 |
} |
| 286 |
|
| 287 |
sub intervall { |
| 288 |
my ($self, $first, $last) = @_; |
| 289 |
my $value = ''; |
| 290 |
my $word = ''; |
| 291 |
my @result; |
| 292 |
|
| 293 |
return unless exists $self->{'intervall'}; |
| 294 |
|
| 295 |
defined $self->{db} or $self->open; |
| 296 |
$self->sync; |
| 297 |
my $dbh = $self->{dbh}; # for convenience |
| 298 |
|
| 299 |
if (ref $self->{'intervall'}) { |
| 300 |
unless (exists $self->{'ifunc'}) { |
| 301 |
$self->{'ifunc'} = |
| 302 |
eval sprintf("sub {grep /./, %s}", _filtergen(@{$self->{intervall}})); |
| 303 |
} |
| 304 |
($first) = &{$self->{'ifunc'}}($first) if $first; |
| 305 |
($last) = &{$self->{'ifunc'}}($last) if $last; |
| 306 |
} |
| 307 |
$first = 'p'.($first||''); |
| 308 |
$last = (defined $last)?'p'.$last:'q'; |
| 309 |
|
| 310 |
# set the cursor to $first |
| 311 |
$dbh->seq($first, $value, R_CURSOR); |
| 312 |
|
| 313 |
# $first would be after the last word |
| 314 |
return () if $first gt $last; |
| 315 |
|
| 316 |
push @result, substr($first,1); |
| 317 |
while (!$dbh->seq($word, $value, R_NEXT)) { |
| 318 |
# We should limit this to a "resonable" number of words |
| 319 |
last if $word gt $last; |
| 320 |
push @result, substr($word,1); |
| 321 |
} |
| 322 |
\@result; # speed |
| 323 |
} |
| 324 |
|
| 325 |
sub prefix { |
| 326 |
my ($self, $prefix) = @_; |
| 327 |
my $value = ''; |
| 328 |
my $word = ''; |
| 329 |
my @result; |
| 330 |
|
| 331 |
return () unless defined $prefix; # Full dictionary requested !! |
| 332 |
return unless exists $self->{'prefix'}; |
| 333 |
defined $self->{db} or $self->open; |
| 334 |
$self->sync; |
| 335 |
my $dbh = $self->{dbh}; |
| 336 |
|
| 337 |
if (ref $self->{'prefix'}) { |
| 338 |
unless (exists $self->{'pfunc'}) { |
| 339 |
$self->{'pfunc'} = |
| 340 |
eval sprintf("sub {grep /./, %s}", _filtergen(@{$self->{prefix}})); |
| 341 |
} |
| 342 |
($prefix) = &{$self->{'pfunc'}}($prefix); |
| 343 |
} |
| 344 |
|
| 345 |
if ($dbh->seq($word = 'p'.$prefix, $value, R_CURSOR)) { |
| 346 |
return (); |
| 347 |
} |
| 348 |
return () if $word !~ /^p$prefix/; |
| 349 |
push @result, substr($word,1); |
| 350 |
|
| 351 |
while (!$dbh->seq($word, $value, R_NEXT)) { |
| 352 |
# We should limit this to a "resonable" number of words |
| 353 |
last if $word !~ /^p$prefix/; |
| 354 |
push @result, substr($word,1); |
| 355 |
} |
| 356 |
\@result; # speed |
| 357 |
} |
| 358 |
|
| 359 |
=head2 search($query) |
| 360 |
|
| 361 |
The search method supports a range of search algorithms. It is |
| 362 |
recommended to tune the index by calling |
| 363 |
C<$table-E<gt>set(top=E<gt>1)> B<after> bulk inserting the documents |
| 364 |
into the table. This is a computing intense operation and all inserts |
| 365 |
and deletes after this optimization are slightly more expensive. Once |
| 366 |
reorganized, the index is kept sorted automatically until you switch |
| 367 |
the optimization off by calling C<$table-E<gt>set(top=E<gt>0)>. |
| 368 |
|
| 369 |
When searching a tuned index, a query can be processed faster if the |
| 370 |
caller requests only the topmost documents. This can be done by |
| 371 |
passing a C<top =E<gt>> I<n> parameter to the search method. |
| 372 |
|
| 373 |
For single term queries, the method returns only the I<n> top ranking |
| 374 |
documents. For multi term queries two optimized algorithms are |
| 375 |
available. The first algorithm computes the top n documents |
| 376 |
approximately but very fast, sacrificing a little bit of precision for |
| 377 |
speed. The second algorithm computes the topmost I<n> documents |
| 378 |
precisely. This algorithm is slower and should be used only for small |
| 379 |
values of I<n>. It can be requested by passing the query attribute |
| 380 |
C<picky =E<gt> 1>. Both algorithms may return more than I<n> hits. |
| 381 |
While the picky version might not be faster than the brute force |
| 382 |
version on average for modest size databases it uses less memory and |
| 383 |
the processing time is almost linear in the number of query terms, not |
| 384 |
in the size of the lists. |
| 385 |
|
| 386 |
=cut |
| 387 |
|
| 388 |
sub search { |
| 389 |
my $self = shift; |
| 390 |
my $query = shift; |
| 391 |
|
| 392 |
defined $self->{db} or $self->open; |
| 393 |
$self->sync; |
| 394 |
$self->search_raw($query, &{$self->{func}}(@_)); # No call to parse() here |
| 395 |
} |
| 396 |
|
| 397 |
sub parse { |
| 398 |
my $self = shift; |
| 399 |
|
| 400 |
defined $self->{db} or $self->open; |
| 401 |
&{$self->{func}}(@_); |
| 402 |
} |
| 403 |
|
| 404 |
sub keys { |
| 405 |
my $self = shift; |
| 406 |
|
| 407 |
defined $self->{db} or $self->open; |
| 408 |
keys %{$self->{db}}; |
| 409 |
} |
| 410 |
|
| 411 |
sub search_prefix { |
| 412 |
my $self = shift; |
| 413 |
|
| 414 |
# print "search_prefix(@_)\n"; |
| 415 |
defined $self->{db} or $self->open; |
| 416 |
$self->search_raw(map($self->prefix($_), @_)); |
| 417 |
} |
| 418 |
|
| 419 |
sub _complain ($$) { |
| 420 |
my ($action, $term) = @_; |
| 421 |
|
| 422 |
require Carp; |
| 423 |
Carp::cluck |
| 424 |
(sprintf("WAIT database inconsistency during $action [%s]: ". |
| 425 |
"Please rebuild index\n", |
| 426 |
$term,)); |
| 427 |
} |
| 428 |
|
| 429 |
sub search_raw { |
| 430 |
my $self = shift; |
| 431 |
my $query = shift; |
| 432 |
my %score; |
| 433 |
|
| 434 |
# Top $wanted documents must be correct. Zero means all matching |
| 435 |
# documents. |
| 436 |
my $wanted = $query->{top}; |
| 437 |
my $strict = $query->{picky}; |
| 438 |
|
| 439 |
# Return at least $minacc documents. Zero means all matching |
| 440 |
# documents. |
| 441 |
# my $minacc = $query->{accus} || $wanted; |
| 442 |
|
| 443 |
# Open index and flush cache if necessary |
| 444 |
defined $self->{db} or $self->open; |
| 445 |
$self->sync; |
| 446 |
|
| 447 |
# We keep duplicates |
| 448 |
my @terms = |
| 449 |
# Sort words by decreasing document frequency |
| 450 |
sort { $self->{db}->{'o'.$a} <=> $self->{db}->{'o'.$b} } |
| 451 |
# check which words occur in the index. |
| 452 |
grep { $self->{db}->{'o'.$_} } @_; |
| 453 |
|
| 454 |
return () unless @terms; # nothing to search for |
| 455 |
|
| 456 |
# We special-case one term queries here. If the index was sorted, |
| 457 |
# choping off the rest of the list will return the same ranking. |
| 458 |
if ($wanted and @terms == 1) { |
| 459 |
my $term = shift @terms; |
| 460 |
my $idf = log($self->{records}/$self->{db}->{'o'.$term}); |
| 461 |
my @res; |
| 462 |
|
| 463 |
if ($self->{reorg}) { # or not $query->{picky} |
| 464 |
@res = unpack "w". int(2*$wanted), $self->{db}->{'p'.$term}; |
| 465 |
} else { |
| 466 |
@res = unpack 'w*', $self->{db}->{'p'.$term}; |
| 467 |
} |
| 468 |
|
| 469 |
for (my $i=1; $i<@res; $i+=2) { |
| 470 |
$res[$i] /= $self->{db}->{'m'. $res[$i-1]} / $idf; |
| 471 |
} |
| 472 |
|
| 473 |
return @res |
| 474 |
} |
| 475 |
|
| 476 |
# We separate exhaustive search here to avoid overhead and make the |
| 477 |
# code more readable. The block can be removed without changing the |
| 478 |
# result. |
| 479 |
unless ($wanted) { |
| 480 |
for (@terms) { |
| 481 |
my $df = $self->{db}->{'o'.$_}; |
| 482 |
|
| 483 |
# The frequency *must* be 1 at least since the posting list is nonempty |
| 484 |
_complain('search for term', $_) and $df = 1 if $df < 1; |
| 485 |
|
| 486 |
# Unpack posting list for current query term $_ |
| 487 |
my %post = unpack 'w*', $self->{db}->{'p'.$_}; |
| 488 |
|
| 489 |
_complain('search for term', $_) if $self->{db}->{'o'.$_} != keys %post; |
| 490 |
# This is the inverse document frequency. The log of the inverse |
| 491 |
# fraction of documents the term occurs in. |
| 492 |
my $idf = log($self->{records}/$df); |
| 493 |
for my $did (keys %post) { |
| 494 |
if (my $freq = $self->{db}->{'m'. $did}) { |
| 495 |
$score{$did} += $post{$did} / $freq * $idf; |
| 496 |
} |
| 497 |
} |
| 498 |
} |
| 499 |
# warn sprintf "Used %d accumulators\n", scalar keys %score; |
| 500 |
return %score; |
| 501 |
} |
| 502 |
|
| 503 |
# A sloppy but fast algorithm for multiple term queries. |
| 504 |
unless ($strict) { |
| 505 |
for (@terms) { |
| 506 |
# Unpack posting list for current query term $_ |
| 507 |
my %post = unpack 'w*', $self->{db}->{'p'.$_}; |
| 508 |
|
| 509 |
# Lookup the number of documents the term occurs in (document frequency) |
| 510 |
my $occ = $self->{db}->{'o'.$_}; |
| 511 |
|
| 512 |
_complain('search for term', $_) if $self->{db}->{'o'.$_} != keys %post; |
| 513 |
# The frequency *must* be 1 at least since the posting list is nonempty |
| 514 |
_complain('search for term', $_) and $occ = 1 if $occ < 1; |
| 515 |
|
| 516 |
# This is the inverse document frequency. The log of the inverse |
| 517 |
# fraction of documents the term occurs in. |
| 518 |
my $idf = log($self->{records}/$occ); |
| 519 |
|
| 520 |
# If we have a reasonable number of accumulators, change the |
| 521 |
# loop to iterate over the accumulators. This will compromise |
| 522 |
# quality for better speed. The algorithm still computes the |
| 523 |
# exact weights, but the result is not guaranteed to contain the |
| 524 |
# *best* results. The database might contain documents better |
| 525 |
# than the worst returned document. |
| 526 |
|
| 527 |
# We process the lists in order of increasing length. When the |
| 528 |
# number of accumulators exceeds $wanted, no new documents are |
| 529 |
# added, only the ranking/weighting of the seen documents is |
| 530 |
# improved. The resulting ranking list must be pruned, since only |
| 531 |
# the top most documents end up near their "optimal" rank. |
| 532 |
|
| 533 |
if (keys %score < $wanted) { |
| 534 |
for my $did (keys %post) { |
| 535 |
if (my $freq = $self->{db}->{'m'. $did}) { |
| 536 |
$score{$did} += $post{$did} / $freq * $idf; |
| 537 |
} |
| 538 |
} |
| 539 |
} else { |
| 540 |
for my $did (keys %score) { |
| 541 |
next unless exists $post{$did}; |
| 542 |
if (my $freq = $self->{db}->{'m'. $did}) { |
| 543 |
$score{$did} += $post{$did} / $freq * $idf; |
| 544 |
} |
| 545 |
} |
| 546 |
} |
| 547 |
} |
| 548 |
return %score; |
| 549 |
} |
| 550 |
my @max; $max[$#terms+1]=0; |
| 551 |
my @idf; |
| 552 |
|
| 553 |
# Preparation loop. This extra loop makes sense only when "reorg" |
| 554 |
# and "wanted" are true. But at the time beeing, keeping the code |
| 555 |
# for the different search algorithms in one place seems more |
| 556 |
# desirable than some minor speedup of the brute force version. We |
| 557 |
# do cache $idf though. |
| 558 |
|
| 559 |
for (my $i = $#terms; $i >=0; $i--) { |
| 560 |
local $_ = $terms[$i]; |
| 561 |
# Lookup the number of documents the term occurs in (document frequency) |
| 562 |
my $df = $self->{db}->{'o'.$_}; |
| 563 |
|
| 564 |
# The frequency *must* be 1 at least since the posting list is nonempty |
| 565 |
_complain('search for term', $_) and $df = 1 if $df < 1; |
| 566 |
|
| 567 |
# This is the inverse document frequency. The log of the inverse |
| 568 |
# fraction of documents the term occurs in. |
| 569 |
$idf[$i] = log($self->{records}/$df); |
| 570 |
|
| 571 |
my ($did,$occ); |
| 572 |
if ($self->{reorg}) { |
| 573 |
($did,$occ) = unpack 'w2', $self->{db}->{'p'.$_}; |
| 574 |
} else { # Maybe this costs more than it helps |
| 575 |
($did,$occ) = unpack 'w2', $self->sort_postings($self->{db}->{'p'.$_}); |
| 576 |
} |
| 577 |
my $freq = $self->{db}->{'m'. $did}; |
| 578 |
my $max = $occ/$freq*$idf[$i]; |
| 579 |
$max[$i] = $max + $max[$i+1]; |
| 580 |
} |
| 581 |
|
| 582 |
# Main loop |
| 583 |
for my $i (0 .. $#terms) { |
| 584 |
my $term = $terms[$i]; |
| 585 |
# Unpack posting list for current query term $term. We loose the |
| 586 |
# sorting order because the assignment to a hash. |
| 587 |
my %post = unpack 'w*', $self->{db}->{'p'.$term}; |
| 588 |
|
| 589 |
_complain('search for term', $term) |
| 590 |
if $self->{db}->{'o'.$term} != keys %post; |
| 591 |
|
| 592 |
my $idf = $idf[$i]; |
| 593 |
my $full; # Need to process all postings |
| 594 |
my $chop; # Score necessary to enter the ranking list |
| 595 |
|
| 596 |
if (# We know that wanted is true since we especial cased the |
| 597 |
# exhaustive search. |
| 598 |
|
| 599 |
$wanted and |
| 600 |
|
| 601 |
# We did sort here if necessary in |
| 602 |
# the preparation loop |
| 603 |
# $self->{reorg} and |
| 604 |
|
| 605 |
scalar keys %score > $wanted) { |
| 606 |
$chop = (sort { $b <=> $a } values %score)[$wanted]; |
| 607 |
$full = $max[$i] > $chop; |
| 608 |
} else { |
| 609 |
$full = 1; |
| 610 |
} |
| 611 |
|
| 612 |
if ($full) { |
| 613 |
# We need to inspect the full list. Either $wanted is not given, |
| 614 |
# the index is not sorted, or we don't have enough accumulators |
| 615 |
# yet. |
| 616 |
if (defined $chop) { |
| 617 |
# We might be able to avoid allocating accumulators |
| 618 |
for my $did (keys %post) { |
| 619 |
if (my $freq = $self->{db}->{'m'. $did}) { |
| 620 |
my $wgt = $post{$did} / $freq * $idf; |
| 621 |
# We add an accumulator if $wgt exeeds $chop |
| 622 |
if (exists $score{$did} or $wgt > $chop) { |
| 623 |
$score{$did} += $wgt; |
| 624 |
} |
| 625 |
} |
| 626 |
} |
| 627 |
} else { |
| 628 |
# Allocate acumulators for each seen document. |
| 629 |
for my $did (keys %post) { |
| 630 |
if (my $freq = $self->{db}->{'m'. $did}) { |
| 631 |
$score{$did} += $post{$did} / $freq * $idf; |
| 632 |
} |
| 633 |
} |
| 634 |
} |
| 635 |
} else { |
| 636 |
# Update existing accumulators |
| 637 |
for my $did (keys %score) { |
| 638 |
next unless exists $post{$did}; |
| 639 |
if (my $freq = $self->{db}->{'m'. $did}) { |
| 640 |
$score{$did} += $post{$did} / $freq * $idf; |
| 641 |
} |
| 642 |
} |
| 643 |
} |
| 644 |
} |
| 645 |
#warn sprintf "Used %d accumulators\n", scalar keys %score; |
| 646 |
%score; |
| 647 |
} |
| 648 |
|
| 649 |
sub set { |
| 650 |
my ($self, $attr, $value) = @_; |
| 651 |
|
| 652 |
die "No such indexy attribute: '$attr'" unless $attr eq 'top'; |
| 653 |
|
| 654 |
return delete $self->{reorg} if $value == 0; |
| 655 |
|
| 656 |
return if $self->{reorg}; # we are sorted already |
| 657 |
return unless $self->{mode} & O_RDWR; |
| 658 |
defined $self->{db} or $self->open; |
| 659 |
|
| 660 |
$self->sync; |
| 661 |
while (my($key, $value) = each %{$self->{db}}) { |
| 662 |
next if $key !~ /^p/; |
| 663 |
$self->{db}->{$key} = $self->sort_postings($value); |
| 664 |
} |
| 665 |
$self->{reorg} = 1; |
| 666 |
} |
| 667 |
|
| 668 |
sub sync { |
| 669 |
my $self = shift; |
| 670 |
|
| 671 |
if ($self->{mode} & O_RDWR) { |
| 672 |
print STDERR "Flushing $self->{cached} postings\n" if $self->{cached}; |
| 673 |
while (my($key, $value) = each %{$self->{cache}}) { |
| 674 |
if ($self->{reorg}) { |
| 675 |
$self->{db}->{'p'.$key} = $self->sort_postings($self->{db}->{'p'.$key} |
| 676 |
. $value); |
| 677 |
} else { |
| 678 |
$self->{db}->{'p'.$key} .= $value; |
| 679 |
} |
| 680 |
} |
| 681 |
while (my($key, $value) = each %{$self->{cdict}}) { |
| 682 |
$self->{db}->{'o'.$key} = 0 unless $self->{db}->{'o'.$key}; |
| 683 |
$self->{db}->{'o'.$key} += $value; |
| 684 |
} |
| 685 |
$self->{cache} = {}; |
| 686 |
$self->{cdict} = {}; |
| 687 |
$self->{cached} = 0; |
| 688 |
} |
| 689 |
} |
| 690 |
|
| 691 |
sub close { |
| 692 |
my $self = shift; |
| 693 |
|
| 694 |
if ($self->{dbh}) { |
| 695 |
$self->sync; |
| 696 |
delete $self->{dbh}; |
| 697 |
untie %{$self->{db}}; |
| 698 |
delete $self->{db}; |
| 699 |
delete $self->{func}; |
| 700 |
delete $self->{cache}; |
| 701 |
delete $self->{cached}; |
| 702 |
delete $self->{cdict}; |
| 703 |
delete $self->{pfunc} if defined $self->{pfunc}; |
| 704 |
delete $self->{ifunc} if defined $self->{ifunc}; |
| 705 |
delete $self->{xfunc} if defined $self->{xfunc}; |
| 706 |
} |
| 707 |
} |
| 708 |
|
| 709 |
1; |
| 710 |
|
Parent Directory
| 
Revision Log