/usr/lib/perl5/KinoSearch1/Search/Similarity.pm is in libkinosearch1-perl 1.00-1build3.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 | package KinoSearch1::Search::Similarity;
use strict;
use warnings;
use KinoSearch1::Util::ToolSet;
use base qw( KinoSearch1::Util::CClass );
BEGIN { __PACKAGE__->init_instance_vars(); }
# See _float_to_byte.
*encode_norm = *_float_to_byte;
*decode_norm = *_byte_to_float;
# Calculate the Inverse Document Frequecy for one or more Term in a given
# collection (the Searcher represents the collection).
#
# If multiple Terms are supplied, their idfs are summed.
sub idf {
my ( $self, $term_or_terms, $searcher ) = @_;
my $max_doc = $searcher->max_doc;
my $terms
= ref $term_or_terms eq 'ARRAY' ? $term_or_terms : [$term_or_terms];
return 1 unless $max_doc; # guard against log of zero error
# accumulate IDF
my $idf = 0;
for my $term (@$terms) {
my $doc_freq = $searcher->doc_freq($term);
$idf += 1 + log( $max_doc / ( 1 + $searcher->doc_freq($term) ) );
}
return $idf;
}
# Normalize a Query's weight so that it is comparable to other Queries.
sub query_norm {
my ( $self, $sum_of_squared_weights ) = @_;
return 0 if ( $sum_of_squared_weights == 0 ); # guard against div by zero
return ( 1 / sqrt($sum_of_squared_weights) );
}
# KLUDGE -- see comment at STORABLE_thaw.
sub STORABLE_freeze {
my ( $self, $cloning ) = @_;
return if $cloning;
return "1";
}
package KinoSearch1::Search::TitleSimilarity;
use strict;
use warnings;
use KinoSearch1::Util::ToolSet;
use base qw( KinoSearch1::Search::Similarity );
sub new {
my $self = shift->SUPER::new(@_);
$self->_use_title_tf;
return $self;
}
sub lengthnorm {
return 0 unless $_[1];
return 1 / sqrt( $_[1] );
}
1;
__END__
__XS__
MODULE = KinoSearch1 PACKAGE = KinoSearch1::Search::Similarity
=begin comment
KLUDGE!!
Rather than attempt to serialize a Similarity, we just create a new one.
=end comment
=cut
void
STORABLE_thaw(blank_obj, cloning, serialized)
SV *blank_obj;
SV *cloning;
SV *serialized;
PPCODE:
{
Similarity *sim = Kino1_Sim_new();
SV *deep_obj = SvRV(blank_obj);
sv_setiv(deep_obj, PTR2IV(sim));
}
void
new(either_sv)
SV *either_sv;
PREINIT:
const char *class;
Similarity *sim;
PPCODE:
/* determine the class */
class = sv_isobject(either_sv)
? sv_reftype(either_sv, 0)
: SvPV_nolen(either_sv);
/* build object */
sim = Kino1_Sim_new();
ST(0) = sv_newmortal();
sv_setref_pv(ST(0), class, (void*)sim);
XSRETURN(1);
=for comment
Provide a normalization factor for a field based on the square-root of the
number of terms in it.
=cut
float
lengthnorm(sim, num_terms)
Similarity *sim;
U32 num_terms;
CODE:
num_terms = num_terms < 100 ? 100 : num_terms;
RETVAL = (float)1 / sqrt(num_terms);
OUTPUT: RETVAL
=for comment
Return a score factor based on the frequency of a term in a given document.
The default implementation is sqrt(freq). Other implementations typically
produce ascending scores with ascending freqs, since the more times a doc
matches, the more relevant it is likely to be.
=cut
float
tf(sim, freq)
Similarity *sim;
U32 freq;
CODE:
RETVAL = sim->tf(sim, freq);
OUTPUT: RETVAL
=for comment
_float_to_byte and _byte_to_float encode and decode between 32-bit IEEE
floating point numbers and a 5-bit exponent, 3-bit mantissa float. The range
covered by the single-byte encoding is 7x10^9 to 2x10^-9. The accuracy is
about one significant decimal digit.
=cut
SV*
_float_to_byte(sim, f)
Similarity *sim;
float f;
PREINIT:
char b;
CODE:
b = Kino1_Sim_float2byte(sim, f);
RETVAL = newSVpv(&b, 1);
OUTPUT: RETVAL
float
_byte_to_float(sim, b)
Similarity *sim;
char b;
CODE:
RETVAL = Kino1_Sim_byte2float(sim, b);
OUTPUT: RETVAL
=for comment
The norm_decoder caches the 256 possible byte => float pairs, obviating the
need to call decode_norm over and over for a scoring implementation that
knows how to use it.
=cut
SV*
get_norm_decoder(sim)
Similarity *sim;
CODE:
RETVAL = newSVpv( (char*)sim->norm_decoder, (256 * sizeof(float)) );
OUTPUT: RETVAL
float
coord(sim, overlap, max_overlap)
Similarity *sim;
U32 overlap;
U32 max_overlap;
CODE:
RETVAL = sim->coord(sim, overlap, max_overlap);
OUTPUT: RETVAL
void
_use_title_tf(sim)
Similarity *sim;
PPCODE:
sim->tf = Kino1_Sim_title_tf;
void
DESTROY(sim)
Similarity *sim;
PPCODE:
Kino1_Sim_destroy(sim);
__H__
#ifndef H_KINO_SIMILARITY
#define H_KINO_SIMILARITY 1
#include "EXTERN.h"
#include "perl.h"
#include "XSUB.h"
#include "KinoSearch1UtilMemManager.h"
typedef struct similarity {
float (*tf)(struct similarity*, float);
float (*coord)(struct similarity*, U32, U32);
float *norm_decoder;
} Similarity;
Similarity* Kino1_Sim_new();
float Kino1_Sim_default_tf(Similarity*, float);
float Kino1_Sim_title_tf(Similarity*, float);
char Kino1_Sim_float2byte(Similarity*, float);
float Kino1_Sim_byte2float(Similarity*, char);
float Kino1_Sim_coord(Similarity*, U32, U32);
void Kino1_Sim_destroy(Similarity*);
#endif /* include guard */
__C__
#include "KinoSearch1SearchSimilarity.h"
Similarity*
Kino1_Sim_new() {
int i;
unsigned char aUChar;
Similarity *sim;
Kino1_New(0, sim, 1, Similarity);
/* cache decoded norms */
Kino1_New(0, sim->norm_decoder, 256, float);
for (i = 0; i < 256; i++) {
aUChar = i;
sim->norm_decoder[i] = Kino1_Sim_byte2float(sim, (char)aUChar);
}
sim->tf = Kino1_Sim_default_tf;
sim->coord = Kino1_Sim_coord;
return sim;
}
float
Kino1_Sim_default_tf(Similarity *sim, float freq) {
return( sqrt(freq) );
}
float
Kino1_Sim_title_tf(Similarity *sim, float freq) {
return 1.0;
}
char
Kino1_Sim_float2byte(Similarity *sim, float f) {
char norm;
I32 mantissa;
I32 exponent;
I32 bits;
if (f < 0.0)
f = 0.0;
if (f == 0.0) {
norm = 0;
}
else {
bits = *(I32*)&f;
mantissa = (bits & 0xffffff) >> 21;
exponent = (((bits >> 24) & 0x7f)-63) + 15;
if (exponent > 31) {
exponent = 31;
mantissa = 7;
}
if (exponent < 0) {
exponent = 0;
mantissa = 1;
}
norm = (char)((exponent << 3) | mantissa);
}
return norm;
}
float
Kino1_Sim_byte2float(Similarity *sim, char b) {
I32 mantissa;
I32 exponent;
I32 result;
if (b == 0) {
result = 0;
}
else {
mantissa = b & 7;
exponent = (b >> 3) & 31;
result = ((exponent+(63-15)) << 24) | (mantissa << 21);
}
return *(float*)&result;
}
/* Calculate a score factor based on the number of terms which match. */
float
Kino1_Sim_coord(Similarity *sim, U32 overlap, U32 max_overlap) {
if (max_overlap == 0)
return 1;
return (float)overlap / (float)max_overlap;
}
void
Kino1_Sim_destroy(Similarity *sim) {
Kino1_Safefree(sim->norm_decoder);
Kino1_Safefree(sim);
}
__POD__
=begin devdocs
=head1 NAME
KinoSearch1::Search::Similarity - calculate how closely two items match
=head1 DESCRIPTION
The Similarity class encapsulates some of the math used when calculating
scores.
TitleSimilarity is tuned for best results with title fields.
=head1 SEE ALSO
The Lucene equivalent of this class provides a thorough discussion of the
Lucene scoring algorithm, which KinoSearch1 implements.
=head1 COPYRIGHT
Copyright 2005-2010 Marvin Humphrey
=head1 LICENSE, DISCLAIMER, BUGS, etc.
See L<KinoSearch1> version 1.00.
=end devdocs
=cut
|