xapian-core/weight/dlhweight.cc

   1 /** @file
   2  * @brief Xapian::DLHWeight class - The DLH weighting scheme of the DFR framework.
   3  */
   4 /* Copyright (C) 2013, 2014 Aarsh Shah
   5  * Copyright (C) 2016,2017,2019 Olly Betts
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License as
   9  * published by the Free Software Foundation; either version 2 of the
  10  * License, or (at your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 #include <config.h>
  23
  24 #include "xapian/weight.h"
  25
  26 #include "xapian/error.h"
  27 #include "common/log2.h"
  28 #include <algorithm>
  29
  30 using namespace std;
  31
  32 namespace Xapian {
  33
  34 DLHWeight *
  35 DLHWeight::clone() const
  36 {
  37     return new DLHWeight();
  38 }
  39
  40 void
  41 DLHWeight::init(double factor)
  42 {
  43     // Avoid warnings about unused private member.
  44     (void)lower_bound;
  45
  46     if (factor == 0.0) {
  47         // This object is for the term-independent contribution, and that's
  48         // always zero for this scheme.
  49         return;
  50     }
  51
  52     double wdf_upper = get_wdf_upper_bound();
  53     if (wdf_upper == 0) {
  54         upper_bound = 0.0;
  55         return;
  56     }
  57
  58     const double wdf_lower = 1.0;
  59     double len_upper = get_doclength_upper_bound();
  60     double len_lower = get_doclength_lower_bound();
  61
  62     double F = get_collection_freq();
  63
  64     // Calculate constant values to be used in get_sumpart().
  65     log_constant = get_total_length() / F;
  66     wqf_product_factor = get_wqf() * factor;
  67
  68     // Calculate values for the upper bound.
  69
  70     // w <= l, so if the allowed ranges overlap, max w/l is 1.0.
  71     double max_wdf_over_l = wdf_upper < len_lower ? wdf_upper / len_lower : 1.0;
  72
  73     // First term A: w/(w+.5)*log2(w/l*L) where L=total_len/coll_freq
  74     // Assume log >= 0:
  75     //   w/(w+.5) = 1-1/(2w+1) and w >= 1 so max A at w=w_max
  76     //   log2(w/l*L) maximised when w/l maximised
  77     //   so max A at w=w_max, l=max(l_min, w_max)
  78     // If log < 0 => then A <= 0, so max A <= 0 and we want to minimise the
  79     // factor outside the log.
  80     double logged_expr = max_wdf_over_l * log_constant;
  81     double w_for_A = logged_expr > 1.0 ? wdf_upper : wdf_lower;
  82     double A = w_for_A / (w_for_A + 0.5) * log2(logged_expr);
  83
  84     // Second term B:
  85     //
  86     // (l-w)*log2(1-w/l)
  87     //
  88     // The log is negative, and w <= l so B <= 0 and its maximum is the value
  89     // as close to zero as possible.  So smaller (l-w) is better and smaller
  90     // w/l is better.
  91     //
  92     // This function is ill defined at w=l, but -> 0 as w -> l.
  93     //
  94     // If w=l is valid (i.e. len_lower > wdf_upper) then B = 0.
  95     double B = 0;
  96     if (len_lower > wdf_upper) {
  97         // If not, then minimising l-w gives us a candidate (i.e. w=wdf_upper
  98         // and l=len_lower).
  99         //
 100         // The function is also 0 at w = 0 (there must be a local mimina at
 101         // some value of w between 0 and l), so the other candidate is at
 102         // w=wdf_lower.
 103         //
 104         // We need to find the optimum value of l in this case, so
 105         // differentiate the formula by l:
 106         //
 107         // d/dl: log2(1-w/l) + (l-w)*(1-w/l)/(l*log(2))
 108         //     = (log(1-w/l) + (1-w/l)²)/log(2)
 109         //     = (log(x) + x²)/log(2)     [x=1-w/l]
 110         //
 111         // which is 0 at approx x=0.65291864
 112         //
 113         // x=1-w/l <=> l*(1-x)=w <=> l=w/(1-x) <=> l ~= 0.34708136*w
 114         //
 115         // but l >= w so we can't attain that (and the log isn't valid there).
 116         //
 117         // Gradient at (without loss of generality) l=2*w is:
 118         //       (log(0.5) + 0.25)/log(2)
 119         // which is < 0 so want to minimise l, i.e. l=len_lower, so the other
 120         // candidate is w=wdf_lower and l=len_lower.
 121         //
 122         // So evaluate both candidates and pick the larger:
 123         double B1 = (len_lower - wdf_lower) * log2(1.0 - wdf_lower / len_lower);
 124         double B2 = (len_lower - wdf_upper) * log2(1.0 - wdf_upper / len_lower);
 125         B = max(B1, B2);
 126     }
 127
 128     /* An upper bound of the term used in the third log can be obtained by:
 129      *
 130      * 0.5 * log2(2.0 * M_PI * wdf * (1 - wdf / len))
 131      *
 132      * An upper bound on wdf * (1 - wdf / len) (and hence on the log, since
 133      * log is a monotonically increasing function on positive numbers) can
 134      * be obtained by plugging in the upper bound of the length and
 135      * differentiating the term w.r.t wdf which gives the value of wdf at which
 136      * the function attains maximum value - at wdf = len_upper / 2 (or if the
 137      * wdf can't be that large, at wdf = wdf_upper): */
 138     double wdf_var = min(wdf_upper, len_upper / 2.0);
 139     double max_product = wdf_var * (1.0 - wdf_var / len_upper);
 140 #if 0
 141     /* An upper bound can also be obtained by taking the minimum and maximum
 142      * wdf value in the formula as shown (which isn't useful now as it's always
 143      * >= the bound above, but could be useful if we tracked bounds on wdf/len):
 144      */
 145     double min_wdf_to_len = wdf_lower / len_upper;
 146     double max_product_2 = wdf_upper * (1.0 - min_wdf_to_len);
 147     /* Take the minimum of the two upper bounds. */
 148     max_product = min(max_product, max_product_2);
 149 #endif
 150     double C = 0.5 * log2(2.0 * M_PI * max_product) / (wdf_lower + 0.5);
 151     upper_bound = A + B + C;
 152
 153     if (rare(upper_bound < 0.0))
 154         upper_bound = 0.0;
 155     else
 156         upper_bound *= wqf_product_factor;
 157 }
 158
 159 string
 160 DLHWeight::name() const
 161 {
 162     return "Xapian::DLHWeight";
 163 }
 164
 165 string
 166 DLHWeight::serialise() const
 167 {
 168     return string();
 169 }
 170
 171 DLHWeight *
 172 DLHWeight::unserialise(const string& s) const
 173 {
 174     if (rare(!s.empty()))
 175         throw Xapian::SerialisationError("Extra data in DLHWeight::unserialise()");
 176     return new DLHWeight();
 177 }
 178
 179 double
 180 DLHWeight::get_sumpart(Xapian::termcount wdf, Xapian::termcount len,
 181                        Xapian::termcount) const
 182 {
 183     if (wdf == 0 || wdf == len) return 0.0;
 184
 185     double wdf_to_len = double(wdf) / len;
 186     double one_minus_wdf_to_len = 1.0 - wdf_to_len;
 187
 188     double wt = wdf * log2(wdf_to_len * log_constant) +
 189                 (len - wdf) * log2(one_minus_wdf_to_len) +
 190                 0.5 * log2(2.0 * M_PI * wdf * one_minus_wdf_to_len);
 191     if (rare(wt <= 0.0)) return 0.0;
 192
 193     return wqf_product_factor * wt / (wdf + 0.5);
 194 }
 195
 196 double
 197 DLHWeight::get_maxpart() const
 198 {
 199     return upper_bound;
 200 }
 201
 202 double
 203 DLHWeight::get_sumextra(Xapian::termcount, Xapian::termcount) const
 204 {
 205     return 0;
 206 }
 207
 208 double
 209 DLHWeight::get_maxextra() const
 210 {
 211     return 0;
 212 }
 213
 214 }