src/math/percentiles.c

   1 /* PSPP - a program for statistical analysis.
   2    Copyright (C) 2008, 2009, 2011 Free Software Foundation, Inc.
   3
   4    This program is free software: you can redistribute it and/or modify
   5    it under the terms of the GNU General Public License as published by
   6    the Free Software Foundation, either version 3 of the License, or
   7    (at your option) any later version.
   8
   9    This program is distributed in the hope that it will be useful,
  10    but WITHOUT ANY WARRANTY; without even the implied warranty of
  11    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12    GNU General Public License for more details.
  13
  14    You should have received a copy of the GNU General Public License
  15    along with this program.  If not, see <http://www.gnu.org/licenses/>. */
  16
  17 #include <config.h>
  18
  19 #include "math/percentiles.h"
  20
  21 #include "data/casereader.h"
  22 #include "data/val-type.h"
  23 #include "data/variable.h"
  24 #include "libpspp/assertion.h"
  25 #include "libpspp/cast.h"
  26 #include "math/order-stats.h"
  27
  28 #include "gl/xalloc.h"
  29
  30 /* Return the value of the percentile. */
  31 double
  32 percentile_calculate (const struct percentile *ptl, enum pc_alg alg)
  33 {
  34   struct percentile *mutable = CONST_CAST (struct percentile *, ptl);
  35   const struct order_stats *os = &ptl->parent;
  36
  37   if (ptl->g1 == SYSMIS)
  38     mutable->g1 = (os->k[0].tc - os->k[0].cc) / os->k[0].c_p1;
  39
  40   if (ptl->g1_star == SYSMIS)
  41     mutable->g1_star = os->k[0].tc - os->k[0].cc;
  42
  43   if (ptl->g2 == SYSMIS)
  44     {
  45       if (os->k[1].c == 0)
  46         mutable->g2 = os->k[1].tc / os->k[1].c_p1;
  47       else if (os->k[1].c_p1 == 0)
  48         mutable->g2 = 0;
  49       else
  50         mutable->g2 = (os->k[1].tc - os->k[1].cc) / os->k[1].c_p1;
  51     }
  52
  53   if (ptl->g2_star == SYSMIS)
  54     {
  55       if (os->k[1].c == 0)
  56         mutable->g2_star = os->k[1].tc;
  57       else if (os->k[1].c_p1 == 0)
  58         mutable->g2_star = 0;
  59       else
  60         mutable->g2_star = os->k[1].tc - os->k[1].cc;
  61     }
  62
  63   switch (alg)
  64     {
  65     case PC_WAVERAGE:
  66       if (ptl->g1_star >= 1.0)
  67         return os->k[0].y_p1;
  68       else
  69         {
  70           double a = (os->k[0].y == SYSMIS) ? 0 : os->k[0].y;
  71
  72           if (os->k[0].c_p1 >= 1.0)
  73             return (1 - ptl->g1_star) * a + ptl->g1_star * os->k[0].y_p1;
  74           else
  75             return (1 - ptl->g1) * a + ptl->g1 * os->k[0].y_p1;
  76         }
  77       break;
  78
  79     case PC_ROUND:
  80       {
  81         double a = (os->k[0].y == SYSMIS) ? 0 : os->k[0].y;
  82
  83         if (os->k[0].c_p1 >= 1.0)
  84           return (ptl->g1_star < 0.5) ? a : os->k[0].y_p1;
  85         else
  86           return (ptl->g1 < 0.5) ? a : os->k[0].y_p1;
  87       }
  88       break;
  89
  90     case PC_EMPIRICAL:
  91       if (ptl->g1_star == 0)
  92         return os->k[0].y;
  93       else
  94         return os->k[0].y_p1;
  95       break;
  96
  97     case PC_HAVERAGE:
  98       if (ptl->g2_star >= 1.0)
  99         {
 100           return os->k[1].y_p1;
 101         }
 102       else
 103         {
 104           double a = (os->k[1].y == SYSMIS) ? 0 : os->k[1].y;
 105
 106           if (os->k[1].c_p1 >= 1.0)
 107             {
 108               if (ptl->g2_star == 0)
 109                 return os->k[1].y;
 110
 111               return (1 - ptl->g2_star) * a + ptl->g2_star * os->k[1].y_p1;
 112             }
 113           else
 114             {
 115               return (1 - ptl->g2) * a + ptl->g2 * os->k[1].y_p1;
 116             }
 117         }
 118
 119       break;
 120
 121     case PC_AEMPIRICAL:
 122       if (ptl->g1_star == 0)
 123         return (os->k[0].y + os->k[0].y_p1)/ 2.0;
 124       else
 125         return os->k[0].y_p1;
 126       break;
 127
 128     default:
 129       NOT_REACHED ();
 130       break;
 131     }
 132
 133   NOT_REACHED ();
 134
 135   return SYSMIS;
 136 }
 137
 138
 139 static void
 140 destroy (struct statistic *stat)
 141 {
 142   struct percentile *ptl = UP_CAST (stat, struct percentile, parent.parent);
 143   free (ptl);
 144 }
 145
 146 /* Create the Pth percentile.
 147    W is the total sum of weights in the data set.
 148 */
 149 struct percentile *
 150 percentile_create (double p, double W)
 151 {
 152   assert (p >= 0);
 153   assert (p <= 1.0);
 154
 155   struct percentile *ptl = xmalloc (sizeof *ptl);
 156   *ptl = (struct percentile) {
 157     .parent = {
 158       .parent = {
 159         .destroy = destroy
 160       },
 161       .k = ptl->k,
 162       .n_k = 2,
 163     },
 164     .ptile = p,
 165     .w = W,
 166     .g1 = SYSMIS,
 167     .g1_star = SYSMIS,
 168     .g2 = SYSMIS,
 169     .g2_star = SYSMIS,
 170     .k[0] = { .tc = W * p, .y = SYSMIS, .y_p1 = SYSMIS },
 171     .k[1] = { .tc = (W + 1.0) * p, .y = SYSMIS, .y_p1 = SYSMIS },
 172   };
 173   return ptl;
 174 }