1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 2004 Free Software Foundation, Inc.
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.
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.
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/>. */
18 #include "factor-stats.h"
19 #include <data/value.h>
20 #include <libpspp/hash.h>
21 #include <libpspp/array.h>
22 #include <libpspp/alloc.h>
24 #include "percentiles.h"
30 #include "histogram.h"
34 metrics_precalc(struct metrics *m)
45 m->moments = moments1_create(MOMENT_KURTOSIS);
47 m->ordered_data = hsh_create(20,
48 (hsh_compare_func *) compare_values,
49 (hsh_hash_func *) hash_value,
50 (hsh_free_func *) weighted_value_free,
55 /* Include val in the calculation for the metrics.
56 If val is null, then treat it as MISSING
59 metrics_calc (struct metrics *fs, const union value *val,
60 double weight, int case_no)
62 struct weighted_value **wv;
67 fs->n_missing += weight;
73 moments1_add(fs->moments, x, weight);
76 if ( x < fs->min) fs->min = x;
77 if ( x > fs->max) fs->max = x;
80 wv = (struct weighted_value **) hsh_probe (fs->ordered_data,(void *) val );
84 /* If this value has already been seen, then simply
85 increase its weight and push a new case number */
89 assert( (*wv)->v.f == val->f );
92 cn = xmalloc ( sizeof *cn);
93 cn->next = (*wv)->case_nos ;
100 struct case_node *cn;
102 *wv = weighted_value_create();
106 cn = xmalloc (sizeof *cn);
109 (*wv)->case_nos = cn;
116 metrics_postcalc(struct metrics *m)
124 moments1_calculate (m->moments, &m->n, &m->mean, &m->var,
125 &m->skewness, &m->kurtosis);
127 moments1_destroy (m->moments);
130 m->stddev = sqrt(m->var);
132 /* FIXME: Check this is correct ???
133 Shouldn't we use the sample variance ??? */
134 m->se_mean = sqrt (m->var / m->n) ;
138 m->wvp = (struct weighted_value **) hsh_sort(m->ordered_data);
139 m->n_data = hsh_count(m->ordered_data);
141 /* Trimmed mean calculation */
142 if ( m->n_data <= 1 )
144 m->trimmed_mean = m->mean;
148 m->histogram = histogram_create(10, m->min, m->max);
150 for ( i = 0 ; i < m->n_data ; ++i )
152 struct weighted_value **wv = (m->wvp) ;
153 gsl_histogram_accumulate(m->histogram, wv[i]->v.f, wv[i]->w);
160 for ( i = 0 ; i < m->n_data ; ++i )
165 m->wvp[i]->rank = j + (m->wvp[i]->w - 1) / 2.0 ;
176 for ( i = m->n_data -1 ; i >= 0; --i )
178 if ( tc > m->n - m->wvp[i]->cc)
183 /* Calculate the percentiles */
184 ptiles (m->ptile_hash, (const struct weighted_value **) m->wvp,
185 m->n_data, m->n, m->ptile_alg);
187 tukey_hinges ((const struct weighted_value **) m->wvp,
188 m->n_data, m->n, m->hinge);
190 /* Special case here */
193 m->trimmed_mean = m->wvp[k2]->v.f;
198 for ( i = k1 + 2 ; i <= k2 - 1 ; ++i )
200 m->trimmed_mean += m->wvp[i]->v.f * m->wvp[i]->w;
204 m->trimmed_mean += (m->n - m->wvp[k2 - 1]->cc - tc) * m->wvp[k2]->v.f ;
205 m->trimmed_mean += (m->wvp[k1 + 1]->cc - tc) * m->wvp[k1 + 1]->v.f ;
206 m->trimmed_mean /= 0.9 * m->n ;
212 struct weighted_value *
213 weighted_value_create(void)
215 struct weighted_value *wv;
216 wv = xmalloc (sizeof *wv);
225 weighted_value_free(struct weighted_value *wv)
227 struct case_node *cn ;
236 struct case_node *next = cn->next;
250 /* Create a factor statistics object with for N dependent vars
251 and ID0 and ID1 as the values of the independent variable */
252 struct factor_statistics *
253 create_factor_statistics (int n,
257 struct factor_statistics *f;
259 f = xmalloc (sizeof *f);
263 f->m = xnmalloc (n, sizeof *f->m);
264 memset (f->m, 0, sizeof(struct metrics) * n);
271 metrics_destroy(struct metrics *m)
273 hsh_destroy(m->ordered_data);
274 hsh_destroy(m->ptile_hash);
276 gsl_histogram_free(m->histogram);
280 factor_statistics_free(struct factor_statistics *f)
286 for ( i = 0 ; i < f->n_var; ++i )
287 metrics_destroy(&f->m[i]);
294 factor_statistics_compare(const struct factor_statistics *f0,
295 const struct factor_statistics *f1, int width)
303 cmp0 = compare_values(f0->id[0], f1->id[0], width);
309 if ( ( f0->id[1]->f == SYSMIS ) && (f1->id[1]->f != SYSMIS) )
312 if ( ( f0->id[1]->f != SYSMIS ) && (f1->id[1]->f == SYSMIS) )
315 return compare_values (f0->id[1], f1->id[1], width);
319 factor_statistics_hash (const struct factor_statistics *f, int width)
323 h = hash_value (f->id[0], width);
325 if ( f->id[1]->f != SYSMIS )
326 h += hash_value(f->id[1], width);