2 PSPP - a program for statistical analysis.
3 Copyright (C) 1997-9, 2000, 2007, 2009, 2010, 2011, 2014, 2015 Free Software Foundation, Inc.
5 This program is free software: you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation, either version 3 of the License, or
8 (at your option) any later version.
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with this program. If not, see <http://www.gnu.org/licenses/>.
21 #include <gsl/gsl_histogram.h>
24 #include "data/case.h"
25 #include "data/casegrouper.h"
26 #include "data/casereader.h"
27 #include "data/dataset.h"
28 #include "data/dictionary.h"
29 #include "data/format.h"
30 #include "data/missing-values.h"
31 #include "data/settings.h"
32 #include "data/value-labels.h"
33 #include "data/variable.h"
35 #include "language/dictionary/split-file.h"
37 #include "language/command.h"
38 #include "language/lexer/lexer.h"
39 #include "language/lexer/variable-parser.h"
40 #include "language/stats/freq.h"
42 #include "libpspp/array.h"
43 #include "libpspp/bit-vector.h"
44 #include "libpspp/compiler.h"
45 #include "libpspp/hmap.h"
46 #include "libpspp/message.h"
47 #include "libpspp/misc.h"
48 #include "libpspp/pool.h"
50 #include "math/histogram.h"
51 #include "math/moments.h"
52 #include "math/chart-geometry.h"
55 #include "output/chart-item.h"
56 #include "output/charts/barchart.h"
57 #include "output/charts/piechart.h"
58 #include "output/charts/plot-hist.h"
59 #include "output/pivot-table.h"
61 #include "gl/minmax.h"
62 #include "gl/xalloc.h"
65 #define _(msgid) gettext (msgid)
66 #define N_(msgid) msgid
68 /* Percentiles to calculate. */
72 double p; /* the %ile to be calculated */
73 double value; /* the %ile's value */
74 bool show; /* True to show this percentile in the statistics box. */
78 ptile_3way (const void *_p1, const void *_p2)
80 const struct percentile *p1 = _p1;
81 const struct percentile *p2 = _p2;
88 if (p1->show > p2->show)
91 return (p1->show < p2->show);
94 return (p1->p > p2->p);
118 /* Array indices for STATISTICS subcommand. */
138 /* Description of statistics. */
139 static const char *st_name[FRQ_ST_count] =
159 struct hmap data; /* Hash table for accumulating counts. */
160 struct freq *valid; /* Valid freqs. */
161 int n_valid; /* Number of total freqs. */
162 const struct dictionary *dict; /* Source of entries in the table. */
164 struct freq *missing; /* Missing freqs. */
165 int n_missing; /* Number of missing freqs. */
168 double total_cases; /* Sum of weights of all cases. */
169 double valid_cases; /* Sum of weights of valid cases. */
174 double x_min; /* X axis minimum value. */
175 double x_max; /* X axis maximum value. */
176 int y_scale; /* Y axis scale: FRQ_FREQ or FRQ_PERCENT. */
178 /* Histograms only. */
179 double y_max; /* Y axis maximum value. */
180 bool draw_normal; /* Whether to draw normal curve. */
182 /* Pie charts only. */
183 bool include_missing; /* Whether to include missing values. */
186 /* Per-variable frequency data. */
189 const struct variable *var;
191 /* Freqency table. */
192 struct freq_tab tab; /* Frequencies table to use. */
195 int n_groups; /* Number of groups. */
196 double *groups; /* Groups. */
199 double stat[FRQ_ST_count];
201 /* Variable attributes. */
209 struct var_freqs *vars;
212 /* Percentiles to calculate and possibly display. */
213 struct percentile *percentiles;
214 const struct percentile *median;
217 /* Frequency table display. */
218 long int max_categories; /* Maximum categories to show. */
219 int sort; /* FRQ_AVALUE or FRQ_DVALUE
220 or FRQ_AFREQ or FRQ_DFREQ. */
222 /* Statistics; number of statistics. */
226 /* Histogram and pie chart settings. */
227 struct frq_chart *hist, *pie, *bar;
233 struct freq_compare_aux
239 bool ascending_value;
242 static void calc_stats (const struct frq_proc *,
243 const struct var_freqs *, double d[FRQ_ST_count]);
245 static void do_piechart(const struct frq_chart *pie,
246 const struct variable *var,
247 const struct freq_tab *frq_tab);
249 static void do_barchart(const struct frq_chart *bar,
250 const struct variable **var,
251 const struct freq_tab *frq_tab);
253 static void dump_statistics (const struct frq_proc *frq,
254 const struct variable *wv);
257 compare_freq (const void *a_, const void *b_, const void *aux_)
259 const struct freq_compare_aux *aux = aux_;
260 const struct freq *a = a_;
261 const struct freq *b = b_;
263 if (aux->by_freq && a->count != b->count)
265 int cmp = a->count > b->count ? 1 : -1;
266 return aux->ascending_freq ? cmp : -cmp;
270 int cmp = value_compare_3way (a->values, b->values, aux->width);
271 return aux->ascending_value ? cmp : -cmp;
275 /* Create a gsl_histogram from a freq_tab */
276 static struct histogram *
277 freq_tab_to_hist (const struct frq_proc *frq, const struct freq_tab *ft,
278 const struct variable *var);
281 put_freq_row (struct pivot_table *table, int var_idx,
282 double frequency, double percent,
283 double valid_percent, double cum_percent)
285 double entries[] = { frequency, percent, valid_percent, cum_percent };
286 for (size_t i = 0; i < sizeof entries / sizeof *entries; i++)
287 if (entries[i] != SYSMIS)
288 pivot_table_put2 (table, i, var_idx,
289 pivot_value_new_number (entries[i]));
292 /* Displays a full frequency table for variable V. */
294 dump_freq_table (const struct var_freqs *vf, const struct variable *wv)
296 const struct freq_tab *ft = &vf->tab;
298 struct pivot_table *table = pivot_table_create__ (pivot_value_new_variable (
300 pivot_table_set_weight_var (table, wv);
302 pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Statistics"),
303 N_("Frequency"), PIVOT_RC_COUNT,
304 N_("Percent"), PIVOT_RC_PERCENT,
305 N_("Valid Percent"), PIVOT_RC_PERCENT,
306 N_("Cumulative Percent"), PIVOT_RC_PERCENT);
308 struct pivot_dimension *variable = pivot_dimension_create__ (
309 table, PIVOT_AXIS_ROW, pivot_value_new_variable (vf->var));
311 double cum_freq = 0.0;
312 double cum_percent = 0.0;
313 struct pivot_category *valid = NULL;
314 for (const struct freq *f = ft->valid; f < ft->missing; f++)
316 cum_freq += f->count;
317 double valid_percent = f->count / ft->valid_cases * 100.0;
318 cum_percent += valid_percent;
321 valid = pivot_category_create_group (variable->root, N_("Valid"));
322 int var_idx = pivot_category_create_leaf (
323 valid, pivot_value_new_var_value (vf->var, &f->values[0]));
324 put_freq_row (table, var_idx, f->count,
325 f->count / ft->total_cases * 100.0,
326 valid_percent, cum_percent);
329 struct pivot_category *missing = NULL;
330 size_t n_categories = ft->n_valid + ft->n_missing;
331 for (const struct freq *f = ft->missing; f < &ft->valid[n_categories]; f++)
333 cum_freq += f->count;
336 missing = pivot_category_create_group (variable->root, N_("Missing"));
337 int var_idx = pivot_category_create_leaf (
338 missing, pivot_value_new_var_value (vf->var, &f->values[0]));
339 put_freq_row (table, var_idx, f->count,
340 f->count / ft->total_cases * 100.0, SYSMIS, SYSMIS);
343 int var_idx = pivot_category_create_leaf (
344 variable->root, pivot_value_new_text (N_("Total")));
345 put_freq_row (table, var_idx, cum_freq, cum_percent, SYSMIS, SYSMIS);
347 pivot_table_submit (table);
350 /* Statistical display. */
353 calc_percentile (double p, double valid_cases, double x1, double x2)
357 s = (settings_get_algorithm () != COMPATIBLE
358 ? modf ((valid_cases - 1) * p, &dummy)
359 : modf ((valid_cases + 1) * p - 1, &dummy));
361 return x1 + (x2 - x1) * s;
364 /* Calculates all of the percentiles for VF within FRQ. */
366 calc_percentiles (const struct frq_proc *frq, const struct var_freqs *vf)
368 const struct freq_tab *ft = &vf->tab;
369 double W = ft->valid_cases;
370 const struct freq *f;
371 int percentile_idx = 0;
374 for (f = ft->valid; f < ft->missing; f++)
377 for (; percentile_idx < frq->n_percentiles; percentile_idx++)
379 struct percentile *pc = &frq->percentiles[percentile_idx];
382 tp = (settings_get_algorithm () == ENHANCED
384 : (W + 1) * pc->p - 1);
389 if (tp + 1 < rank || f + 1 >= ft->missing)
390 pc->value = f->values[0].f;
392 pc->value = calc_percentile (pc->p, W, f->values[0].f, f[1].values[0].f);
395 for (; percentile_idx < frq->n_percentiles; percentile_idx++)
397 struct percentile *pc = &frq->percentiles[percentile_idx];
398 pc->value = (ft->n_valid > 0
399 ? ft->valid[ft->n_valid - 1].values[0].f
404 /* Returns true iff the value in struct freq F is non-missing
407 not_missing (const void *f_, const void *v_)
409 const struct freq *f = f_;
410 const struct variable *v = v_;
412 return !var_is_value_missing (v, f->values, MV_ANY);
416 /* Summarizes the frequency table data for variable V. */
418 postprocess_freq_tab (const struct frq_proc *frq, struct var_freqs *vf)
420 struct freq_tab *ft = &vf->tab;
421 struct freq_compare_aux aux;
423 struct freq *freqs, *f;
426 /* Extract data from hash table. */
427 count = hmap_count (&ft->data);
428 freqs = freq_hmap_extract (&ft->data);
430 /* Put data into ft. */
432 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, vf->var);
433 ft->missing = freqs + ft->n_valid;
434 ft->n_missing = count - ft->n_valid;
437 aux.by_freq = frq->sort == FRQ_AFREQ || frq->sort == FRQ_DFREQ;
438 aux.ascending_freq = frq->sort != FRQ_DFREQ;
439 aux.width = vf->width;
440 aux.ascending_value = frq->sort != FRQ_DVALUE;
441 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare_freq, &aux);
442 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare_freq, &aux);
444 /* Summary statistics. */
445 ft->valid_cases = 0.0;
446 for(i = 0 ; i < ft->n_valid ; ++i )
449 ft->valid_cases += f->count;
453 ft->total_cases = ft->valid_cases ;
454 for(i = 0 ; i < ft->n_missing ; ++i )
457 ft->total_cases += f->count;
462 /* Frees the frequency table for variable V. */
464 cleanup_freq_tab (struct var_freqs *vf)
466 free (vf->tab.valid);
467 freq_hmap_destroy (&vf->tab.data, vf->width);
470 /* Add data from case C to the frequency table. */
472 calc (struct frq_proc *frq, const struct ccase *c, const struct dataset *ds)
474 double weight = dict_get_case_weight (dataset_dict (ds), c, &frq->warn);
477 for (i = 0; i < frq->n_vars; i++)
479 struct var_freqs *vf = &frq->vars[i];
480 const union value *value = case_data (c, vf->var);
481 size_t hash = value_hash (value, vf->width, 0);
484 f = freq_hmap_search (&vf->tab.data, value, vf->width, hash);
486 f = freq_hmap_insert (&vf->tab.data, value, vf->width, hash);
492 /* Prepares each variable that is the target of FREQUENCIES by setting
493 up its hash table. */
495 precalc (struct frq_proc *frq, struct casereader *input, struct dataset *ds)
500 c = casereader_peek (input, 0);
503 output_split_file_values (ds, c);
507 for (i = 0; i < frq->n_vars; i++)
508 hmap_init (&frq->vars[i].tab.data);
511 /* Finishes up with the variables after frequencies have been
512 calculated. Displays statistics, percentiles, ... */
514 postcalc (struct frq_proc *frq, const struct dataset *ds)
516 const struct dictionary *dict = dataset_dict (ds);
517 const struct variable *wv = dict_get_weight (dict);
520 for (i = 0; i < frq->n_vars; i++)
522 struct var_freqs *vf = &frq->vars[i];
523 postprocess_freq_tab (frq, vf);
524 calc_percentiles (frq, vf);
528 dump_statistics (frq, wv);
530 for (i = 0; i < frq->n_vars; i++)
532 struct var_freqs *vf = &frq->vars[i];
534 /* Frequencies tables. */
535 if (vf->tab.n_valid + vf->tab.n_missing <= frq->max_categories)
536 dump_freq_table (vf, wv);
539 if (frq->hist && var_is_numeric (vf->var) && vf->tab.n_valid > 0)
541 double d[FRQ_ST_count];
542 struct histogram *histogram;
544 calc_stats (frq, vf, d);
546 histogram = freq_tab_to_hist (frq, &vf->tab, vf->var);
550 chart_item_submit (histogram_chart_create (
551 histogram->gsl_hist, var_to_string(vf->var),
555 frq->hist->draw_normal));
557 statistic_destroy (&histogram->parent);
562 do_piechart(frq->pie, vf->var, &vf->tab);
565 do_barchart(frq->bar, &vf->var, &vf->tab);
567 cleanup_freq_tab (vf);
572 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
576 const struct variable **vars = NULL;
578 bool sbc_barchart = false;
579 bool sbc_piechart = false;
580 bool sbc_histogram = false;
582 double pie_min = -DBL_MAX;
583 double pie_max = DBL_MAX;
584 bool pie_missing = true;
586 double bar_min = -DBL_MAX;
587 double bar_max = DBL_MAX;
588 bool bar_freq = true;
590 double hi_min = -DBL_MAX;
591 double hi_max = DBL_MAX;
592 int hi_scale = FRQ_FREQ;
593 int hi_freq = INT_MIN;
594 int hi_pcnt = INT_MIN;
595 int hi_norm = FRQ_NONORMAL;
597 frq.pool = pool_create ();
598 frq.sort = FRQ_AVALUE;
603 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
604 | BIT_INDEX (FRQ_ST_STDDEV)
605 | BIT_INDEX (FRQ_ST_MINIMUM)
606 | BIT_INDEX (FRQ_ST_MAXIMUM);
610 frq.max_categories = LONG_MAX;
612 frq.percentiles = NULL;
613 frq.n_percentiles = 0;
621 /* Accept an optional, completely pointless "/VARIABLES=" */
622 lex_match (lexer, T_SLASH);
623 if (lex_match_id (lexer, "VARIABLES"))
625 if (! lex_force_match (lexer, T_EQUALS) )
629 if (!parse_variables_const (lexer, dataset_dict (ds),
635 frq.vars = xzalloc (frq.n_vars * sizeof (*frq.vars));
636 for (i = 0; i < frq.n_vars; ++i)
638 frq.vars[i].var = vars[i];
639 frq.vars[i].width = var_get_width (vars[i]);
642 while (lex_token (lexer) != T_ENDCMD)
644 lex_match (lexer, T_SLASH);
646 if (lex_match_id (lexer, "STATISTICS"))
648 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
649 | BIT_INDEX (FRQ_ST_STDDEV)
650 | BIT_INDEX (FRQ_ST_MINIMUM)
651 | BIT_INDEX (FRQ_ST_MAXIMUM);
655 if (lex_match (lexer, T_EQUALS))
661 while (lex_token (lexer) != T_ENDCMD
662 && lex_token (lexer) != T_SLASH)
664 if (lex_match_id (lexer, "DEFAULT"))
666 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
667 | BIT_INDEX (FRQ_ST_STDDEV)
668 | BIT_INDEX (FRQ_ST_MINIMUM)
669 | BIT_INDEX (FRQ_ST_MAXIMUM);
673 else if (lex_match_id (lexer, "MEAN"))
675 frq.stats |= BIT_INDEX (FRQ_ST_MEAN);
678 else if (lex_match_id (lexer, "SEMEAN"))
680 frq.stats |= BIT_INDEX (FRQ_ST_SEMEAN);
683 else if (lex_match_id (lexer, "MEDIAN"))
685 frq.stats |= BIT_INDEX (FRQ_ST_MEDIAN);
688 else if (lex_match_id (lexer, "MODE"))
690 frq.stats |= BIT_INDEX (FRQ_ST_MODE);
693 else if (lex_match_id (lexer, "STDDEV"))
695 frq.stats |= BIT_INDEX (FRQ_ST_STDDEV);
698 else if (lex_match_id (lexer, "VARIANCE"))
700 frq.stats |= BIT_INDEX (FRQ_ST_VARIANCE);
703 else if (lex_match_id (lexer, "KURTOSIS"))
705 frq.stats |= BIT_INDEX (FRQ_ST_KURTOSIS);
708 else if (lex_match_id (lexer, "SKEWNESS"))
710 frq.stats |= BIT_INDEX (FRQ_ST_SKEWNESS);
713 else if (lex_match_id (lexer, "RANGE"))
715 frq.stats |= BIT_INDEX (FRQ_ST_RANGE);
718 else if (lex_match_id (lexer, "MINIMUM"))
720 frq.stats |= BIT_INDEX (FRQ_ST_MINIMUM);
723 else if (lex_match_id (lexer, "MAXIMUM"))
725 frq.stats |= BIT_INDEX (FRQ_ST_MAXIMUM);
728 else if (lex_match_id (lexer, "SUM"))
730 frq.stats |= BIT_INDEX (FRQ_ST_SUM);
733 else if (lex_match_id (lexer, "SESKEWNESS"))
735 frq.stats |= BIT_INDEX (FRQ_ST_SESKEWNESS);
738 else if (lex_match_id (lexer, "SEKURTOSIS"))
740 frq.stats |= BIT_INDEX (FRQ_ST_SEKURTOSIS);
743 else if (lex_match_id (lexer, "NONE"))
748 else if (lex_match (lexer, T_ALL))
751 frq.n_stats = FRQ_ST_count;
755 lex_error (lexer, NULL);
760 else if (lex_match_id (lexer, "PERCENTILES"))
762 lex_match (lexer, T_EQUALS);
763 while (lex_token (lexer) != T_ENDCMD
764 && lex_token (lexer) != T_SLASH)
766 if (lex_force_num (lexer))
769 xrealloc (frq.percentiles,
770 (frq.n_percentiles + 1)
771 * sizeof (*frq.percentiles));
772 frq.percentiles[frq.n_percentiles].p = lex_number (lexer) / 100.0;
773 frq.percentiles[frq.n_percentiles].show = true;
779 lex_error (lexer, NULL);
782 lex_match (lexer, T_COMMA);
785 else if (lex_match_id (lexer, "FORMAT"))
787 lex_match (lexer, T_EQUALS);
788 while (lex_token (lexer) != T_ENDCMD
789 && lex_token (lexer) != T_SLASH)
791 if (lex_match_id (lexer, "TABLE"))
794 else if (lex_match_id (lexer, "NOTABLE"))
796 frq.max_categories = 0;
798 else if (lex_match_id (lexer, "LIMIT"))
800 if (!lex_force_match (lexer, T_LPAREN)
801 || !lex_force_int (lexer))
804 frq.max_categories = lex_integer (lexer);
807 if (!lex_force_match (lexer, T_RPAREN))
810 else if (lex_match_id (lexer, "AVALUE"))
812 frq.sort = FRQ_AVALUE;
814 else if (lex_match_id (lexer, "DVALUE"))
816 frq.sort = FRQ_DVALUE;
818 else if (lex_match_id (lexer, "AFREQ"))
820 frq.sort = FRQ_AFREQ;
822 else if (lex_match_id (lexer, "DFREQ"))
824 frq.sort = FRQ_DFREQ;
828 lex_error (lexer, NULL);
833 else if (lex_match_id (lexer, "NTILES"))
835 lex_match (lexer, T_EQUALS);
837 if (lex_force_int (lexer))
840 int n = lex_integer (lexer);
842 for (i = 0; i < n + 1; ++i)
845 xrealloc (frq.percentiles,
846 (frq.n_percentiles + 1)
847 * sizeof (*frq.percentiles));
848 frq.percentiles[frq.n_percentiles].p =
850 frq.percentiles[frq.n_percentiles].show = true;
857 lex_error (lexer, NULL);
861 else if (lex_match_id (lexer, "ALGORITHM"))
863 lex_match (lexer, T_EQUALS);
865 if (lex_match_id (lexer, "COMPATIBLE"))
867 settings_set_cmd_algorithm (COMPATIBLE);
869 else if (lex_match_id (lexer, "ENHANCED"))
871 settings_set_cmd_algorithm (ENHANCED);
875 lex_error (lexer, NULL);
879 else if (lex_match_id (lexer, "HISTOGRAM"))
881 lex_match (lexer, T_EQUALS);
882 sbc_histogram = true;
884 while (lex_token (lexer) != T_ENDCMD
885 && lex_token (lexer) != T_SLASH)
887 if (lex_match_id (lexer, "NORMAL"))
889 hi_norm = FRQ_NORMAL;
891 else if (lex_match_id (lexer, "NONORMAL"))
893 hi_norm = FRQ_NONORMAL;
895 else if (lex_match_id (lexer, "FREQ"))
898 if (lex_match (lexer, T_LPAREN))
900 if (lex_force_int (lexer))
902 hi_freq = lex_integer (lexer);
905 lex_error (lexer, _("Histogram frequency must be greater than zero."));
908 if (! lex_force_match (lexer, T_RPAREN))
913 else if (lex_match_id (lexer, "PERCENT"))
915 hi_scale = FRQ_PERCENT;
916 if (lex_match (lexer, T_LPAREN))
918 if (lex_force_int (lexer))
920 hi_pcnt = lex_integer (lexer);
923 lex_error (lexer, _("Histogram percentage must be greater than zero."));
926 if (! lex_force_match (lexer, T_RPAREN))
931 else if (lex_match_id (lexer, "MINIMUM"))
933 if (! lex_force_match (lexer, T_LPAREN))
935 if (lex_force_num (lexer))
937 hi_min = lex_number (lexer);
940 if (! lex_force_match (lexer, T_RPAREN))
943 else if (lex_match_id (lexer, "MAXIMUM"))
945 if (! lex_force_match (lexer, T_LPAREN))
947 if (lex_force_num (lexer))
949 hi_max = lex_number (lexer);
952 if (! lex_force_match (lexer, T_RPAREN))
957 lex_error (lexer, NULL);
962 else if (lex_match_id (lexer, "PIECHART"))
964 lex_match (lexer, T_EQUALS);
965 while (lex_token (lexer) != T_ENDCMD
966 && lex_token (lexer) != T_SLASH)
968 if (lex_match_id (lexer, "MINIMUM"))
970 if (! lex_force_match (lexer, T_LPAREN))
972 if (lex_force_num (lexer))
974 pie_min = lex_number (lexer);
977 if (! lex_force_match (lexer, T_RPAREN))
980 else if (lex_match_id (lexer, "MAXIMUM"))
982 if (! lex_force_match (lexer, T_LPAREN))
984 if (lex_force_num (lexer))
986 pie_max = lex_number (lexer);
989 if (! lex_force_match (lexer, T_RPAREN))
992 else if (lex_match_id (lexer, "MISSING"))
996 else if (lex_match_id (lexer, "NOMISSING"))
1002 lex_error (lexer, NULL);
1006 sbc_piechart = true;
1008 else if (lex_match_id (lexer, "BARCHART"))
1010 lex_match (lexer, T_EQUALS);
1011 while (lex_token (lexer) != T_ENDCMD
1012 && lex_token (lexer) != T_SLASH)
1014 if (lex_match_id (lexer, "MINIMUM"))
1016 if (! lex_force_match (lexer, T_LPAREN))
1018 if (lex_force_num (lexer))
1020 bar_min = lex_number (lexer);
1023 if (! lex_force_match (lexer, T_RPAREN))
1026 else if (lex_match_id (lexer, "MAXIMUM"))
1028 if (! lex_force_match (lexer, T_LPAREN))
1030 if (lex_force_num (lexer))
1032 bar_max = lex_number (lexer);
1035 if (! lex_force_match (lexer, T_RPAREN))
1038 else if (lex_match_id (lexer, "FREQ"))
1040 if ( lex_match (lexer, T_LPAREN))
1042 if (lex_force_num (lexer))
1047 if (! lex_force_match (lexer, T_RPAREN))
1052 else if (lex_match_id (lexer, "PERCENT"))
1054 if ( lex_match (lexer, T_LPAREN))
1056 if (lex_force_num (lexer))
1061 if (! lex_force_match (lexer, T_RPAREN))
1068 lex_error (lexer, NULL);
1072 sbc_barchart = true;
1074 else if (lex_match_id (lexer, "MISSING"))
1076 lex_match (lexer, T_EQUALS);
1078 while (lex_token (lexer) != T_ENDCMD
1079 && lex_token (lexer) != T_SLASH)
1081 if (lex_match_id (lexer, "EXCLUDE"))
1084 else if (lex_match_id (lexer, "INCLUDE"))
1089 lex_error (lexer, NULL);
1094 else if (lex_match_id (lexer, "ORDER"))
1096 lex_match (lexer, T_EQUALS);
1097 if (!lex_match_id (lexer, "ANALYSIS"))
1098 lex_match_id (lexer, "VARIABLE");
1102 lex_error (lexer, NULL);
1107 if (frq.stats & BIT_INDEX (FRQ_ST_MEDIAN))
1110 xrealloc (frq.percentiles,
1111 (frq.n_percentiles + 1)
1112 * sizeof (*frq.percentiles));
1114 frq.percentiles[frq.n_percentiles].p = 0.50;
1115 frq.percentiles[frq.n_percentiles].show = false;
1117 frq.n_percentiles++;
1121 /* Figure out which charts the user requested. */
1126 struct frq_chart *hist;
1128 hist = frq.hist = xmalloc (sizeof *frq.hist);
1129 hist->x_min = hi_min;
1130 hist->x_max = hi_max;
1131 hist->y_scale = hi_scale;
1132 hist->y_max = hi_scale == FRQ_FREQ ? hi_freq : hi_pcnt;
1133 hist->draw_normal = hi_norm != FRQ_NONORMAL;
1134 hist->include_missing = false;
1136 if (hist->x_min != SYSMIS && hist->x_max != SYSMIS
1137 && hist->x_min >= hist->x_max)
1139 msg (SE, _("%s for histogram must be greater than or equal to %s, "
1140 "but %s was specified as %.15g and %s as %.15g. "
1141 "%s and %s will be ignored."),
1146 hist->x_min = hist->x_max = SYSMIS;
1150 xrealloc (frq.percentiles,
1151 (frq.n_percentiles + 2)
1152 * sizeof (*frq.percentiles));
1154 frq.percentiles[frq.n_percentiles].p = 0.25;
1155 frq.percentiles[frq.n_percentiles].show = false;
1157 frq.percentiles[frq.n_percentiles + 1].p = 0.75;
1158 frq.percentiles[frq.n_percentiles + 1].show = false;
1160 frq.n_percentiles+=2;
1165 frq.bar = xmalloc (sizeof *frq.bar);
1166 frq.bar->x_min = bar_min;
1167 frq.bar->x_max = bar_max;
1168 frq.bar->include_missing = false;
1169 frq.bar->y_scale = bar_freq ? FRQ_FREQ : FRQ_PERCENT;
1174 struct frq_chart *pie;
1176 pie = frq.pie = xmalloc (sizeof *frq.pie);
1177 pie->x_min = pie_min;
1178 pie->x_max = pie_max;
1179 pie->include_missing = pie_missing;
1181 if (pie->x_min != SYSMIS && pie->x_max != SYSMIS
1182 && pie->x_min >= pie->x_max)
1184 msg (SE, _("%s for pie chart must be greater than or equal to %s, "
1185 "but %s was specified as %.15g and %s as %.15g. "
1186 "%s and %s will be ignored."),
1191 pie->x_min = pie->x_max = SYSMIS;
1198 double previous_p = -1;
1199 qsort (frq.percentiles, frq.n_percentiles,
1200 sizeof (*frq.percentiles),
1203 for (i = o = 0; i < frq.n_percentiles; ++i)
1205 if (frq.percentiles[i].p != previous_p)
1207 frq.percentiles[o].p = frq.percentiles[i].p;
1208 frq.percentiles[o].show = frq.percentiles[i].show;
1211 else if (frq.percentiles[i].show &&
1212 !frq.percentiles[o].show)
1214 frq.percentiles[o].show = true;
1216 previous_p = frq.percentiles[i].p;
1219 frq.n_percentiles = o;
1222 for (i = 0; i < frq.n_percentiles; i++)
1223 if (frq.percentiles[i].p == 0.5)
1225 frq.median = &frq.percentiles[i];
1231 struct casegrouper *grouper;
1232 struct casereader *group;
1235 grouper = casegrouper_create_splits (proc_open (ds), dataset_dict (ds));
1236 while (casegrouper_get_next_group (grouper, &group))
1239 precalc (&frq, group, ds);
1241 for (; (c = casereader_read (group)) != NULL; case_unref (c))
1243 postcalc (&frq, ds);
1244 casereader_destroy (group);
1246 ok = casegrouper_destroy (grouper);
1247 ok = proc_commit (ds) && ok;
1256 free (frq.percentiles);
1257 pool_destroy (frq.pool);
1268 free (frq.percentiles);
1269 pool_destroy (frq.pool);
1275 calculate_iqr (const struct frq_proc *frq)
1281 /* This cannot work unless the 25th and 75th percentile are calculated */
1282 assert (frq->n_percentiles >= 2);
1283 for (i = 0; i < frq->n_percentiles; i++)
1285 struct percentile *pc = &frq->percentiles[i];
1287 if (fabs (0.25 - pc->p) < DBL_EPSILON)
1289 else if (fabs (0.75 - pc->p) < DBL_EPSILON)
1293 return q1 == SYSMIS || q3 == SYSMIS ? SYSMIS : q3 - q1;
1297 chart_includes_value (const struct frq_chart *chart,
1298 const struct variable *var,
1299 const union value *value)
1301 if (!chart->include_missing && var_is_value_missing (var, value, MV_ANY))
1304 if (var_is_numeric (var)
1305 && ((chart->x_min != SYSMIS && value->f < chart->x_min)
1306 || (chart->x_max != SYSMIS && value->f > chart->x_max)))
1312 /* Create a gsl_histogram from a freq_tab */
1313 static struct histogram *
1314 freq_tab_to_hist (const struct frq_proc *frq, const struct freq_tab *ft,
1315 const struct variable *var)
1317 double x_min, x_max, valid_freq;
1320 struct histogram *histogram;
1323 /* Find out the extremes of the x value, within the range to be included in
1324 the histogram, and sum the total frequency of those values. */
1328 for (i = 0; i < ft->n_valid; i++)
1330 const struct freq *f = &ft->valid[i];
1331 if (chart_includes_value (frq->hist, var, f->values))
1333 x_min = MIN (x_min, f->values[0].f);
1334 x_max = MAX (x_max, f->values[0].f);
1335 valid_freq += f->count;
1339 if (valid_freq <= 0)
1342 iqr = calculate_iqr (frq);
1345 /* Freedman-Diaconis' choice of bin width. */
1346 bin_width = 2 * iqr / pow (valid_freq, 1.0 / 3.0);
1350 bin_width = (x_max - x_min) / (1 + log2 (valid_freq));
1352 histogram = histogram_create (bin_width, x_min, x_max);
1354 if ( histogram == NULL)
1357 for (i = 0; i < ft->n_valid; i++)
1359 const struct freq *f = &ft->valid[i];
1360 if (chart_includes_value (frq->hist, var, f->values))
1361 histogram_add (histogram, f->values[0].f, f->count);
1368 /* Allocate an array of struct freqs and fill them from the data in FRQ_TAB,
1369 according to the parameters of CATCHART
1370 N_SLICES will contain the number of slices allocated.
1371 The caller is responsible for freeing slices
1373 static struct freq *
1374 pick_cat_counts (const struct frq_chart *catchart,
1375 const struct freq_tab *frq_tab,
1380 struct freq *slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1382 for (i = 0; i < frq_tab->n_valid; i++)
1384 const struct freq *f = &frq_tab->valid[i];
1385 if (f->count > catchart->x_max)
1388 if (f->count < catchart->x_min)
1391 slices[n_slices] = *f;
1396 if (catchart->include_missing)
1398 for (i = 0; i < frq_tab->n_missing; i++)
1400 const struct freq *f = &frq_tab->missing[i];
1401 slices[n_slices].count += f->count;
1404 slices[n_slices].values[0] = f->values[0];
1407 if (frq_tab->n_missing > 0)
1411 *n_slicesp = n_slices;
1416 /* Allocate an array of struct freqs and fill them from the data in FRQ_TAB,
1417 according to the parameters of CATCHART
1418 N_SLICES will contain the number of slices allocated.
1419 The caller is responsible for freeing slices
1421 static struct freq **
1422 pick_cat_counts_ptr (const struct frq_chart *catchart,
1423 const struct freq_tab *frq_tab,
1428 struct freq **slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1430 for (i = 0; i < frq_tab->n_valid; i++)
1432 struct freq *f = &frq_tab->valid[i];
1433 if (f->count > catchart->x_max)
1436 if (f->count < catchart->x_min)
1439 slices[n_slices] = f;
1444 if (catchart->include_missing)
1446 for (i = 0; i < frq_tab->n_missing; i++)
1448 const struct freq *f = &frq_tab->missing[i];
1451 slices[n_slices] = xmalloc (sizeof (struct freq));
1452 slices[n_slices]->values[0] = f->values[0];
1455 slices[n_slices]->count += f->count;
1460 *n_slicesp = n_slices;
1467 do_piechart(const struct frq_chart *pie, const struct variable *var,
1468 const struct freq_tab *frq_tab)
1471 struct freq *slices = pick_cat_counts (pie, frq_tab, &n_slices);
1474 msg (SW, _("Omitting pie chart for %s, which has only %d unique values."),
1475 var_get_name (var), n_slices);
1476 else if (n_slices > 50)
1477 msg (SW, _("Omitting pie chart for %s, which has over 50 unique values."),
1478 var_get_name (var));
1480 chart_item_submit (piechart_create (var, slices, n_slices));
1487 do_barchart(const struct frq_chart *bar, const struct variable **var,
1488 const struct freq_tab *frq_tab)
1491 struct freq **slices = pick_cat_counts_ptr (bar, frq_tab, &n_slices);
1493 chart_item_submit (barchart_create (var, 1,
1494 (bar->y_scale == FRQ_FREQ) ? _("Count") : _("Percent"),
1495 (bar->y_scale == FRQ_PERCENT),
1501 /* Calculates all the pertinent statistics for VF, putting them in array
1504 calc_stats (const struct frq_proc *frq, const struct var_freqs *vf,
1505 double d[FRQ_ST_count])
1507 const struct freq_tab *ft = &vf->tab;
1508 double W = ft->valid_cases;
1509 const struct freq *f;
1511 int most_often = -1;
1512 double X_mode = SYSMIS;
1514 /* Calculate the mode. */
1515 for (f = ft->valid; f < ft->missing; f++)
1517 if (most_often < f->count)
1519 most_often = f->count;
1520 X_mode = f->values[0].f;
1522 else if (most_often == f->count)
1524 /* A duplicate mode is undefined.
1525 FIXME: keep track of *all* the modes. */
1530 /* Calculate moments. */
1531 m = moments_create (MOMENT_KURTOSIS);
1532 for (f = ft->valid; f < ft->missing; f++)
1533 moments_pass_one (m, f->values[0].f, f->count);
1534 for (f = ft->valid; f < ft->missing; f++)
1535 moments_pass_two (m, f->values[0].f, f->count);
1536 moments_calculate (m, NULL, &d[FRQ_ST_MEAN], &d[FRQ_ST_VARIANCE],
1537 &d[FRQ_ST_SKEWNESS], &d[FRQ_ST_KURTOSIS]);
1538 moments_destroy (m);
1540 /* Formulae below are taken from _SPSS Statistical Algorithms_. */
1541 if (ft->n_valid > 0)
1543 d[FRQ_ST_MINIMUM] = ft->valid[0].values[0].f;
1544 d[FRQ_ST_MAXIMUM] = ft->valid[ft->n_valid - 1].values[0].f;
1545 d[FRQ_ST_RANGE] = d[FRQ_ST_MAXIMUM] - d[FRQ_ST_MINIMUM];
1549 d[FRQ_ST_MINIMUM] = SYSMIS;
1550 d[FRQ_ST_MAXIMUM] = SYSMIS;
1551 d[FRQ_ST_RANGE] = SYSMIS;
1553 d[FRQ_ST_MODE] = X_mode;
1554 d[FRQ_ST_SUM] = d[FRQ_ST_MEAN] * W;
1555 d[FRQ_ST_STDDEV] = sqrt (d[FRQ_ST_VARIANCE]);
1556 d[FRQ_ST_SEMEAN] = d[FRQ_ST_STDDEV] / sqrt (W);
1557 d[FRQ_ST_SESKEWNESS] = calc_seskew (W);
1558 d[FRQ_ST_SEKURTOSIS] = calc_sekurt (W);
1559 d[FRQ_ST_MEDIAN] = frq->median ? frq->median->value : SYSMIS;
1563 all_string_variables (const struct frq_proc *frq)
1565 for (size_t i = 0; i < frq->n_vars; i++)
1566 if (var_is_numeric (frq->vars[i].var))
1572 /* Displays a table of all the statistics requested. */
1574 dump_statistics (const struct frq_proc *frq, const struct variable *wv)
1576 if (all_string_variables (frq))
1579 struct pivot_table *table = pivot_table_create (N_("Statistics"));
1580 pivot_table_set_weight_var (table, wv);
1582 struct pivot_dimension *variables
1583 = pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Variables"));
1585 struct pivot_dimension *statistics = pivot_dimension_create (
1586 table, PIVOT_AXIS_ROW, N_("Statistics"));
1587 struct pivot_category *n = pivot_category_create_group (
1588 statistics->root, N_("N"));
1589 pivot_category_create_leaves (n,
1590 N_("Valid"), PIVOT_RC_COUNT,
1591 N_("Missing"), PIVOT_RC_COUNT);
1592 for (int i = 0; i < FRQ_ST_count; i++)
1593 if (frq->stats & BIT_INDEX (i))
1594 pivot_category_create_leaf (statistics->root,
1595 pivot_value_new_text (st_name[i]));
1596 struct pivot_category *percentiles = NULL;
1597 for (size_t i = 0; i < frq->n_percentiles; i++)
1599 const struct percentile *pc = &frq->percentiles[i];
1605 percentiles = pivot_category_create_group (
1606 statistics->root, N_("Percentiles"));
1607 pivot_category_create_leaf (percentiles, pivot_value_new_integer (
1611 for (size_t i = 0; i < frq->n_vars; i++)
1613 struct var_freqs *vf = &frq->vars[i];
1614 if (var_is_alpha (vf->var))
1617 const struct freq_tab *ft = &vf->tab;
1619 int var_idx = pivot_category_create_leaf (
1620 variables->root, pivot_value_new_variable (vf->var));
1623 pivot_table_put2 (table, var_idx, row++,
1624 pivot_value_new_number (ft->valid_cases));
1625 pivot_table_put2 (table, var_idx, row++,
1626 pivot_value_new_number (
1627 ft->total_cases - ft->valid_cases));
1629 double stat_values[FRQ_ST_count];
1630 calc_stats (frq, vf, stat_values);
1631 for (int j = 0; j < FRQ_ST_count; j++)
1633 if (!(frq->stats & BIT_INDEX (j)))
1636 union value v = { .f = vf->tab.n_valid ? stat_values[j] : SYSMIS };
1637 struct pivot_value *pv
1638 = (j == FRQ_ST_MODE || j == FRQ_ST_MINIMUM || j == FRQ_ST_MAXIMUM
1639 ? pivot_value_new_var_value (vf->var, &v)
1640 : pivot_value_new_number (v.f));
1641 pivot_table_put2 (table, var_idx, row++, pv);
1644 for (size_t j = 0; j < frq->n_percentiles; j++)
1646 const struct percentile *pc = &frq->percentiles[j];
1650 union value v = { .f = vf->tab.n_valid ? pc->value : SYSMIS };
1651 pivot_table_put2 (table, var_idx, row++,
1652 pivot_value_new_var_value (vf->var, &v));
1656 pivot_table_submit (table);