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"
49 #include "math/histogram.h"
50 #include "math/moments.h"
51 #include "math/chart-geometry.h"
54 #include "output/charts/barchart.h"
55 #include "output/charts/piechart.h"
56 #include "output/charts/plot-hist.h"
57 #include "output/pivot-table.h"
59 #include "gl/minmax.h"
60 #include "gl/xalloc.h"
63 #define _(msgid) gettext (msgid)
64 #define N_(msgid) msgid
66 /* Percentiles to calculate. */
70 double p; /* the %ile to be calculated */
71 bool show; /* True to show this percentile in the statistics box. */
75 ptile_3way (const void *_p1, const void *_p2)
77 const struct percentile *p1 = _p1;
78 const struct percentile *p2 = _p2;
85 if (p1->show > p2->show)
88 return (p1->show < p2->show);
91 return (p1->p > p2->p);
115 /* Array indices for STATISTICS subcommand. */
135 /* Description of statistics. */
136 static const char *st_name[FRQ_ST_count] =
156 struct hmap data; /* Hash table for accumulating counts. */
157 struct freq *valid; /* Valid freqs. */
158 int n_valid; /* Number of total freqs. */
159 const struct dictionary *dict; /* Source of entries in the table. */
161 struct freq *missing; /* Missing freqs. */
162 int n_missing; /* Number of missing freqs. */
165 double total_cases; /* Sum of weights of all cases. */
166 double valid_cases; /* Sum of weights of valid cases. */
171 double x_min; /* X axis minimum value. */
172 double x_max; /* X axis maximum value. */
173 int y_scale; /* Y axis scale: FRQ_FREQ or FRQ_PERCENT. */
175 /* Histograms only. */
176 double y_max; /* Y axis maximum value. */
177 bool draw_normal; /* Whether to draw normal curve. */
179 /* Pie charts only. */
180 bool include_missing; /* Whether to include missing values. */
183 /* Per-variable frequency data. */
186 const struct variable *var;
188 /* Freqency table. */
189 struct freq_tab tab; /* Frequencies table to use. */
192 double stat[FRQ_ST_count];
195 /* Variable attributes. */
201 struct var_freqs *vars;
204 /* Percentiles to calculate and possibly display. */
205 struct percentile *percentiles;
207 size_t n_percentiles;
209 /* Frequency table display. */
210 long int max_categories; /* Maximum categories to show. */
211 int sort; /* FRQ_AVALUE or FRQ_DVALUE
212 or FRQ_AFREQ or FRQ_DFREQ. */
214 /* Statistics; number of statistics. */
218 /* Histogram and pie chart settings. */
219 struct frq_chart *hist, *pie, *bar;
225 struct freq_compare_aux
231 bool ascending_value;
234 static void calc_stats (const struct frq_proc *,
235 const struct var_freqs *, double d[FRQ_ST_count]);
237 static void do_piechart(const struct frq_chart *pie,
238 const struct variable *var,
239 const struct freq_tab *frq_tab);
241 static void do_barchart(const struct frq_chart *bar,
242 const struct variable **var,
243 const struct freq_tab *frq_tab);
245 static struct frq_stats_table *frq_stats_table_submit (
246 struct frq_stats_table *, const struct frq_proc *,
247 const struct dictionary *, const struct variable *wv);
248 static void frq_stats_table_destroy (struct frq_stats_table *);
251 compare_freq (const void *a_, const void *b_, const void *aux_)
253 const struct freq_compare_aux *aux = aux_;
254 const struct freq *a = a_;
255 const struct freq *b = b_;
257 if (aux->by_freq && a->count != b->count)
259 int cmp = a->count > b->count ? 1 : -1;
260 return aux->ascending_freq ? cmp : -cmp;
264 int cmp = value_compare_3way (a->values, b->values, aux->width);
265 return aux->ascending_value ? cmp : -cmp;
269 /* Create a gsl_histogram from a freq_tab */
270 static struct histogram *freq_tab_to_hist (const struct frq_proc *,
271 const struct var_freqs *);
274 put_freq_row (struct pivot_table *table, int var_idx,
275 double frequency, double percent,
276 double valid_percent, double cum_percent)
278 double entries[] = { frequency, percent, valid_percent, cum_percent };
279 for (size_t i = 0; i < sizeof entries / sizeof *entries; i++)
280 if (entries[i] != SYSMIS)
281 pivot_table_put2 (table, i, var_idx,
282 pivot_value_new_number (entries[i]));
285 /* Displays a full frequency table for variable V. */
287 dump_freq_table (const struct var_freqs *vf, const struct variable *wv)
289 const struct freq_tab *ft = &vf->tab;
291 struct pivot_table *table = pivot_table_create__ (pivot_value_new_variable (
292 vf->var), "Frequencies");
293 pivot_table_set_weight_var (table, wv);
295 pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Statistics"),
296 N_("Frequency"), PIVOT_RC_COUNT,
297 N_("Percent"), PIVOT_RC_PERCENT,
298 N_("Valid Percent"), PIVOT_RC_PERCENT,
299 N_("Cumulative Percent"), PIVOT_RC_PERCENT);
301 struct pivot_dimension *variable = pivot_dimension_create__ (
302 table, PIVOT_AXIS_ROW, pivot_value_new_variable (vf->var));
304 double cum_freq = 0.0;
305 double cum_percent = 0.0;
306 struct pivot_category *valid = NULL;
307 for (const struct freq *f = ft->valid; f < ft->missing; f++)
309 cum_freq += f->count;
310 double valid_percent = f->count / ft->valid_cases * 100.0;
311 cum_percent += valid_percent;
314 valid = pivot_category_create_group (variable->root, N_("Valid"));
315 int var_idx = pivot_category_create_leaf (
316 valid, pivot_value_new_var_value (vf->var, &f->values[0]));
317 put_freq_row (table, var_idx, f->count,
318 f->count / ft->total_cases * 100.0,
319 valid_percent, cum_percent);
322 struct pivot_category *missing = NULL;
323 size_t n_categories = ft->n_valid + ft->n_missing;
324 for (const struct freq *f = ft->missing; f < &ft->valid[n_categories]; f++)
326 cum_freq += f->count;
329 missing = pivot_category_create_group (variable->root, N_("Missing"));
330 int var_idx = pivot_category_create_leaf (
331 missing, pivot_value_new_var_value (vf->var, &f->values[0]));
332 put_freq_row (table, var_idx, f->count,
333 f->count / ft->total_cases * 100.0, SYSMIS, SYSMIS);
336 int var_idx = pivot_category_create_leaf (
337 variable->root, pivot_value_new_text (N_("Total")));
338 put_freq_row (table, var_idx, cum_freq, cum_percent, SYSMIS, SYSMIS);
340 pivot_table_submit (table);
343 /* Statistical display. */
346 calc_percentile (double p, double valid_cases, double x1, double x2)
350 s = (settings_get_algorithm () != COMPATIBLE
351 ? modf ((valid_cases - 1) * p, &dummy)
352 : modf ((valid_cases + 1) * p - 1, &dummy));
354 return x1 + (x2 - x1) * s;
357 /* Calculates all of the percentiles for VF within FRQ. */
359 calc_percentiles (const struct frq_proc *frq, struct var_freqs *vf)
361 if (!frq->n_percentiles)
364 if (!vf->percentiles)
365 vf->percentiles = xnmalloc (frq->n_percentiles, sizeof *vf->percentiles);
367 const struct freq_tab *ft = &vf->tab;
368 const double W = ft->valid_cases;
372 for (const struct freq *f = ft->valid; f < ft->missing; f++)
375 for (; idx < frq->n_percentiles; idx++)
377 struct percentile *pc = &frq->percentiles[idx];
380 tp = (settings_get_algorithm () == ENHANCED
382 : (W + 1) * pc->p - 1);
387 if (tp + 1 < rank || f + 1 >= ft->missing)
388 vf->percentiles[idx] = f->values[0].f;
390 vf->percentiles[idx] = calc_percentile (pc->p, W, f->values[0].f,
394 for (; idx < frq->n_percentiles; idx++)
395 vf->percentiles[idx] = (ft->n_valid > 0
396 ? ft->valid[ft->n_valid - 1].values[0].f
400 /* Returns true iff the value in struct freq F is non-missing
403 not_missing (const void *f_, const void *v_)
405 const struct freq *f = f_;
406 const struct variable *v = v_;
408 return !var_is_value_missing (v, f->values);
412 /* Summarizes the frequency table data for variable V. */
414 postprocess_freq_tab (const struct frq_proc *frq, struct var_freqs *vf)
416 struct freq_tab *ft = &vf->tab;
417 struct freq_compare_aux aux;
419 struct freq *freqs, *f;
422 /* Extract data from hash table. */
423 count = hmap_count (&ft->data);
424 freqs = freq_hmap_extract (&ft->data);
426 /* Put data into ft. */
428 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, vf->var);
429 ft->missing = freqs + ft->n_valid;
430 ft->n_missing = count - ft->n_valid;
433 aux.by_freq = frq->sort == FRQ_AFREQ || frq->sort == FRQ_DFREQ;
434 aux.ascending_freq = frq->sort != FRQ_DFREQ;
435 aux.width = vf->width;
436 aux.ascending_value = frq->sort != FRQ_DVALUE;
437 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare_freq, &aux);
438 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare_freq, &aux);
440 /* Summary statistics. */
441 ft->valid_cases = 0.0;
442 for(i = 0 ; i < ft->n_valid ; ++i)
445 ft->valid_cases += f->count;
449 ft->total_cases = ft->valid_cases ;
450 for(i = 0 ; i < ft->n_missing ; ++i)
453 ft->total_cases += f->count;
458 /* Add data from case C to the frequency table. */
460 calc (struct frq_proc *frq, const struct ccase *c, const struct dataset *ds)
462 double weight = dict_get_case_weight (dataset_dict (ds), c, &frq->warn);
465 for (i = 0; i < frq->n_vars; i++)
467 struct var_freqs *vf = &frq->vars[i];
468 const union value *value = case_data (c, vf->var);
469 size_t hash = value_hash (value, vf->width, 0);
472 f = freq_hmap_search (&vf->tab.data, value, vf->width, hash);
474 f = freq_hmap_insert (&vf->tab.data, value, vf->width, hash);
481 output_splits_once (bool *need_splits, const struct dataset *ds,
482 const struct ccase *c)
486 output_split_file_values (ds, c);
487 *need_splits = false;
491 /* Finishes up with the variables after frequencies have been
492 calculated. Displays statistics, percentiles, ... */
493 static struct frq_stats_table *
494 postcalc (struct frq_proc *frq, const struct dataset *ds,
495 struct ccase *first, struct frq_stats_table *fst)
497 const struct dictionary *dict = dataset_dict (ds);
498 const struct variable *wv = dict_get_weight (dict);
500 for (size_t i = 0; i < frq->n_vars; i++)
502 struct var_freqs *vf = &frq->vars[i];
503 postprocess_freq_tab (frq, vf);
504 calc_percentiles (frq, vf);
507 enum split_type st = dict_get_split_type (dict);
508 bool need_splits = true;
511 if (st != SPLIT_LAYERED)
512 output_splits_once (&need_splits, ds, first);
513 fst = frq_stats_table_submit (fst, frq, dict, wv);
516 for (size_t i = 0; i < frq->n_vars; i++)
518 struct var_freqs *vf = &frq->vars[i];
520 /* Frequencies tables. */
521 if (vf->tab.n_valid + vf->tab.n_missing <= frq->max_categories)
523 output_splits_once (&need_splits, ds, first);
524 dump_freq_table (vf, wv);
527 if (frq->hist && var_is_numeric (vf->var) && vf->tab.n_valid > 0)
529 double d[FRQ_ST_count];
530 struct histogram *histogram;
532 calc_stats (frq, vf, d);
534 histogram = freq_tab_to_hist (frq, vf);
538 output_splits_once (&need_splits, ds, first);
539 chart_submit (histogram_chart_create (
540 histogram->gsl_hist, var_to_string(vf->var),
544 frq->hist->draw_normal));
546 statistic_destroy (&histogram->parent);
552 output_splits_once (&need_splits, ds, first);
553 do_piechart(frq->pie, vf->var, &vf->tab);
558 output_splits_once (&need_splits, ds, first);
559 do_barchart(frq->bar, &vf->var, &vf->tab);
562 free (vf->tab.valid);
563 freq_hmap_destroy (&vf->tab.data, vf->width);
570 frq_run (struct frq_proc *frq, struct dataset *ds)
572 struct frq_stats_table *fst = NULL;
573 struct casegrouper *grouper = casegrouper_create_splits (proc_open (ds),
575 struct casereader *group;
576 while (casegrouper_get_next_group (grouper, &group))
578 for (size_t i = 0; i < frq->n_vars; i++)
579 hmap_init (&frq->vars[i].tab.data);
581 struct ccase *first = casereader_peek (group, 0);
584 for (; (c = casereader_read (group)) != NULL; case_unref (c))
586 fst = postcalc (frq, ds, first, fst);
587 casereader_destroy (group);
591 frq_stats_table_destroy (fst);
592 casegrouper_destroy (grouper);
597 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
601 const struct variable **vars = NULL;
603 bool sbc_barchart = false;
604 bool sbc_piechart = false;
605 bool sbc_histogram = false;
607 double pie_min = -DBL_MAX;
608 double pie_max = DBL_MAX;
609 bool pie_missing = true;
611 double bar_min = -DBL_MAX;
612 double bar_max = DBL_MAX;
613 bool bar_freq = true;
615 double hi_min = -DBL_MAX;
616 double hi_max = DBL_MAX;
617 int hi_scale = FRQ_FREQ;
618 int hi_freq = INT_MIN;
619 int hi_pcnt = INT_MIN;
620 int hi_norm = FRQ_NONORMAL;
622 frq.sort = FRQ_AVALUE;
627 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
628 | BIT_INDEX (FRQ_ST_STDDEV)
629 | BIT_INDEX (FRQ_ST_MINIMUM)
630 | BIT_INDEX (FRQ_ST_MAXIMUM);
634 frq.max_categories = LONG_MAX;
636 frq.percentiles = NULL;
637 frq.n_percentiles = 0;
645 /* Accept an optional, completely pointless "/VARIABLES=" */
646 lex_match (lexer, T_SLASH);
647 if (lex_match_id (lexer, "VARIABLES"))
649 if (! lex_force_match (lexer, T_EQUALS))
653 if (!parse_variables_const (lexer, dataset_dict (ds),
659 frq.vars = xcalloc (frq.n_vars, sizeof (*frq.vars));
660 for (i = 0; i < frq.n_vars; ++i)
662 frq.vars[i].var = vars[i];
663 frq.vars[i].width = var_get_width (vars[i]);
666 while (lex_token (lexer) != T_ENDCMD)
668 lex_match (lexer, T_SLASH);
670 if (lex_match_id (lexer, "STATISTICS"))
672 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
673 | BIT_INDEX (FRQ_ST_STDDEV)
674 | BIT_INDEX (FRQ_ST_MINIMUM)
675 | BIT_INDEX (FRQ_ST_MAXIMUM);
679 if (lex_match (lexer, T_EQUALS))
685 while (lex_token (lexer) != T_ENDCMD
686 && lex_token (lexer) != T_SLASH)
688 if (lex_match_id (lexer, "DEFAULT"))
690 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
691 | BIT_INDEX (FRQ_ST_STDDEV)
692 | BIT_INDEX (FRQ_ST_MINIMUM)
693 | BIT_INDEX (FRQ_ST_MAXIMUM);
697 else if (lex_match_id (lexer, "MEAN"))
699 frq.stats |= BIT_INDEX (FRQ_ST_MEAN);
702 else if (lex_match_id (lexer, "SEMEAN"))
704 frq.stats |= BIT_INDEX (FRQ_ST_SEMEAN);
707 else if (lex_match_id (lexer, "MEDIAN"))
709 frq.stats |= BIT_INDEX (FRQ_ST_MEDIAN);
712 else if (lex_match_id (lexer, "MODE"))
714 frq.stats |= BIT_INDEX (FRQ_ST_MODE);
717 else if (lex_match_id (lexer, "STDDEV"))
719 frq.stats |= BIT_INDEX (FRQ_ST_STDDEV);
722 else if (lex_match_id (lexer, "VARIANCE"))
724 frq.stats |= BIT_INDEX (FRQ_ST_VARIANCE);
727 else if (lex_match_id (lexer, "KURTOSIS"))
729 frq.stats |= BIT_INDEX (FRQ_ST_KURTOSIS);
732 else if (lex_match_id (lexer, "SKEWNESS"))
734 frq.stats |= BIT_INDEX (FRQ_ST_SKEWNESS);
737 else if (lex_match_id (lexer, "RANGE"))
739 frq.stats |= BIT_INDEX (FRQ_ST_RANGE);
742 else if (lex_match_id (lexer, "MINIMUM"))
744 frq.stats |= BIT_INDEX (FRQ_ST_MINIMUM);
747 else if (lex_match_id (lexer, "MAXIMUM"))
749 frq.stats |= BIT_INDEX (FRQ_ST_MAXIMUM);
752 else if (lex_match_id (lexer, "SUM"))
754 frq.stats |= BIT_INDEX (FRQ_ST_SUM);
757 else if (lex_match_id (lexer, "SESKEWNESS"))
759 frq.stats |= BIT_INDEX (FRQ_ST_SESKEWNESS);
762 else if (lex_match_id (lexer, "SEKURTOSIS"))
764 frq.stats |= BIT_INDEX (FRQ_ST_SEKURTOSIS);
767 else if (lex_match_id (lexer, "NONE"))
772 else if (lex_match (lexer, T_ALL))
775 frq.n_stats = FRQ_ST_count;
779 lex_error (lexer, NULL);
784 else if (lex_match_id (lexer, "PERCENTILES"))
786 lex_match (lexer, T_EQUALS);
787 while (lex_token (lexer) != T_ENDCMD
788 && lex_token (lexer) != T_SLASH)
790 if (lex_force_num (lexer))
793 xrealloc (frq.percentiles,
794 (frq.n_percentiles + 1)
795 * sizeof (*frq.percentiles));
796 frq.percentiles[frq.n_percentiles].p = lex_number (lexer) / 100.0;
797 frq.percentiles[frq.n_percentiles].show = true;
803 lex_error (lexer, NULL);
806 lex_match (lexer, T_COMMA);
809 else if (lex_match_id (lexer, "FORMAT"))
811 lex_match (lexer, T_EQUALS);
812 while (lex_token (lexer) != T_ENDCMD
813 && lex_token (lexer) != T_SLASH)
815 if (lex_match_id (lexer, "TABLE"))
818 else if (lex_match_id (lexer, "NOTABLE"))
820 frq.max_categories = 0;
822 else if (lex_match_id (lexer, "LIMIT"))
824 if (!lex_force_match (lexer, T_LPAREN)
825 || !lex_force_int_range (lexer, "LIMIT", 0, INT_MAX))
828 frq.max_categories = lex_integer (lexer);
831 if (!lex_force_match (lexer, T_RPAREN))
834 else if (lex_match_id (lexer, "AVALUE"))
836 frq.sort = FRQ_AVALUE;
838 else if (lex_match_id (lexer, "DVALUE"))
840 frq.sort = FRQ_DVALUE;
842 else if (lex_match_id (lexer, "AFREQ"))
844 frq.sort = FRQ_AFREQ;
846 else if (lex_match_id (lexer, "DFREQ"))
848 frq.sort = FRQ_DFREQ;
852 lex_error (lexer, NULL);
857 else if (lex_match_id (lexer, "NTILES"))
859 lex_match (lexer, T_EQUALS);
861 if (lex_force_int_range (lexer, "NTILES", 0, INT_MAX))
863 int n = lex_integer (lexer);
865 for (int i = 0; i < n + 1; ++i)
868 xrealloc (frq.percentiles,
869 (frq.n_percentiles + 1)
870 * sizeof (*frq.percentiles));
871 frq.percentiles[frq.n_percentiles].p =
873 frq.percentiles[frq.n_percentiles].show = true;
880 lex_error (lexer, NULL);
884 else if (lex_match_id (lexer, "ALGORITHM"))
886 lex_match (lexer, T_EQUALS);
888 if (lex_match_id (lexer, "COMPATIBLE"))
890 settings_set_cmd_algorithm (COMPATIBLE);
892 else if (lex_match_id (lexer, "ENHANCED"))
894 settings_set_cmd_algorithm (ENHANCED);
898 lex_error (lexer, NULL);
902 else if (lex_match_id (lexer, "HISTOGRAM"))
904 lex_match (lexer, T_EQUALS);
905 sbc_histogram = true;
907 while (lex_token (lexer) != T_ENDCMD
908 && lex_token (lexer) != T_SLASH)
910 if (lex_match_id (lexer, "NORMAL"))
912 hi_norm = FRQ_NORMAL;
914 else if (lex_match_id (lexer, "NONORMAL"))
916 hi_norm = FRQ_NONORMAL;
918 else if (lex_match_id (lexer, "FREQ"))
921 if (lex_match (lexer, T_LPAREN))
923 if (lex_force_int_range (lexer, "FREQ", 1, INT_MAX))
925 hi_freq = lex_integer (lexer);
927 if (! lex_force_match (lexer, T_RPAREN))
932 else if (lex_match_id (lexer, "PERCENT"))
934 hi_scale = FRQ_PERCENT;
935 if (lex_match (lexer, T_LPAREN))
937 if (lex_force_int_range (lexer, "PERCENT", 1, INT_MAX))
939 hi_pcnt = lex_integer (lexer);
941 if (! lex_force_match (lexer, T_RPAREN))
946 else if (lex_match_id (lexer, "MINIMUM"))
948 if (! lex_force_match (lexer, T_LPAREN))
950 if (lex_force_num (lexer))
952 hi_min = lex_number (lexer);
955 if (! lex_force_match (lexer, T_RPAREN))
958 else if (lex_match_id (lexer, "MAXIMUM"))
960 if (! lex_force_match (lexer, T_LPAREN))
962 if (lex_force_num (lexer))
964 hi_max = lex_number (lexer);
967 if (! lex_force_match (lexer, T_RPAREN))
972 lex_error (lexer, NULL);
977 else if (lex_match_id (lexer, "PIECHART"))
979 lex_match (lexer, T_EQUALS);
980 while (lex_token (lexer) != T_ENDCMD
981 && lex_token (lexer) != T_SLASH)
983 if (lex_match_id (lexer, "MINIMUM"))
985 if (! lex_force_match (lexer, T_LPAREN))
987 if (lex_force_num (lexer))
989 pie_min = lex_number (lexer);
992 if (! lex_force_match (lexer, T_RPAREN))
995 else if (lex_match_id (lexer, "MAXIMUM"))
997 if (! lex_force_match (lexer, T_LPAREN))
999 if (lex_force_num (lexer))
1001 pie_max = lex_number (lexer);
1004 if (! lex_force_match (lexer, T_RPAREN))
1007 else if (lex_match_id (lexer, "MISSING"))
1011 else if (lex_match_id (lexer, "NOMISSING"))
1013 pie_missing = false;
1017 lex_error (lexer, NULL);
1021 sbc_piechart = true;
1023 else if (lex_match_id (lexer, "BARCHART"))
1025 lex_match (lexer, T_EQUALS);
1026 while (lex_token (lexer) != T_ENDCMD
1027 && lex_token (lexer) != T_SLASH)
1029 if (lex_match_id (lexer, "MINIMUM"))
1031 if (! lex_force_match (lexer, T_LPAREN))
1033 if (lex_force_num (lexer))
1035 bar_min = lex_number (lexer);
1038 if (! lex_force_match (lexer, T_RPAREN))
1041 else if (lex_match_id (lexer, "MAXIMUM"))
1043 if (! lex_force_match (lexer, T_LPAREN))
1045 if (lex_force_num (lexer))
1047 bar_max = lex_number (lexer);
1050 if (! lex_force_match (lexer, T_RPAREN))
1053 else if (lex_match_id (lexer, "FREQ"))
1055 if (lex_match (lexer, T_LPAREN))
1057 if (lex_force_num (lexer))
1062 if (! lex_force_match (lexer, T_RPAREN))
1067 else if (lex_match_id (lexer, "PERCENT"))
1069 if (lex_match (lexer, T_LPAREN))
1071 if (lex_force_num (lexer))
1076 if (! lex_force_match (lexer, T_RPAREN))
1083 lex_error (lexer, NULL);
1087 sbc_barchart = true;
1089 else if (lex_match_id (lexer, "MISSING"))
1091 lex_match (lexer, T_EQUALS);
1093 while (lex_token (lexer) != T_ENDCMD
1094 && lex_token (lexer) != T_SLASH)
1096 if (lex_match_id (lexer, "EXCLUDE"))
1099 else if (lex_match_id (lexer, "INCLUDE"))
1104 lex_error (lexer, NULL);
1109 else if (lex_match_id (lexer, "ORDER"))
1111 lex_match (lexer, T_EQUALS);
1112 if (!lex_match_id (lexer, "ANALYSIS"))
1113 lex_match_id (lexer, "VARIABLE");
1117 lex_error (lexer, NULL);
1122 if (frq.stats & BIT_INDEX (FRQ_ST_MEDIAN))
1125 xrealloc (frq.percentiles,
1126 (frq.n_percentiles + 1)
1127 * sizeof (*frq.percentiles));
1129 frq.percentiles[frq.n_percentiles].p = 0.50;
1130 frq.percentiles[frq.n_percentiles].show = false;
1132 frq.n_percentiles++;
1136 /* Figure out which charts the user requested. */
1141 struct frq_chart *hist;
1143 hist = frq.hist = xmalloc (sizeof *frq.hist);
1144 hist->x_min = hi_min;
1145 hist->x_max = hi_max;
1146 hist->y_scale = hi_scale;
1147 hist->y_max = hi_scale == FRQ_FREQ ? hi_freq : hi_pcnt;
1148 hist->draw_normal = hi_norm != FRQ_NONORMAL;
1149 hist->include_missing = false;
1151 if (hist->x_min != SYSMIS && hist->x_max != SYSMIS
1152 && hist->x_min >= hist->x_max)
1154 msg (SE, _("%s for histogram must be greater than or equal to %s, "
1155 "but %s was specified as %.15g and %s as %.15g. "
1156 "%s and %s will be ignored."),
1161 hist->x_min = hist->x_max = SYSMIS;
1165 xrealloc (frq.percentiles,
1166 (frq.n_percentiles + 2)
1167 * sizeof (*frq.percentiles));
1169 frq.percentiles[frq.n_percentiles].p = 0.25;
1170 frq.percentiles[frq.n_percentiles].show = false;
1172 frq.percentiles[frq.n_percentiles + 1].p = 0.75;
1173 frq.percentiles[frq.n_percentiles + 1].show = false;
1175 frq.n_percentiles+=2;
1180 frq.bar = xmalloc (sizeof *frq.bar);
1181 frq.bar->x_min = bar_min;
1182 frq.bar->x_max = bar_max;
1183 frq.bar->include_missing = false;
1184 frq.bar->y_scale = bar_freq ? FRQ_FREQ : FRQ_PERCENT;
1189 struct frq_chart *pie;
1191 pie = frq.pie = xmalloc (sizeof *frq.pie);
1192 pie->x_min = pie_min;
1193 pie->x_max = pie_max;
1194 pie->include_missing = pie_missing;
1196 if (pie->x_min != SYSMIS && pie->x_max != SYSMIS
1197 && pie->x_min >= pie->x_max)
1199 msg (SE, _("%s for pie chart must be greater than or equal to %s, "
1200 "but %s was specified as %.15g and %s as %.15g. "
1201 "%s and %s will be ignored."),
1206 pie->x_min = pie->x_max = SYSMIS;
1213 double previous_p = -1;
1214 qsort (frq.percentiles, frq.n_percentiles,
1215 sizeof (*frq.percentiles),
1218 for (i = o = 0; i < frq.n_percentiles; ++i)
1220 if (frq.percentiles[i].p != previous_p)
1222 frq.percentiles[o].p = frq.percentiles[i].p;
1223 frq.percentiles[o].show = frq.percentiles[i].show;
1226 else if (frq.percentiles[i].show &&
1227 !frq.percentiles[o].show)
1229 frq.percentiles[o].show = true;
1231 previous_p = frq.percentiles[i].p;
1234 frq.n_percentiles = o;
1236 frq.median_idx = SIZE_MAX;
1237 for (i = 0; i < frq.n_percentiles; i++)
1238 if (frq.percentiles[i].p == 0.5)
1248 for (size_t i = 0; i < frq.n_vars; i++)
1249 free (frq.vars[i].percentiles);
1254 free (frq.percentiles);
1261 for (size_t i = 0; i < frq.n_vars; i++)
1262 free (frq.vars[i].percentiles);
1267 free (frq.percentiles);
1273 calculate_iqr (const struct frq_proc *frq, const struct var_freqs *vf)
1279 /* This cannot work unless the 25th and 75th percentile are calculated */
1280 assert (frq->n_percentiles >= 2);
1281 for (i = 0; i < frq->n_percentiles; i++)
1283 struct percentile *pc = &frq->percentiles[i];
1285 if (fabs (0.25 - pc->p) < DBL_EPSILON)
1286 q1 = vf->percentiles[i];
1287 else if (fabs (0.75 - pc->p) < DBL_EPSILON)
1288 q3 = vf->percentiles[i];
1291 return q1 == SYSMIS || q3 == SYSMIS ? SYSMIS : q3 - q1;
1295 chart_includes_value (const struct frq_chart *chart,
1296 const struct variable *var,
1297 const union value *value)
1299 if (!chart->include_missing && var_is_value_missing (var, value))
1302 if (var_is_numeric (var)
1303 && ((chart->x_min != SYSMIS && value->f < chart->x_min)
1304 || (chart->x_max != SYSMIS && value->f > chart->x_max)))
1310 /* Create a gsl_histogram from a freq_tab */
1311 static struct histogram *
1312 freq_tab_to_hist (const struct frq_proc *frq, const struct var_freqs *vf)
1314 /* Find out the extremes of the x value, within the range to be included in
1315 the histogram, and sum the total frequency of those values. */
1316 double x_min = DBL_MAX;
1317 double x_max = -DBL_MAX;
1318 double valid_freq = 0;
1319 for (int i = 0; i < vf->tab.n_valid; i++)
1321 const struct freq *f = &vf->tab.valid[i];
1322 if (chart_includes_value (frq->hist, vf->var, f->values))
1324 x_min = MIN (x_min, f->values[0].f);
1325 x_max = MAX (x_max, f->values[0].f);
1326 valid_freq += f->count;
1330 if (valid_freq <= 0)
1333 double iqr = calculate_iqr (frq, vf);
1337 ? 2 * iqr / pow (valid_freq, 1.0 / 3.0) /* Freedman-Diaconis. */
1338 : (x_max - x_min) / (1 + log2 (valid_freq))); /* Sturges */
1340 struct histogram *histogram = histogram_create (bin_width, x_min, x_max);
1341 if (histogram == NULL)
1344 for (int i = 0; i < vf->tab.n_valid; i++)
1346 const struct freq *f = &vf->tab.valid[i];
1347 if (chart_includes_value (frq->hist, vf->var, f->values))
1348 histogram_add (histogram, f->values[0].f, f->count);
1355 /* Allocate an array of struct freqs and fill them from the data in FRQ_TAB,
1356 according to the parameters of CATCHART
1357 N_SLICES will contain the number of slices allocated.
1358 The caller is responsible for freeing slices
1360 static struct freq *
1361 pick_cat_counts (const struct frq_chart *catchart,
1362 const struct freq_tab *frq_tab,
1367 struct freq *slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1369 for (i = 0; i < frq_tab->n_valid; i++)
1371 const struct freq *f = &frq_tab->valid[i];
1372 if (f->count > catchart->x_max)
1375 if (f->count < catchart->x_min)
1378 slices[n_slices] = *f;
1383 if (catchart->include_missing)
1385 for (i = 0; i < frq_tab->n_missing; i++)
1387 const struct freq *f = &frq_tab->missing[i];
1388 slices[n_slices].count += f->count;
1391 slices[n_slices].values[0] = f->values[0];
1394 if (frq_tab->n_missing > 0)
1398 *n_slicesp = n_slices;
1403 /* Allocate an array of struct freqs and fill them from the data in FRQ_TAB,
1404 according to the parameters of CATCHART
1405 N_SLICES will contain the number of slices allocated.
1406 The caller is responsible for freeing slices
1408 static struct freq **
1409 pick_cat_counts_ptr (const struct frq_chart *catchart,
1410 const struct freq_tab *frq_tab,
1415 struct freq **slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1417 for (i = 0; i < frq_tab->n_valid; i++)
1419 struct freq *f = &frq_tab->valid[i];
1420 if (f->count > catchart->x_max)
1423 if (f->count < catchart->x_min)
1426 slices[n_slices] = f;
1431 if (catchart->include_missing)
1433 for (i = 0; i < frq_tab->n_missing; i++)
1435 const struct freq *f = &frq_tab->missing[i];
1438 slices[n_slices] = xmalloc (sizeof (struct freq));
1439 slices[n_slices]->values[0] = f->values[0];
1442 slices[n_slices]->count += f->count;
1447 *n_slicesp = n_slices;
1454 do_piechart(const struct frq_chart *pie, const struct variable *var,
1455 const struct freq_tab *frq_tab)
1458 struct freq *slices = pick_cat_counts (pie, frq_tab, &n_slices);
1461 msg (SW, _("Omitting pie chart for %s, which has only %d unique values."),
1462 var_get_name (var), n_slices);
1463 else if (n_slices > 50)
1464 msg (SW, _("Omitting pie chart for %s, which has over 50 unique values."),
1465 var_get_name (var));
1467 chart_submit (piechart_create (var, slices, n_slices));
1474 do_barchart(const struct frq_chart *bar, const struct variable **var,
1475 const struct freq_tab *frq_tab)
1478 struct freq **slices = pick_cat_counts_ptr (bar, frq_tab, &n_slices);
1481 msg (SW, _("Omitting bar chart, which has no values."));
1483 chart_submit (barchart_create (var, 1,
1484 (bar->y_scale == FRQ_FREQ) ? _("Count") : _("Percent"),
1485 (bar->y_scale == FRQ_PERCENT),
1491 /* Calculates all the pertinent statistics for VF, putting them in array
1494 calc_stats (const struct frq_proc *frq, const struct var_freqs *vf,
1495 double d[FRQ_ST_count])
1497 const struct freq_tab *ft = &vf->tab;
1498 double W = ft->valid_cases;
1499 const struct freq *f;
1501 int most_often = -1;
1502 double X_mode = SYSMIS;
1504 /* Calculate the mode. */
1505 for (f = ft->valid; f < ft->missing; f++)
1507 if (most_often < f->count)
1509 most_often = f->count;
1510 X_mode = f->values[0].f;
1512 else if (most_often == f->count)
1514 /* A duplicate mode is undefined.
1515 FIXME: keep track of *all* the modes. */
1520 /* Calculate moments. */
1521 m = moments_create (MOMENT_KURTOSIS);
1522 for (f = ft->valid; f < ft->missing; f++)
1523 moments_pass_one (m, f->values[0].f, f->count);
1524 for (f = ft->valid; f < ft->missing; f++)
1525 moments_pass_two (m, f->values[0].f, f->count);
1526 moments_calculate (m, NULL, &d[FRQ_ST_MEAN], &d[FRQ_ST_VARIANCE],
1527 &d[FRQ_ST_SKEWNESS], &d[FRQ_ST_KURTOSIS]);
1528 moments_destroy (m);
1530 /* Formulae below are taken from _SPSS Statistical Algorithms_. */
1531 if (ft->n_valid > 0)
1533 d[FRQ_ST_MINIMUM] = ft->valid[0].values[0].f;
1534 d[FRQ_ST_MAXIMUM] = ft->valid[ft->n_valid - 1].values[0].f;
1535 d[FRQ_ST_RANGE] = d[FRQ_ST_MAXIMUM] - d[FRQ_ST_MINIMUM];
1539 d[FRQ_ST_MINIMUM] = SYSMIS;
1540 d[FRQ_ST_MAXIMUM] = SYSMIS;
1541 d[FRQ_ST_RANGE] = SYSMIS;
1543 d[FRQ_ST_MODE] = X_mode;
1544 d[FRQ_ST_SUM] = d[FRQ_ST_MEAN] * W;
1545 d[FRQ_ST_STDDEV] = sqrt (d[FRQ_ST_VARIANCE]);
1546 d[FRQ_ST_SEMEAN] = d[FRQ_ST_STDDEV] / sqrt (W);
1547 d[FRQ_ST_SESKEWNESS] = calc_seskew (W);
1548 d[FRQ_ST_SEKURTOSIS] = calc_sekurt (W);
1549 d[FRQ_ST_MEDIAN] = (frq->median_idx != SIZE_MAX
1550 ? vf->percentiles[frq->median_idx]
1555 all_string_variables (const struct frq_proc *frq)
1557 for (size_t i = 0; i < frq->n_vars; i++)
1558 if (var_is_numeric (frq->vars[i].var))
1564 struct frq_stats_table
1566 struct pivot_table *table;
1567 struct pivot_splits *splits;
1570 /* Displays a table of all the statistics requested. */
1571 static struct frq_stats_table *
1572 frq_stats_table_create (const struct frq_proc *frq,
1573 const struct dictionary *dict,
1574 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_splits *splits = pivot_splits_create (table, dict);
1584 struct pivot_dimension *variables
1585 = pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Variables"));
1586 for (size_t i = 0; i < frq->n_vars; i++)
1587 if (!var_is_alpha (frq->vars[i].var))
1588 pivot_category_create_leaf (variables->root,
1589 pivot_value_new_variable (frq->vars[i].var));
1591 struct pivot_dimension *statistics = pivot_dimension_create (
1592 table, PIVOT_AXIS_ROW, N_("Statistics"));
1593 struct pivot_category *n = pivot_category_create_group (
1594 statistics->root, N_("N"));
1595 pivot_category_create_leaves (n,
1596 N_("Valid"), PIVOT_RC_COUNT,
1597 N_("Missing"), PIVOT_RC_COUNT);
1598 for (int i = 0; i < FRQ_ST_count; i++)
1599 if (frq->stats & BIT_INDEX (i))
1600 pivot_category_create_leaf (statistics->root,
1601 pivot_value_new_text (st_name[i]));
1602 struct pivot_category *percentiles = NULL;
1603 for (size_t i = 0; i < frq->n_percentiles; i++)
1605 const struct percentile *pc = &frq->percentiles[i];
1611 percentiles = pivot_category_create_group (
1612 statistics->root, N_("Percentiles"));
1613 pivot_category_create_leaf (percentiles, pivot_value_new_integer (
1617 struct frq_stats_table *fst = xmalloc (sizeof *fst);
1618 *fst = (struct frq_stats_table) { .table = table, .splits = splits };
1622 static struct frq_stats_table *
1623 frq_stats_table_submit (struct frq_stats_table *fst,
1624 const struct frq_proc *frq,
1625 const struct dictionary *dict,
1626 const struct variable *wv)
1630 fst = frq_stats_table_create (frq, dict, wv);
1636 for (size_t i = 0; i < frq->n_vars; i++)
1638 struct var_freqs *vf = &frq->vars[i];
1639 if (var_is_alpha (vf->var))
1642 const struct freq_tab *ft = &vf->tab;
1645 pivot_table_put2 (fst->table, var_idx, row++,
1646 pivot_value_new_number (ft->valid_cases));
1647 pivot_table_put2 (fst->table, var_idx, row++,
1648 pivot_value_new_number (
1649 ft->total_cases - ft->valid_cases));
1651 double stat_values[FRQ_ST_count];
1652 calc_stats (frq, vf, stat_values);
1653 for (int j = 0; j < FRQ_ST_count; j++)
1655 if (!(frq->stats & BIT_INDEX (j)))
1658 union value v = { .f = vf->tab.n_valid ? stat_values[j] : SYSMIS };
1659 struct pivot_value *pv
1660 = (j == FRQ_ST_MODE || j == FRQ_ST_MINIMUM || j == FRQ_ST_MAXIMUM
1661 ? pivot_value_new_var_value (vf->var, &v)
1662 : pivot_value_new_number (v.f));
1663 pivot_table_put2 (fst->table, var_idx, row++, pv);
1666 for (size_t j = 0; j < frq->n_percentiles; j++)
1668 const struct percentile *pc = &frq->percentiles[j];
1673 .f = vf->tab.n_valid ? vf->percentiles[j] : SYSMIS
1675 pivot_table_put2 (fst->table, var_idx, row++,
1676 pivot_value_new_var_value (vf->var, &v));
1684 frq_stats_table_destroy (fst);
1691 frq_stats_table_destroy (struct frq_stats_table *fst)
1696 pivot_table_submit (fst->table);
1697 pivot_splits_destroy (fst->splits);