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 void dump_statistics (const struct frq_proc *frq,
246 const struct variable *wv);
249 compare_freq (const void *a_, const void *b_, const void *aux_)
251 const struct freq_compare_aux *aux = aux_;
252 const struct freq *a = a_;
253 const struct freq *b = b_;
255 if (aux->by_freq && a->count != b->count)
257 int cmp = a->count > b->count ? 1 : -1;
258 return aux->ascending_freq ? cmp : -cmp;
262 int cmp = value_compare_3way (a->values, b->values, aux->width);
263 return aux->ascending_value ? cmp : -cmp;
267 /* Create a gsl_histogram from a freq_tab */
268 static struct histogram *freq_tab_to_hist (const struct frq_proc *,
269 const struct var_freqs *);
272 put_freq_row (struct pivot_table *table, int var_idx,
273 double frequency, double percent,
274 double valid_percent, double cum_percent)
276 double entries[] = { frequency, percent, valid_percent, cum_percent };
277 for (size_t i = 0; i < sizeof entries / sizeof *entries; i++)
278 if (entries[i] != SYSMIS)
279 pivot_table_put2 (table, i, var_idx,
280 pivot_value_new_number (entries[i]));
283 /* Displays a full frequency table for variable V. */
285 dump_freq_table (const struct var_freqs *vf, const struct variable *wv)
287 const struct freq_tab *ft = &vf->tab;
289 struct pivot_table *table = pivot_table_create__ (pivot_value_new_variable (
290 vf->var), "Frequencies");
291 pivot_table_set_weight_var (table, wv);
293 pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Statistics"),
294 N_("Frequency"), PIVOT_RC_COUNT,
295 N_("Percent"), PIVOT_RC_PERCENT,
296 N_("Valid Percent"), PIVOT_RC_PERCENT,
297 N_("Cumulative Percent"), PIVOT_RC_PERCENT);
299 struct pivot_dimension *variable = pivot_dimension_create__ (
300 table, PIVOT_AXIS_ROW, pivot_value_new_variable (vf->var));
302 double cum_freq = 0.0;
303 double cum_percent = 0.0;
304 struct pivot_category *valid = NULL;
305 for (const struct freq *f = ft->valid; f < ft->missing; f++)
307 cum_freq += f->count;
308 double valid_percent = f->count / ft->valid_cases * 100.0;
309 cum_percent += valid_percent;
312 valid = pivot_category_create_group (variable->root, N_("Valid"));
313 int var_idx = pivot_category_create_leaf (
314 valid, pivot_value_new_var_value (vf->var, &f->values[0]));
315 put_freq_row (table, var_idx, f->count,
316 f->count / ft->total_cases * 100.0,
317 valid_percent, cum_percent);
320 struct pivot_category *missing = NULL;
321 size_t n_categories = ft->n_valid + ft->n_missing;
322 for (const struct freq *f = ft->missing; f < &ft->valid[n_categories]; f++)
324 cum_freq += f->count;
327 missing = pivot_category_create_group (variable->root, N_("Missing"));
328 int var_idx = pivot_category_create_leaf (
329 missing, pivot_value_new_var_value (vf->var, &f->values[0]));
330 put_freq_row (table, var_idx, f->count,
331 f->count / ft->total_cases * 100.0, SYSMIS, SYSMIS);
334 int var_idx = pivot_category_create_leaf (
335 variable->root, pivot_value_new_text (N_("Total")));
336 put_freq_row (table, var_idx, cum_freq, cum_percent, SYSMIS, SYSMIS);
338 pivot_table_submit (table);
341 /* Statistical display. */
344 calc_percentile (double p, double valid_cases, double x1, double x2)
348 s = (settings_get_algorithm () != COMPATIBLE
349 ? modf ((valid_cases - 1) * p, &dummy)
350 : modf ((valid_cases + 1) * p - 1, &dummy));
352 return x1 + (x2 - x1) * s;
355 /* Calculates all of the percentiles for VF within FRQ. */
357 calc_percentiles (const struct frq_proc *frq, struct var_freqs *vf)
359 if (!frq->n_percentiles)
362 vf->percentiles = xnmalloc (frq->n_percentiles, sizeof *vf->percentiles);
364 const struct freq_tab *ft = &vf->tab;
365 const double W = ft->valid_cases;
369 for (const struct freq *f = ft->valid; f < ft->missing; f++)
372 for (; idx < frq->n_percentiles; idx++)
374 struct percentile *pc = &frq->percentiles[idx];
377 tp = (settings_get_algorithm () == ENHANCED
379 : (W + 1) * pc->p - 1);
384 if (tp + 1 < rank || f + 1 >= ft->missing)
385 vf->percentiles[idx] = f->values[0].f;
387 vf->percentiles[idx] = calc_percentile (pc->p, W, f->values[0].f,
391 for (; idx < frq->n_percentiles; idx++)
392 vf->percentiles[idx] = (ft->n_valid > 0
393 ? ft->valid[ft->n_valid - 1].values[0].f
397 /* Returns true iff the value in struct freq F is non-missing
400 not_missing (const void *f_, const void *v_)
402 const struct freq *f = f_;
403 const struct variable *v = v_;
405 return !var_is_value_missing (v, f->values);
409 /* Summarizes the frequency table data for variable V. */
411 postprocess_freq_tab (const struct frq_proc *frq, struct var_freqs *vf)
413 struct freq_tab *ft = &vf->tab;
414 struct freq_compare_aux aux;
416 struct freq *freqs, *f;
419 /* Extract data from hash table. */
420 count = hmap_count (&ft->data);
421 freqs = freq_hmap_extract (&ft->data);
423 /* Put data into ft. */
425 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, vf->var);
426 ft->missing = freqs + ft->n_valid;
427 ft->n_missing = count - ft->n_valid;
430 aux.by_freq = frq->sort == FRQ_AFREQ || frq->sort == FRQ_DFREQ;
431 aux.ascending_freq = frq->sort != FRQ_DFREQ;
432 aux.width = vf->width;
433 aux.ascending_value = frq->sort != FRQ_DVALUE;
434 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare_freq, &aux);
435 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare_freq, &aux);
437 /* Summary statistics. */
438 ft->valid_cases = 0.0;
439 for(i = 0 ; i < ft->n_valid ; ++i)
442 ft->valid_cases += f->count;
446 ft->total_cases = ft->valid_cases ;
447 for(i = 0 ; i < ft->n_missing ; ++i)
450 ft->total_cases += f->count;
455 /* Frees the frequency table for variable V. */
457 cleanup_freq_tab (struct var_freqs *vf)
459 free (vf->tab.valid);
460 freq_hmap_destroy (&vf->tab.data, vf->width);
463 /* Add data from case C to the frequency table. */
465 calc (struct frq_proc *frq, const struct ccase *c, const struct dataset *ds)
467 double weight = dict_get_case_weight (dataset_dict (ds), c, &frq->warn);
470 for (i = 0; i < frq->n_vars; i++)
472 struct var_freqs *vf = &frq->vars[i];
473 const union value *value = case_data (c, vf->var);
474 size_t hash = value_hash (value, vf->width, 0);
477 f = freq_hmap_search (&vf->tab.data, value, vf->width, hash);
479 f = freq_hmap_insert (&vf->tab.data, value, vf->width, hash);
485 /* Prepares each variable that is the target of FREQUENCIES by setting
486 up its hash table. */
488 precalc (struct frq_proc *frq, struct casereader *input, struct dataset *ds)
493 c = casereader_peek (input, 0);
496 output_split_file_values (ds, c);
500 for (i = 0; i < frq->n_vars; i++)
501 hmap_init (&frq->vars[i].tab.data);
504 /* Finishes up with the variables after frequencies have been
505 calculated. Displays statistics, percentiles, ... */
507 postcalc (struct frq_proc *frq, const struct dataset *ds)
509 const struct dictionary *dict = dataset_dict (ds);
510 const struct variable *wv = dict_get_weight (dict);
513 for (i = 0; i < frq->n_vars; i++)
515 struct var_freqs *vf = &frq->vars[i];
516 postprocess_freq_tab (frq, vf);
517 calc_percentiles (frq, vf);
521 dump_statistics (frq, wv);
523 for (i = 0; i < frq->n_vars; i++)
525 struct var_freqs *vf = &frq->vars[i];
527 /* Frequencies tables. */
528 if (vf->tab.n_valid + vf->tab.n_missing <= frq->max_categories)
529 dump_freq_table (vf, wv);
532 if (frq->hist && var_is_numeric (vf->var) && vf->tab.n_valid > 0)
534 double d[FRQ_ST_count];
535 struct histogram *histogram;
537 calc_stats (frq, vf, d);
539 histogram = freq_tab_to_hist (frq, vf);
543 chart_submit (histogram_chart_create (
544 histogram->gsl_hist, var_to_string(vf->var),
548 frq->hist->draw_normal));
550 statistic_destroy (&histogram->parent);
555 do_piechart(frq->pie, vf->var, &vf->tab);
558 do_barchart(frq->bar, &vf->var, &vf->tab);
560 cleanup_freq_tab (vf);
565 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
569 const struct variable **vars = NULL;
571 bool sbc_barchart = false;
572 bool sbc_piechart = false;
573 bool sbc_histogram = false;
575 double pie_min = -DBL_MAX;
576 double pie_max = DBL_MAX;
577 bool pie_missing = true;
579 double bar_min = -DBL_MAX;
580 double bar_max = DBL_MAX;
581 bool bar_freq = true;
583 double hi_min = -DBL_MAX;
584 double hi_max = DBL_MAX;
585 int hi_scale = FRQ_FREQ;
586 int hi_freq = INT_MIN;
587 int hi_pcnt = INT_MIN;
588 int hi_norm = FRQ_NONORMAL;
590 frq.sort = FRQ_AVALUE;
595 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
596 | BIT_INDEX (FRQ_ST_STDDEV)
597 | BIT_INDEX (FRQ_ST_MINIMUM)
598 | BIT_INDEX (FRQ_ST_MAXIMUM);
602 frq.max_categories = LONG_MAX;
604 frq.percentiles = NULL;
605 frq.n_percentiles = 0;
613 /* Accept an optional, completely pointless "/VARIABLES=" */
614 lex_match (lexer, T_SLASH);
615 if (lex_match_id (lexer, "VARIABLES"))
617 if (! lex_force_match (lexer, T_EQUALS))
621 if (!parse_variables_const (lexer, dataset_dict (ds),
627 frq.vars = xcalloc (frq.n_vars, sizeof (*frq.vars));
628 for (i = 0; i < frq.n_vars; ++i)
630 frq.vars[i].var = vars[i];
631 frq.vars[i].width = var_get_width (vars[i]);
634 while (lex_token (lexer) != T_ENDCMD)
636 lex_match (lexer, T_SLASH);
638 if (lex_match_id (lexer, "STATISTICS"))
640 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
641 | BIT_INDEX (FRQ_ST_STDDEV)
642 | BIT_INDEX (FRQ_ST_MINIMUM)
643 | BIT_INDEX (FRQ_ST_MAXIMUM);
647 if (lex_match (lexer, T_EQUALS))
653 while (lex_token (lexer) != T_ENDCMD
654 && lex_token (lexer) != T_SLASH)
656 if (lex_match_id (lexer, "DEFAULT"))
658 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
659 | BIT_INDEX (FRQ_ST_STDDEV)
660 | BIT_INDEX (FRQ_ST_MINIMUM)
661 | BIT_INDEX (FRQ_ST_MAXIMUM);
665 else if (lex_match_id (lexer, "MEAN"))
667 frq.stats |= BIT_INDEX (FRQ_ST_MEAN);
670 else if (lex_match_id (lexer, "SEMEAN"))
672 frq.stats |= BIT_INDEX (FRQ_ST_SEMEAN);
675 else if (lex_match_id (lexer, "MEDIAN"))
677 frq.stats |= BIT_INDEX (FRQ_ST_MEDIAN);
680 else if (lex_match_id (lexer, "MODE"))
682 frq.stats |= BIT_INDEX (FRQ_ST_MODE);
685 else if (lex_match_id (lexer, "STDDEV"))
687 frq.stats |= BIT_INDEX (FRQ_ST_STDDEV);
690 else if (lex_match_id (lexer, "VARIANCE"))
692 frq.stats |= BIT_INDEX (FRQ_ST_VARIANCE);
695 else if (lex_match_id (lexer, "KURTOSIS"))
697 frq.stats |= BIT_INDEX (FRQ_ST_KURTOSIS);
700 else if (lex_match_id (lexer, "SKEWNESS"))
702 frq.stats |= BIT_INDEX (FRQ_ST_SKEWNESS);
705 else if (lex_match_id (lexer, "RANGE"))
707 frq.stats |= BIT_INDEX (FRQ_ST_RANGE);
710 else if (lex_match_id (lexer, "MINIMUM"))
712 frq.stats |= BIT_INDEX (FRQ_ST_MINIMUM);
715 else if (lex_match_id (lexer, "MAXIMUM"))
717 frq.stats |= BIT_INDEX (FRQ_ST_MAXIMUM);
720 else if (lex_match_id (lexer, "SUM"))
722 frq.stats |= BIT_INDEX (FRQ_ST_SUM);
725 else if (lex_match_id (lexer, "SESKEWNESS"))
727 frq.stats |= BIT_INDEX (FRQ_ST_SESKEWNESS);
730 else if (lex_match_id (lexer, "SEKURTOSIS"))
732 frq.stats |= BIT_INDEX (FRQ_ST_SEKURTOSIS);
735 else if (lex_match_id (lexer, "NONE"))
740 else if (lex_match (lexer, T_ALL))
743 frq.n_stats = FRQ_ST_count;
747 lex_error (lexer, NULL);
752 else if (lex_match_id (lexer, "PERCENTILES"))
754 lex_match (lexer, T_EQUALS);
755 while (lex_token (lexer) != T_ENDCMD
756 && lex_token (lexer) != T_SLASH)
758 if (lex_force_num (lexer))
761 xrealloc (frq.percentiles,
762 (frq.n_percentiles + 1)
763 * sizeof (*frq.percentiles));
764 frq.percentiles[frq.n_percentiles].p = lex_number (lexer) / 100.0;
765 frq.percentiles[frq.n_percentiles].show = true;
771 lex_error (lexer, NULL);
774 lex_match (lexer, T_COMMA);
777 else if (lex_match_id (lexer, "FORMAT"))
779 lex_match (lexer, T_EQUALS);
780 while (lex_token (lexer) != T_ENDCMD
781 && lex_token (lexer) != T_SLASH)
783 if (lex_match_id (lexer, "TABLE"))
786 else if (lex_match_id (lexer, "NOTABLE"))
788 frq.max_categories = 0;
790 else if (lex_match_id (lexer, "LIMIT"))
792 if (!lex_force_match (lexer, T_LPAREN)
793 || !lex_force_int_range (lexer, "LIMIT", 0, INT_MAX))
796 frq.max_categories = lex_integer (lexer);
799 if (!lex_force_match (lexer, T_RPAREN))
802 else if (lex_match_id (lexer, "AVALUE"))
804 frq.sort = FRQ_AVALUE;
806 else if (lex_match_id (lexer, "DVALUE"))
808 frq.sort = FRQ_DVALUE;
810 else if (lex_match_id (lexer, "AFREQ"))
812 frq.sort = FRQ_AFREQ;
814 else if (lex_match_id (lexer, "DFREQ"))
816 frq.sort = FRQ_DFREQ;
820 lex_error (lexer, NULL);
825 else if (lex_match_id (lexer, "NTILES"))
827 lex_match (lexer, T_EQUALS);
829 if (lex_force_int_range (lexer, "NTILES", 0, INT_MAX))
831 int n = lex_integer (lexer);
833 for (int i = 0; i < n + 1; ++i)
836 xrealloc (frq.percentiles,
837 (frq.n_percentiles + 1)
838 * sizeof (*frq.percentiles));
839 frq.percentiles[frq.n_percentiles].p =
841 frq.percentiles[frq.n_percentiles].show = true;
848 lex_error (lexer, NULL);
852 else if (lex_match_id (lexer, "ALGORITHM"))
854 lex_match (lexer, T_EQUALS);
856 if (lex_match_id (lexer, "COMPATIBLE"))
858 settings_set_cmd_algorithm (COMPATIBLE);
860 else if (lex_match_id (lexer, "ENHANCED"))
862 settings_set_cmd_algorithm (ENHANCED);
866 lex_error (lexer, NULL);
870 else if (lex_match_id (lexer, "HISTOGRAM"))
872 lex_match (lexer, T_EQUALS);
873 sbc_histogram = true;
875 while (lex_token (lexer) != T_ENDCMD
876 && lex_token (lexer) != T_SLASH)
878 if (lex_match_id (lexer, "NORMAL"))
880 hi_norm = FRQ_NORMAL;
882 else if (lex_match_id (lexer, "NONORMAL"))
884 hi_norm = FRQ_NONORMAL;
886 else if (lex_match_id (lexer, "FREQ"))
889 if (lex_match (lexer, T_LPAREN))
891 if (lex_force_int_range (lexer, "FREQ", 1, INT_MAX))
893 hi_freq = lex_integer (lexer);
895 if (! lex_force_match (lexer, T_RPAREN))
900 else if (lex_match_id (lexer, "PERCENT"))
902 hi_scale = FRQ_PERCENT;
903 if (lex_match (lexer, T_LPAREN))
905 if (lex_force_int_range (lexer, "PERCENT", 1, INT_MAX))
907 hi_pcnt = lex_integer (lexer);
909 if (! lex_force_match (lexer, T_RPAREN))
914 else if (lex_match_id (lexer, "MINIMUM"))
916 if (! lex_force_match (lexer, T_LPAREN))
918 if (lex_force_num (lexer))
920 hi_min = lex_number (lexer);
923 if (! lex_force_match (lexer, T_RPAREN))
926 else if (lex_match_id (lexer, "MAXIMUM"))
928 if (! lex_force_match (lexer, T_LPAREN))
930 if (lex_force_num (lexer))
932 hi_max = lex_number (lexer);
935 if (! lex_force_match (lexer, T_RPAREN))
940 lex_error (lexer, NULL);
945 else if (lex_match_id (lexer, "PIECHART"))
947 lex_match (lexer, T_EQUALS);
948 while (lex_token (lexer) != T_ENDCMD
949 && lex_token (lexer) != T_SLASH)
951 if (lex_match_id (lexer, "MINIMUM"))
953 if (! lex_force_match (lexer, T_LPAREN))
955 if (lex_force_num (lexer))
957 pie_min = lex_number (lexer);
960 if (! lex_force_match (lexer, T_RPAREN))
963 else if (lex_match_id (lexer, "MAXIMUM"))
965 if (! lex_force_match (lexer, T_LPAREN))
967 if (lex_force_num (lexer))
969 pie_max = lex_number (lexer);
972 if (! lex_force_match (lexer, T_RPAREN))
975 else if (lex_match_id (lexer, "MISSING"))
979 else if (lex_match_id (lexer, "NOMISSING"))
985 lex_error (lexer, NULL);
991 else if (lex_match_id (lexer, "BARCHART"))
993 lex_match (lexer, T_EQUALS);
994 while (lex_token (lexer) != T_ENDCMD
995 && lex_token (lexer) != T_SLASH)
997 if (lex_match_id (lexer, "MINIMUM"))
999 if (! lex_force_match (lexer, T_LPAREN))
1001 if (lex_force_num (lexer))
1003 bar_min = lex_number (lexer);
1006 if (! lex_force_match (lexer, T_RPAREN))
1009 else if (lex_match_id (lexer, "MAXIMUM"))
1011 if (! lex_force_match (lexer, T_LPAREN))
1013 if (lex_force_num (lexer))
1015 bar_max = lex_number (lexer);
1018 if (! lex_force_match (lexer, T_RPAREN))
1021 else if (lex_match_id (lexer, "FREQ"))
1023 if (lex_match (lexer, T_LPAREN))
1025 if (lex_force_num (lexer))
1030 if (! lex_force_match (lexer, T_RPAREN))
1035 else if (lex_match_id (lexer, "PERCENT"))
1037 if (lex_match (lexer, T_LPAREN))
1039 if (lex_force_num (lexer))
1044 if (! lex_force_match (lexer, T_RPAREN))
1051 lex_error (lexer, NULL);
1055 sbc_barchart = true;
1057 else if (lex_match_id (lexer, "MISSING"))
1059 lex_match (lexer, T_EQUALS);
1061 while (lex_token (lexer) != T_ENDCMD
1062 && lex_token (lexer) != T_SLASH)
1064 if (lex_match_id (lexer, "EXCLUDE"))
1067 else if (lex_match_id (lexer, "INCLUDE"))
1072 lex_error (lexer, NULL);
1077 else if (lex_match_id (lexer, "ORDER"))
1079 lex_match (lexer, T_EQUALS);
1080 if (!lex_match_id (lexer, "ANALYSIS"))
1081 lex_match_id (lexer, "VARIABLE");
1085 lex_error (lexer, NULL);
1090 if (frq.stats & BIT_INDEX (FRQ_ST_MEDIAN))
1093 xrealloc (frq.percentiles,
1094 (frq.n_percentiles + 1)
1095 * sizeof (*frq.percentiles));
1097 frq.percentiles[frq.n_percentiles].p = 0.50;
1098 frq.percentiles[frq.n_percentiles].show = false;
1100 frq.n_percentiles++;
1104 /* Figure out which charts the user requested. */
1109 struct frq_chart *hist;
1111 hist = frq.hist = xmalloc (sizeof *frq.hist);
1112 hist->x_min = hi_min;
1113 hist->x_max = hi_max;
1114 hist->y_scale = hi_scale;
1115 hist->y_max = hi_scale == FRQ_FREQ ? hi_freq : hi_pcnt;
1116 hist->draw_normal = hi_norm != FRQ_NONORMAL;
1117 hist->include_missing = false;
1119 if (hist->x_min != SYSMIS && hist->x_max != SYSMIS
1120 && hist->x_min >= hist->x_max)
1122 msg (SE, _("%s for histogram must be greater than or equal to %s, "
1123 "but %s was specified as %.15g and %s as %.15g. "
1124 "%s and %s will be ignored."),
1129 hist->x_min = hist->x_max = SYSMIS;
1133 xrealloc (frq.percentiles,
1134 (frq.n_percentiles + 2)
1135 * sizeof (*frq.percentiles));
1137 frq.percentiles[frq.n_percentiles].p = 0.25;
1138 frq.percentiles[frq.n_percentiles].show = false;
1140 frq.percentiles[frq.n_percentiles + 1].p = 0.75;
1141 frq.percentiles[frq.n_percentiles + 1].show = false;
1143 frq.n_percentiles+=2;
1148 frq.bar = xmalloc (sizeof *frq.bar);
1149 frq.bar->x_min = bar_min;
1150 frq.bar->x_max = bar_max;
1151 frq.bar->include_missing = false;
1152 frq.bar->y_scale = bar_freq ? FRQ_FREQ : FRQ_PERCENT;
1157 struct frq_chart *pie;
1159 pie = frq.pie = xmalloc (sizeof *frq.pie);
1160 pie->x_min = pie_min;
1161 pie->x_max = pie_max;
1162 pie->include_missing = pie_missing;
1164 if (pie->x_min != SYSMIS && pie->x_max != SYSMIS
1165 && pie->x_min >= pie->x_max)
1167 msg (SE, _("%s for pie chart must be greater than or equal to %s, "
1168 "but %s was specified as %.15g and %s as %.15g. "
1169 "%s and %s will be ignored."),
1174 pie->x_min = pie->x_max = SYSMIS;
1181 double previous_p = -1;
1182 qsort (frq.percentiles, frq.n_percentiles,
1183 sizeof (*frq.percentiles),
1186 for (i = o = 0; i < frq.n_percentiles; ++i)
1188 if (frq.percentiles[i].p != previous_p)
1190 frq.percentiles[o].p = frq.percentiles[i].p;
1191 frq.percentiles[o].show = frq.percentiles[i].show;
1194 else if (frq.percentiles[i].show &&
1195 !frq.percentiles[o].show)
1197 frq.percentiles[o].show = true;
1199 previous_p = frq.percentiles[i].p;
1202 frq.n_percentiles = o;
1204 frq.median_idx = SIZE_MAX;
1205 for (i = 0; i < frq.n_percentiles; i++)
1206 if (frq.percentiles[i].p == 0.5)
1214 struct casegrouper *grouper;
1215 struct casereader *group;
1218 grouper = casegrouper_create_splits (proc_open (ds), dataset_dict (ds));
1219 while (casegrouper_get_next_group (grouper, &group))
1222 precalc (&frq, group, ds);
1224 for (; (c = casereader_read (group)) != NULL; case_unref (c))
1226 postcalc (&frq, ds);
1227 casereader_destroy (group);
1229 ok = casegrouper_destroy (grouper);
1230 ok = proc_commit (ds) && ok;
1234 for (size_t i = 0; i < frq.n_vars; i++)
1235 free (frq.vars[i].percentiles);
1240 free (frq.percentiles);
1247 for (size_t i = 0; i < frq.n_vars; i++)
1248 free (frq.vars[i].percentiles);
1253 free (frq.percentiles);
1259 calculate_iqr (const struct frq_proc *frq, const struct var_freqs *vf)
1265 /* This cannot work unless the 25th and 75th percentile are calculated */
1266 assert (frq->n_percentiles >= 2);
1267 for (i = 0; i < frq->n_percentiles; i++)
1269 struct percentile *pc = &frq->percentiles[i];
1271 if (fabs (0.25 - pc->p) < DBL_EPSILON)
1272 q1 = vf->percentiles[i];
1273 else if (fabs (0.75 - pc->p) < DBL_EPSILON)
1274 q3 = vf->percentiles[i];
1277 return q1 == SYSMIS || q3 == SYSMIS ? SYSMIS : q3 - q1;
1281 chart_includes_value (const struct frq_chart *chart,
1282 const struct variable *var,
1283 const union value *value)
1285 if (!chart->include_missing && var_is_value_missing (var, value))
1288 if (var_is_numeric (var)
1289 && ((chart->x_min != SYSMIS && value->f < chart->x_min)
1290 || (chart->x_max != SYSMIS && value->f > chart->x_max)))
1296 /* Create a gsl_histogram from a freq_tab */
1297 static struct histogram *
1298 freq_tab_to_hist (const struct frq_proc *frq, const struct var_freqs *vf)
1300 /* Find out the extremes of the x value, within the range to be included in
1301 the histogram, and sum the total frequency of those values. */
1302 double x_min = DBL_MAX;
1303 double x_max = -DBL_MAX;
1304 double valid_freq = 0;
1305 for (int i = 0; i < vf->tab.n_valid; i++)
1307 const struct freq *f = &vf->tab.valid[i];
1308 if (chart_includes_value (frq->hist, vf->var, f->values))
1310 x_min = MIN (x_min, f->values[0].f);
1311 x_max = MAX (x_max, f->values[0].f);
1312 valid_freq += f->count;
1316 if (valid_freq <= 0)
1319 double iqr = calculate_iqr (frq, vf);
1323 ? 2 * iqr / pow (valid_freq, 1.0 / 3.0) /* Freedman-Diaconis. */
1324 : (x_max - x_min) / (1 + log2 (valid_freq))); /* Sturges */
1326 struct histogram *histogram = histogram_create (bin_width, x_min, x_max);
1327 if (histogram == NULL)
1330 for (int i = 0; i < vf->tab.n_valid; i++)
1332 const struct freq *f = &vf->tab.valid[i];
1333 if (chart_includes_value (frq->hist, vf->var, f->values))
1334 histogram_add (histogram, f->values[0].f, f->count);
1341 /* Allocate an array of struct freqs and fill them from the data in FRQ_TAB,
1342 according to the parameters of CATCHART
1343 N_SLICES will contain the number of slices allocated.
1344 The caller is responsible for freeing slices
1346 static struct freq *
1347 pick_cat_counts (const struct frq_chart *catchart,
1348 const struct freq_tab *frq_tab,
1353 struct freq *slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1355 for (i = 0; i < frq_tab->n_valid; i++)
1357 const struct freq *f = &frq_tab->valid[i];
1358 if (f->count > catchart->x_max)
1361 if (f->count < catchart->x_min)
1364 slices[n_slices] = *f;
1369 if (catchart->include_missing)
1371 for (i = 0; i < frq_tab->n_missing; i++)
1373 const struct freq *f = &frq_tab->missing[i];
1374 slices[n_slices].count += f->count;
1377 slices[n_slices].values[0] = f->values[0];
1380 if (frq_tab->n_missing > 0)
1384 *n_slicesp = n_slices;
1389 /* Allocate an array of struct freqs and fill them from the data in FRQ_TAB,
1390 according to the parameters of CATCHART
1391 N_SLICES will contain the number of slices allocated.
1392 The caller is responsible for freeing slices
1394 static struct freq **
1395 pick_cat_counts_ptr (const struct frq_chart *catchart,
1396 const struct freq_tab *frq_tab,
1401 struct freq **slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1403 for (i = 0; i < frq_tab->n_valid; i++)
1405 struct freq *f = &frq_tab->valid[i];
1406 if (f->count > catchart->x_max)
1409 if (f->count < catchart->x_min)
1412 slices[n_slices] = f;
1417 if (catchart->include_missing)
1419 for (i = 0; i < frq_tab->n_missing; i++)
1421 const struct freq *f = &frq_tab->missing[i];
1424 slices[n_slices] = xmalloc (sizeof (struct freq));
1425 slices[n_slices]->values[0] = f->values[0];
1428 slices[n_slices]->count += f->count;
1433 *n_slicesp = n_slices;
1440 do_piechart(const struct frq_chart *pie, const struct variable *var,
1441 const struct freq_tab *frq_tab)
1444 struct freq *slices = pick_cat_counts (pie, frq_tab, &n_slices);
1447 msg (SW, _("Omitting pie chart for %s, which has only %d unique values."),
1448 var_get_name (var), n_slices);
1449 else if (n_slices > 50)
1450 msg (SW, _("Omitting pie chart for %s, which has over 50 unique values."),
1451 var_get_name (var));
1453 chart_submit (piechart_create (var, slices, n_slices));
1460 do_barchart(const struct frq_chart *bar, const struct variable **var,
1461 const struct freq_tab *frq_tab)
1464 struct freq **slices = pick_cat_counts_ptr (bar, frq_tab, &n_slices);
1467 msg (SW, _("Omitting bar chart, which has no values."));
1469 chart_submit (barchart_create (var, 1,
1470 (bar->y_scale == FRQ_FREQ) ? _("Count") : _("Percent"),
1471 (bar->y_scale == FRQ_PERCENT),
1477 /* Calculates all the pertinent statistics for VF, putting them in array
1480 calc_stats (const struct frq_proc *frq, const struct var_freqs *vf,
1481 double d[FRQ_ST_count])
1483 const struct freq_tab *ft = &vf->tab;
1484 double W = ft->valid_cases;
1485 const struct freq *f;
1487 int most_often = -1;
1488 double X_mode = SYSMIS;
1490 /* Calculate the mode. */
1491 for (f = ft->valid; f < ft->missing; f++)
1493 if (most_often < f->count)
1495 most_often = f->count;
1496 X_mode = f->values[0].f;
1498 else if (most_often == f->count)
1500 /* A duplicate mode is undefined.
1501 FIXME: keep track of *all* the modes. */
1506 /* Calculate moments. */
1507 m = moments_create (MOMENT_KURTOSIS);
1508 for (f = ft->valid; f < ft->missing; f++)
1509 moments_pass_one (m, f->values[0].f, f->count);
1510 for (f = ft->valid; f < ft->missing; f++)
1511 moments_pass_two (m, f->values[0].f, f->count);
1512 moments_calculate (m, NULL, &d[FRQ_ST_MEAN], &d[FRQ_ST_VARIANCE],
1513 &d[FRQ_ST_SKEWNESS], &d[FRQ_ST_KURTOSIS]);
1514 moments_destroy (m);
1516 /* Formulae below are taken from _SPSS Statistical Algorithms_. */
1517 if (ft->n_valid > 0)
1519 d[FRQ_ST_MINIMUM] = ft->valid[0].values[0].f;
1520 d[FRQ_ST_MAXIMUM] = ft->valid[ft->n_valid - 1].values[0].f;
1521 d[FRQ_ST_RANGE] = d[FRQ_ST_MAXIMUM] - d[FRQ_ST_MINIMUM];
1525 d[FRQ_ST_MINIMUM] = SYSMIS;
1526 d[FRQ_ST_MAXIMUM] = SYSMIS;
1527 d[FRQ_ST_RANGE] = SYSMIS;
1529 d[FRQ_ST_MODE] = X_mode;
1530 d[FRQ_ST_SUM] = d[FRQ_ST_MEAN] * W;
1531 d[FRQ_ST_STDDEV] = sqrt (d[FRQ_ST_VARIANCE]);
1532 d[FRQ_ST_SEMEAN] = d[FRQ_ST_STDDEV] / sqrt (W);
1533 d[FRQ_ST_SESKEWNESS] = calc_seskew (W);
1534 d[FRQ_ST_SEKURTOSIS] = calc_sekurt (W);
1535 d[FRQ_ST_MEDIAN] = (frq->median_idx != SIZE_MAX
1536 ? vf->percentiles[frq->median_idx]
1541 all_string_variables (const struct frq_proc *frq)
1543 for (size_t i = 0; i < frq->n_vars; i++)
1544 if (var_is_numeric (frq->vars[i].var))
1550 /* Displays a table of all the statistics requested. */
1552 dump_statistics (const struct frq_proc *frq, const struct variable *wv)
1554 if (all_string_variables (frq))
1557 struct pivot_table *table = pivot_table_create (N_("Statistics"));
1558 pivot_table_set_weight_var (table, wv);
1560 struct pivot_dimension *variables
1561 = pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Variables"));
1563 struct pivot_dimension *statistics = pivot_dimension_create (
1564 table, PIVOT_AXIS_ROW, N_("Statistics"));
1565 struct pivot_category *n = pivot_category_create_group (
1566 statistics->root, N_("N"));
1567 pivot_category_create_leaves (n,
1568 N_("Valid"), PIVOT_RC_COUNT,
1569 N_("Missing"), PIVOT_RC_COUNT);
1570 for (int i = 0; i < FRQ_ST_count; i++)
1571 if (frq->stats & BIT_INDEX (i))
1572 pivot_category_create_leaf (statistics->root,
1573 pivot_value_new_text (st_name[i]));
1574 struct pivot_category *percentiles = NULL;
1575 for (size_t i = 0; i < frq->n_percentiles; i++)
1577 const struct percentile *pc = &frq->percentiles[i];
1583 percentiles = pivot_category_create_group (
1584 statistics->root, N_("Percentiles"));
1585 pivot_category_create_leaf (percentiles, pivot_value_new_integer (
1589 for (size_t i = 0; i < frq->n_vars; i++)
1591 struct var_freqs *vf = &frq->vars[i];
1592 if (var_is_alpha (vf->var))
1595 const struct freq_tab *ft = &vf->tab;
1597 int var_idx = pivot_category_create_leaf (
1598 variables->root, pivot_value_new_variable (vf->var));
1601 pivot_table_put2 (table, var_idx, row++,
1602 pivot_value_new_number (ft->valid_cases));
1603 pivot_table_put2 (table, var_idx, row++,
1604 pivot_value_new_number (
1605 ft->total_cases - ft->valid_cases));
1607 double stat_values[FRQ_ST_count];
1608 calc_stats (frq, vf, stat_values);
1609 for (int j = 0; j < FRQ_ST_count; j++)
1611 if (!(frq->stats & BIT_INDEX (j)))
1614 union value v = { .f = vf->tab.n_valid ? stat_values[j] : SYSMIS };
1615 struct pivot_value *pv
1616 = (j == FRQ_ST_MODE || j == FRQ_ST_MINIMUM || j == FRQ_ST_MAXIMUM
1617 ? pivot_value_new_var_value (vf->var, &v)
1618 : pivot_value_new_number (v.f));
1619 pivot_table_put2 (table, var_idx, row++, pv);
1622 for (size_t j = 0; j < frq->n_percentiles; j++)
1624 const struct percentile *pc = &frq->percentiles[j];
1629 .f = vf->tab.n_valid ? vf->percentiles[j] : SYSMIS
1631 pivot_table_put2 (table, var_idx, row++,
1632 pivot_value_new_var_value (vf->var, &v));
1636 pivot_table_submit (table);