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;
231 struct freq_compare_aux
237 bool ascending_value;
240 static void calc_stats (const struct frq_proc *,
241 const struct var_freqs *, double d[FRQ_ST_count]);
243 static void do_piechart(const struct frq_chart *pie,
244 const struct variable *var,
245 const struct freq_tab *frq_tab);
247 static void do_barchart(const struct frq_chart *bar,
248 const struct variable **var,
249 const struct freq_tab *frq_tab);
251 static void dump_statistics (const struct frq_proc *frq,
252 const struct variable *wv);
255 compare_freq (const void *a_, const void *b_, const void *aux_)
257 const struct freq_compare_aux *aux = aux_;
258 const struct freq *a = a_;
259 const struct freq *b = b_;
261 if (aux->by_freq && a->count != b->count)
263 int cmp = a->count > b->count ? 1 : -1;
264 return aux->ascending_freq ? cmp : -cmp;
268 int cmp = value_compare_3way (a->values, b->values, aux->width);
269 return aux->ascending_value ? cmp : -cmp;
273 /* Create a gsl_histogram from a freq_tab */
274 static struct histogram *
275 freq_tab_to_hist (const struct frq_proc *frq, const struct freq_tab *ft,
276 const struct variable *var);
279 put_freq_row (struct pivot_table *table, int var_idx,
280 double frequency, double percent,
281 double valid_percent, double cum_percent)
283 double entries[] = { frequency, percent, valid_percent, cum_percent };
284 for (size_t i = 0; i < sizeof entries / sizeof *entries; i++)
285 if (entries[i] != SYSMIS)
286 pivot_table_put2 (table, i, var_idx,
287 pivot_value_new_number (entries[i]));
290 /* Displays a full frequency table for variable V. */
292 dump_freq_table (const struct var_freqs *vf, const struct variable *wv)
294 const struct freq_tab *ft = &vf->tab;
296 struct pivot_table *table = pivot_table_create__ (pivot_value_new_variable (
298 pivot_table_set_weight_var (table, wv);
300 pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Statistics"),
301 N_("Frequency"), PIVOT_RC_COUNT,
302 N_("Percent"), PIVOT_RC_PERCENT,
303 N_("Valid Percent"), PIVOT_RC_PERCENT,
304 N_("Cumulative Percent"), PIVOT_RC_PERCENT);
306 struct pivot_dimension *variable = pivot_dimension_create__ (
307 table, PIVOT_AXIS_ROW, pivot_value_new_variable (vf->var));
309 double cum_freq = 0.0;
310 double cum_percent = 0.0;
311 struct pivot_category *valid = NULL;
312 for (const struct freq *f = ft->valid; f < ft->missing; f++)
314 cum_freq += f->count;
315 double valid_percent = f->count / ft->valid_cases * 100.0;
316 cum_percent += valid_percent;
319 valid = pivot_category_create_group (variable->root, N_("Valid"));
320 int var_idx = pivot_category_create_leaf (
321 valid, pivot_value_new_var_value (vf->var, &f->values[0]));
322 put_freq_row (table, var_idx, f->count,
323 f->count / ft->total_cases * 100.0,
324 valid_percent, cum_percent);
327 struct pivot_category *missing = NULL;
328 size_t n_categories = ft->n_valid + ft->n_missing;
329 for (const struct freq *f = ft->missing; f < &ft->valid[n_categories]; f++)
331 cum_freq += f->count;
334 missing = pivot_category_create_group (variable->root, N_("Missing"));
335 int var_idx = pivot_category_create_leaf (
336 missing, pivot_value_new_var_value (vf->var, &f->values[0]));
337 put_freq_row (table, var_idx, f->count,
338 f->count / ft->total_cases * 100.0, SYSMIS, SYSMIS);
341 int var_idx = pivot_category_create_leaf (
342 variable->root, pivot_value_new_text (N_("Total")));
343 put_freq_row (table, var_idx, cum_freq, cum_percent, SYSMIS, SYSMIS);
345 pivot_table_submit (table);
348 /* Statistical display. */
351 calc_percentile (double p, double valid_cases, double x1, double x2)
355 s = (settings_get_algorithm () != COMPATIBLE
356 ? modf ((valid_cases - 1) * p, &dummy)
357 : modf ((valid_cases + 1) * p - 1, &dummy));
359 return x1 + (x2 - x1) * s;
362 /* Calculates all of the percentiles for VF within FRQ. */
364 calc_percentiles (const struct frq_proc *frq, const struct var_freqs *vf)
366 const struct freq_tab *ft = &vf->tab;
367 double W = ft->valid_cases;
368 const struct freq *f;
369 int percentile_idx = 0;
372 for (f = ft->valid; f < ft->missing; f++)
375 for (; percentile_idx < frq->n_percentiles; percentile_idx++)
377 struct percentile *pc = &frq->percentiles[percentile_idx];
380 tp = (settings_get_algorithm () == ENHANCED
382 : (W + 1) * pc->p - 1);
387 if (tp + 1 < rank || f + 1 >= ft->missing)
388 pc->value = f->values[0].f;
390 pc->value = calc_percentile (pc->p, W, f->values[0].f, f[1].values[0].f);
393 for (; percentile_idx < frq->n_percentiles; percentile_idx++)
395 struct percentile *pc = &frq->percentiles[percentile_idx];
396 pc->value = (ft->n_valid > 0
397 ? ft->valid[ft->n_valid - 1].values[0].f
402 /* Returns true iff the value in struct freq F is non-missing
405 not_missing (const void *f_, const void *v_)
407 const struct freq *f = f_;
408 const struct variable *v = v_;
410 return !var_is_value_missing (v, f->values, MV_ANY);
414 /* Summarizes the frequency table data for variable V. */
416 postprocess_freq_tab (const struct frq_proc *frq, struct var_freqs *vf)
418 struct freq_tab *ft = &vf->tab;
419 struct freq_compare_aux aux;
421 struct freq *freqs, *f;
424 /* Extract data from hash table. */
425 count = hmap_count (&ft->data);
426 freqs = freq_hmap_extract (&ft->data);
428 /* Put data into ft. */
430 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, vf->var);
431 ft->missing = freqs + ft->n_valid;
432 ft->n_missing = count - ft->n_valid;
435 aux.by_freq = frq->sort == FRQ_AFREQ || frq->sort == FRQ_DFREQ;
436 aux.ascending_freq = frq->sort != FRQ_DFREQ;
437 aux.width = vf->width;
438 aux.ascending_value = frq->sort != FRQ_DVALUE;
439 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare_freq, &aux);
440 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare_freq, &aux);
442 /* Summary statistics. */
443 ft->valid_cases = 0.0;
444 for(i = 0 ; i < ft->n_valid ; ++i )
447 ft->valid_cases += f->count;
451 ft->total_cases = ft->valid_cases ;
452 for(i = 0 ; i < ft->n_missing ; ++i )
455 ft->total_cases += f->count;
460 /* Frees the frequency table for variable V. */
462 cleanup_freq_tab (struct var_freqs *vf)
464 free (vf->tab.valid);
465 freq_hmap_destroy (&vf->tab.data, vf->width);
468 /* Add data from case C to the frequency table. */
470 calc (struct frq_proc *frq, const struct ccase *c, const struct dataset *ds)
472 double weight = dict_get_case_weight (dataset_dict (ds), c, NULL);
475 for (i = 0; i < frq->n_vars; i++)
477 struct var_freqs *vf = &frq->vars[i];
478 const union value *value = case_data (c, vf->var);
479 size_t hash = value_hash (value, vf->width, 0);
482 f = freq_hmap_search (&vf->tab.data, value, vf->width, hash);
484 f = freq_hmap_insert (&vf->tab.data, value, vf->width, hash);
490 /* Prepares each variable that is the target of FREQUENCIES by setting
491 up its hash table. */
493 precalc (struct frq_proc *frq, struct casereader *input, struct dataset *ds)
498 c = casereader_peek (input, 0);
501 output_split_file_values (ds, c);
505 for (i = 0; i < frq->n_vars; i++)
506 hmap_init (&frq->vars[i].tab.data);
509 /* Finishes up with the variables after frequencies have been
510 calculated. Displays statistics, percentiles, ... */
512 postcalc (struct frq_proc *frq, const struct dataset *ds)
514 const struct dictionary *dict = dataset_dict (ds);
515 const struct variable *wv = dict_get_weight (dict);
518 for (i = 0; i < frq->n_vars; i++)
520 struct var_freqs *vf = &frq->vars[i];
521 postprocess_freq_tab (frq, vf);
522 calc_percentiles (frq, vf);
526 dump_statistics (frq, wv);
528 for (i = 0; i < frq->n_vars; i++)
530 struct var_freqs *vf = &frq->vars[i];
532 /* Frequencies tables. */
533 if (vf->tab.n_valid + vf->tab.n_missing <= frq->max_categories)
534 dump_freq_table (vf, wv);
537 if (frq->hist && var_is_numeric (vf->var) && vf->tab.n_valid > 0)
539 double d[FRQ_ST_count];
540 struct histogram *histogram;
542 calc_stats (frq, vf, d);
544 histogram = freq_tab_to_hist (frq, &vf->tab, vf->var);
548 chart_item_submit (histogram_chart_create (
549 histogram->gsl_hist, var_to_string(vf->var),
553 frq->hist->draw_normal));
555 statistic_destroy (&histogram->parent);
560 do_piechart(frq->pie, vf->var, &vf->tab);
563 do_barchart(frq->bar, &vf->var, &vf->tab);
565 cleanup_freq_tab (vf);
570 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
574 const struct variable **vars = NULL;
576 bool sbc_barchart = false;
577 bool sbc_piechart = false;
578 bool sbc_histogram = false;
580 double pie_min = -DBL_MAX;
581 double pie_max = DBL_MAX;
582 bool pie_missing = true;
584 double bar_min = -DBL_MAX;
585 double bar_max = DBL_MAX;
586 bool bar_freq = true;
588 double hi_min = -DBL_MAX;
589 double hi_max = DBL_MAX;
590 int hi_scale = FRQ_FREQ;
591 int hi_freq = INT_MIN;
592 int hi_pcnt = INT_MIN;
593 int hi_norm = FRQ_NONORMAL;
595 frq.pool = pool_create ();
596 frq.sort = FRQ_AVALUE;
601 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
602 | BIT_INDEX (FRQ_ST_STDDEV)
603 | BIT_INDEX (FRQ_ST_MINIMUM)
604 | BIT_INDEX (FRQ_ST_MAXIMUM);
608 frq.max_categories = LONG_MAX;
610 frq.percentiles = NULL;
611 frq.n_percentiles = 0;
618 /* Accept an optional, completely pointless "/VARIABLES=" */
619 lex_match (lexer, T_SLASH);
620 if (lex_match_id (lexer, "VARIABLES"))
622 if (! lex_force_match (lexer, T_EQUALS) )
626 if (!parse_variables_const (lexer, dataset_dict (ds),
632 frq.vars = xzalloc (frq.n_vars * sizeof (*frq.vars));
633 for (i = 0; i < frq.n_vars; ++i)
635 frq.vars[i].var = vars[i];
636 frq.vars[i].width = var_get_width (vars[i]);
639 while (lex_token (lexer) != T_ENDCMD)
641 lex_match (lexer, T_SLASH);
643 if (lex_match_id (lexer, "STATISTICS"))
645 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
646 | BIT_INDEX (FRQ_ST_STDDEV)
647 | BIT_INDEX (FRQ_ST_MINIMUM)
648 | BIT_INDEX (FRQ_ST_MAXIMUM);
652 if (lex_match (lexer, T_EQUALS))
658 while (lex_token (lexer) != T_ENDCMD
659 && lex_token (lexer) != T_SLASH)
661 if (lex_match_id (lexer, "DEFAULT"))
663 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
664 | BIT_INDEX (FRQ_ST_STDDEV)
665 | BIT_INDEX (FRQ_ST_MINIMUM)
666 | BIT_INDEX (FRQ_ST_MAXIMUM);
670 else if (lex_match_id (lexer, "MEAN"))
672 frq.stats |= BIT_INDEX (FRQ_ST_MEAN);
675 else if (lex_match_id (lexer, "SEMEAN"))
677 frq.stats |= BIT_INDEX (FRQ_ST_SEMEAN);
680 else if (lex_match_id (lexer, "MEDIAN"))
682 frq.stats |= BIT_INDEX (FRQ_ST_MEDIAN);
685 else if (lex_match_id (lexer, "MODE"))
687 frq.stats |= BIT_INDEX (FRQ_ST_MODE);
690 else if (lex_match_id (lexer, "STDDEV"))
692 frq.stats |= BIT_INDEX (FRQ_ST_STDDEV);
695 else if (lex_match_id (lexer, "VARIANCE"))
697 frq.stats |= BIT_INDEX (FRQ_ST_VARIANCE);
700 else if (lex_match_id (lexer, "KURTOSIS"))
702 frq.stats |= BIT_INDEX (FRQ_ST_KURTOSIS);
705 else if (lex_match_id (lexer, "SKEWNESS"))
707 frq.stats |= BIT_INDEX (FRQ_ST_SKEWNESS);
710 else if (lex_match_id (lexer, "RANGE"))
712 frq.stats |= BIT_INDEX (FRQ_ST_RANGE);
715 else if (lex_match_id (lexer, "MINIMUM"))
717 frq.stats |= BIT_INDEX (FRQ_ST_MINIMUM);
720 else if (lex_match_id (lexer, "MAXIMUM"))
722 frq.stats |= BIT_INDEX (FRQ_ST_MAXIMUM);
725 else if (lex_match_id (lexer, "SUM"))
727 frq.stats |= BIT_INDEX (FRQ_ST_SUM);
730 else if (lex_match_id (lexer, "SESKEWNESS"))
732 frq.stats |= BIT_INDEX (FRQ_ST_SESKEWNESS);
735 else if (lex_match_id (lexer, "SEKURTOSIS"))
737 frq.stats |= BIT_INDEX (FRQ_ST_SEKURTOSIS);
740 else if (lex_match_id (lexer, "NONE"))
745 else if (lex_match (lexer, T_ALL))
748 frq.n_stats = FRQ_ST_count;
752 lex_error (lexer, NULL);
757 else if (lex_match_id (lexer, "PERCENTILES"))
759 lex_match (lexer, T_EQUALS);
760 while (lex_token (lexer) != T_ENDCMD
761 && lex_token (lexer) != T_SLASH)
763 if (lex_force_num (lexer))
766 xrealloc (frq.percentiles,
767 (frq.n_percentiles + 1)
768 * sizeof (*frq.percentiles));
769 frq.percentiles[frq.n_percentiles].p = lex_number (lexer) / 100.0;
770 frq.percentiles[frq.n_percentiles].show = true;
776 lex_error (lexer, NULL);
779 lex_match (lexer, T_COMMA);
782 else if (lex_match_id (lexer, "FORMAT"))
784 lex_match (lexer, T_EQUALS);
785 while (lex_token (lexer) != T_ENDCMD
786 && lex_token (lexer) != T_SLASH)
788 if (lex_match_id (lexer, "TABLE"))
791 else if (lex_match_id (lexer, "NOTABLE"))
793 frq.max_categories = 0;
795 else if (lex_match_id (lexer, "LIMIT"))
797 if (!lex_force_match (lexer, T_LPAREN)
798 || !lex_force_int (lexer))
801 frq.max_categories = lex_integer (lexer);
804 if (!lex_force_match (lexer, T_RPAREN))
807 else if (lex_match_id (lexer, "AVALUE"))
809 frq.sort = FRQ_AVALUE;
811 else if (lex_match_id (lexer, "DVALUE"))
813 frq.sort = FRQ_DVALUE;
815 else if (lex_match_id (lexer, "AFREQ"))
817 frq.sort = FRQ_AFREQ;
819 else if (lex_match_id (lexer, "DFREQ"))
821 frq.sort = FRQ_DFREQ;
825 lex_error (lexer, NULL);
830 else if (lex_match_id (lexer, "NTILES"))
832 lex_match (lexer, T_EQUALS);
834 if (lex_force_int (lexer))
837 int n = lex_integer (lexer);
839 for (i = 0; i < n + 1; ++i)
842 xrealloc (frq.percentiles,
843 (frq.n_percentiles + 1)
844 * sizeof (*frq.percentiles));
845 frq.percentiles[frq.n_percentiles].p =
847 frq.percentiles[frq.n_percentiles].show = true;
854 lex_error (lexer, NULL);
858 else if (lex_match_id (lexer, "ALGORITHM"))
860 lex_match (lexer, T_EQUALS);
862 if (lex_match_id (lexer, "COMPATIBLE"))
864 settings_set_cmd_algorithm (COMPATIBLE);
866 else if (lex_match_id (lexer, "ENHANCED"))
868 settings_set_cmd_algorithm (ENHANCED);
872 lex_error (lexer, NULL);
876 else if (lex_match_id (lexer, "HISTOGRAM"))
878 lex_match (lexer, T_EQUALS);
879 sbc_histogram = true;
881 while (lex_token (lexer) != T_ENDCMD
882 && lex_token (lexer) != T_SLASH)
884 if (lex_match_id (lexer, "NORMAL"))
886 hi_norm = FRQ_NORMAL;
888 else if (lex_match_id (lexer, "NONORMAL"))
890 hi_norm = FRQ_NONORMAL;
892 else if (lex_match_id (lexer, "FREQ"))
895 if (lex_match (lexer, T_LPAREN))
897 if (lex_force_int (lexer))
899 hi_freq = lex_integer (lexer);
902 lex_error (lexer, _("Histogram frequency must be greater than zero."));
905 if (! lex_force_match (lexer, T_RPAREN))
910 else if (lex_match_id (lexer, "PERCENT"))
912 hi_scale = FRQ_PERCENT;
913 if (lex_match (lexer, T_LPAREN))
915 if (lex_force_int (lexer))
917 hi_pcnt = lex_integer (lexer);
920 lex_error (lexer, _("Histogram percentage must be greater than zero."));
923 if (! lex_force_match (lexer, T_RPAREN))
928 else if (lex_match_id (lexer, "MINIMUM"))
930 if (! lex_force_match (lexer, T_LPAREN))
932 if (lex_force_num (lexer))
934 hi_min = lex_number (lexer);
937 if (! lex_force_match (lexer, T_RPAREN))
940 else if (lex_match_id (lexer, "MAXIMUM"))
942 if (! lex_force_match (lexer, T_LPAREN))
944 if (lex_force_num (lexer))
946 hi_max = lex_number (lexer);
949 if (! lex_force_match (lexer, T_RPAREN))
954 lex_error (lexer, NULL);
959 else if (lex_match_id (lexer, "PIECHART"))
961 lex_match (lexer, T_EQUALS);
962 while (lex_token (lexer) != T_ENDCMD
963 && lex_token (lexer) != T_SLASH)
965 if (lex_match_id (lexer, "MINIMUM"))
967 if (! lex_force_match (lexer, T_LPAREN))
969 if (lex_force_num (lexer))
971 pie_min = lex_number (lexer);
974 if (! lex_force_match (lexer, T_RPAREN))
977 else if (lex_match_id (lexer, "MAXIMUM"))
979 if (! lex_force_match (lexer, T_LPAREN))
981 if (lex_force_num (lexer))
983 pie_max = lex_number (lexer);
986 if (! lex_force_match (lexer, T_RPAREN))
989 else if (lex_match_id (lexer, "MISSING"))
993 else if (lex_match_id (lexer, "NOMISSING"))
999 lex_error (lexer, NULL);
1003 sbc_piechart = true;
1005 else if (lex_match_id (lexer, "BARCHART"))
1007 lex_match (lexer, T_EQUALS);
1008 while (lex_token (lexer) != T_ENDCMD
1009 && lex_token (lexer) != T_SLASH)
1011 if (lex_match_id (lexer, "MINIMUM"))
1013 if (! lex_force_match (lexer, T_LPAREN))
1015 if (lex_force_num (lexer))
1017 bar_min = lex_number (lexer);
1020 if (! lex_force_match (lexer, T_RPAREN))
1023 else if (lex_match_id (lexer, "MAXIMUM"))
1025 if (! lex_force_match (lexer, T_LPAREN))
1027 if (lex_force_num (lexer))
1029 bar_max = lex_number (lexer);
1032 if (! lex_force_match (lexer, T_RPAREN))
1035 else if (lex_match_id (lexer, "FREQ"))
1037 if ( lex_match (lexer, T_LPAREN))
1039 if (lex_force_num (lexer))
1044 if (! lex_force_match (lexer, T_RPAREN))
1049 else if (lex_match_id (lexer, "PERCENT"))
1051 if ( lex_match (lexer, T_LPAREN))
1053 if (lex_force_num (lexer))
1058 if (! lex_force_match (lexer, T_RPAREN))
1065 lex_error (lexer, NULL);
1069 sbc_barchart = true;
1071 else if (lex_match_id (lexer, "MISSING"))
1073 lex_match (lexer, T_EQUALS);
1075 while (lex_token (lexer) != T_ENDCMD
1076 && lex_token (lexer) != T_SLASH)
1078 if (lex_match_id (lexer, "EXCLUDE"))
1081 else if (lex_match_id (lexer, "INCLUDE"))
1086 lex_error (lexer, NULL);
1091 else if (lex_match_id (lexer, "ORDER"))
1093 lex_match (lexer, T_EQUALS);
1094 if (!lex_match_id (lexer, "ANALYSIS"))
1095 lex_match_id (lexer, "VARIABLE");
1099 lex_error (lexer, NULL);
1104 if (frq.stats & BIT_INDEX (FRQ_ST_MEDIAN))
1107 xrealloc (frq.percentiles,
1108 (frq.n_percentiles + 1)
1109 * sizeof (*frq.percentiles));
1111 frq.percentiles[frq.n_percentiles].p = 0.50;
1112 frq.percentiles[frq.n_percentiles].show = false;
1114 frq.n_percentiles++;
1118 /* Figure out which charts the user requested. */
1123 struct frq_chart *hist;
1125 hist = frq.hist = xmalloc (sizeof *frq.hist);
1126 hist->x_min = hi_min;
1127 hist->x_max = hi_max;
1128 hist->y_scale = hi_scale;
1129 hist->y_max = hi_scale == FRQ_FREQ ? hi_freq : hi_pcnt;
1130 hist->draw_normal = hi_norm != FRQ_NONORMAL;
1131 hist->include_missing = false;
1133 if (hist->x_min != SYSMIS && hist->x_max != SYSMIS
1134 && hist->x_min >= hist->x_max)
1136 msg (SE, _("%s for histogram must be greater than or equal to %s, "
1137 "but %s was specified as %.15g and %s as %.15g. "
1138 "%s and %s will be ignored."),
1143 hist->x_min = hist->x_max = SYSMIS;
1147 xrealloc (frq.percentiles,
1148 (frq.n_percentiles + 2)
1149 * sizeof (*frq.percentiles));
1151 frq.percentiles[frq.n_percentiles].p = 0.25;
1152 frq.percentiles[frq.n_percentiles].show = false;
1154 frq.percentiles[frq.n_percentiles + 1].p = 0.75;
1155 frq.percentiles[frq.n_percentiles + 1].show = false;
1157 frq.n_percentiles+=2;
1162 frq.bar = xmalloc (sizeof *frq.bar);
1163 frq.bar->x_min = bar_min;
1164 frq.bar->x_max = bar_max;
1165 frq.bar->include_missing = false;
1166 frq.bar->y_scale = bar_freq ? FRQ_FREQ : FRQ_PERCENT;
1171 struct frq_chart *pie;
1173 pie = frq.pie = xmalloc (sizeof *frq.pie);
1174 pie->x_min = pie_min;
1175 pie->x_max = pie_max;
1176 pie->include_missing = pie_missing;
1178 if (pie->x_min != SYSMIS && pie->x_max != SYSMIS
1179 && pie->x_min >= pie->x_max)
1181 msg (SE, _("%s for pie chart must be greater than or equal to %s, "
1182 "but %s was specified as %.15g and %s as %.15g. "
1183 "%s and %s will be ignored."),
1188 pie->x_min = pie->x_max = SYSMIS;
1195 double previous_p = -1;
1196 qsort (frq.percentiles, frq.n_percentiles,
1197 sizeof (*frq.percentiles),
1200 for (i = o = 0; i < frq.n_percentiles; ++i)
1202 if (frq.percentiles[i].p != previous_p)
1204 frq.percentiles[o].p = frq.percentiles[i].p;
1205 frq.percentiles[o].show = frq.percentiles[i].show;
1208 else if (frq.percentiles[i].show &&
1209 !frq.percentiles[o].show)
1211 frq.percentiles[o].show = true;
1213 previous_p = frq.percentiles[i].p;
1216 frq.n_percentiles = o;
1219 for (i = 0; i < frq.n_percentiles; i++)
1220 if (frq.percentiles[i].p == 0.5)
1222 frq.median = &frq.percentiles[i];
1228 struct casegrouper *grouper;
1229 struct casereader *group;
1232 grouper = casegrouper_create_splits (proc_open (ds), dataset_dict (ds));
1233 while (casegrouper_get_next_group (grouper, &group))
1236 precalc (&frq, group, ds);
1238 for (; (c = casereader_read (group)) != NULL; case_unref (c))
1240 postcalc (&frq, ds);
1241 casereader_destroy (group);
1243 ok = casegrouper_destroy (grouper);
1244 ok = proc_commit (ds) && ok;
1253 free (frq.percentiles);
1254 pool_destroy (frq.pool);
1265 free (frq.percentiles);
1266 pool_destroy (frq.pool);
1272 calculate_iqr (const struct frq_proc *frq)
1278 /* This cannot work unless the 25th and 75th percentile are calculated */
1279 assert (frq->n_percentiles >= 2);
1280 for (i = 0; i < frq->n_percentiles; i++)
1282 struct percentile *pc = &frq->percentiles[i];
1284 if (fabs (0.25 - pc->p) < DBL_EPSILON)
1286 else if (fabs (0.75 - pc->p) < DBL_EPSILON)
1290 return q1 == SYSMIS || q3 == SYSMIS ? SYSMIS : q3 - q1;
1294 chart_includes_value (const struct frq_chart *chart,
1295 const struct variable *var,
1296 const union value *value)
1298 if (!chart->include_missing && var_is_value_missing (var, value, MV_ANY))
1301 if (var_is_numeric (var)
1302 && ((chart->x_min != SYSMIS && value->f < chart->x_min)
1303 || (chart->x_max != SYSMIS && value->f > chart->x_max)))
1309 /* Create a gsl_histogram from a freq_tab */
1310 static struct histogram *
1311 freq_tab_to_hist (const struct frq_proc *frq, const struct freq_tab *ft,
1312 const struct variable *var)
1314 double x_min, x_max, valid_freq;
1317 struct histogram *histogram;
1320 /* Find out the extremes of the x value, within the range to be included in
1321 the histogram, and sum the total frequency of those values. */
1325 for (i = 0; i < ft->n_valid; i++)
1327 const struct freq *f = &ft->valid[i];
1328 if (chart_includes_value (frq->hist, var, f->values))
1330 x_min = MIN (x_min, f->values[0].f);
1331 x_max = MAX (x_max, f->values[0].f);
1332 valid_freq += f->count;
1336 if (valid_freq <= 0)
1339 iqr = calculate_iqr (frq);
1342 /* Freedman-Diaconis' choice of bin width. */
1343 bin_width = 2 * iqr / pow (valid_freq, 1.0 / 3.0);
1347 bin_width = (x_max - x_min) / (1 + log2 (valid_freq));
1349 histogram = histogram_create (bin_width, x_min, x_max);
1351 if ( histogram == NULL)
1354 for (i = 0; i < ft->n_valid; i++)
1356 const struct freq *f = &ft->valid[i];
1357 if (chart_includes_value (frq->hist, var, f->values))
1358 histogram_add (histogram, f->values[0].f, f->count);
1365 /* Allocate an array of struct freqs and fill them from the data in FRQ_TAB,
1366 according to the parameters of CATCHART
1367 N_SLICES will contain the number of slices allocated.
1368 The caller is responsible for freeing slices
1370 static struct freq *
1371 pick_cat_counts (const struct frq_chart *catchart,
1372 const struct freq_tab *frq_tab,
1377 struct freq *slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1379 for (i = 0; i < frq_tab->n_valid; i++)
1381 const struct freq *f = &frq_tab->valid[i];
1382 if (f->count > catchart->x_max)
1385 if (f->count < catchart->x_min)
1388 slices[n_slices] = *f;
1393 if (catchart->include_missing)
1395 for (i = 0; i < frq_tab->n_missing; i++)
1397 const struct freq *f = &frq_tab->missing[i];
1398 slices[n_slices].count += f->count;
1401 slices[n_slices].values[0] = f->values[0];
1404 if (frq_tab->n_missing > 0)
1408 *n_slicesp = n_slices;
1413 /* Allocate an array of struct freqs and fill them from the data in FRQ_TAB,
1414 according to the parameters of CATCHART
1415 N_SLICES will contain the number of slices allocated.
1416 The caller is responsible for freeing slices
1418 static struct freq **
1419 pick_cat_counts_ptr (const struct frq_chart *catchart,
1420 const struct freq_tab *frq_tab,
1425 struct freq **slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1427 for (i = 0; i < frq_tab->n_valid; i++)
1429 struct freq *f = &frq_tab->valid[i];
1430 if (f->count > catchart->x_max)
1433 if (f->count < catchart->x_min)
1436 slices[n_slices] = f;
1441 if (catchart->include_missing)
1443 for (i = 0; i < frq_tab->n_missing; i++)
1445 const struct freq *f = &frq_tab->missing[i];
1448 slices[n_slices] = xmalloc (sizeof (struct freq));
1449 slices[n_slices]->values[0] = f->values[0];
1452 slices[n_slices]->count += f->count;
1457 *n_slicesp = n_slices;
1464 do_piechart(const struct frq_chart *pie, const struct variable *var,
1465 const struct freq_tab *frq_tab)
1468 struct freq *slices = pick_cat_counts (pie, frq_tab, &n_slices);
1471 msg (SW, _("Omitting pie chart for %s, which has only %d unique values."),
1472 var_get_name (var), n_slices);
1473 else if (n_slices > 50)
1474 msg (SW, _("Omitting pie chart for %s, which has over 50 unique values."),
1475 var_get_name (var));
1477 chart_item_submit (piechart_create (var, slices, n_slices));
1484 do_barchart(const struct frq_chart *bar, const struct variable **var,
1485 const struct freq_tab *frq_tab)
1488 struct freq **slices = pick_cat_counts_ptr (bar, frq_tab, &n_slices);
1490 chart_item_submit (barchart_create (var, 1,
1491 (bar->y_scale == FRQ_FREQ) ? _("Count") : _("Percent"),
1492 (bar->y_scale == FRQ_PERCENT),
1498 /* Calculates all the pertinent statistics for VF, putting them in array
1501 calc_stats (const struct frq_proc *frq, const struct var_freqs *vf,
1502 double d[FRQ_ST_count])
1504 const struct freq_tab *ft = &vf->tab;
1505 double W = ft->valid_cases;
1506 const struct freq *f;
1508 int most_often = -1;
1509 double X_mode = SYSMIS;
1511 /* Calculate the mode. */
1512 for (f = ft->valid; f < ft->missing; f++)
1514 if (most_often < f->count)
1516 most_often = f->count;
1517 X_mode = f->values[0].f;
1519 else if (most_often == f->count)
1521 /* A duplicate mode is undefined.
1522 FIXME: keep track of *all* the modes. */
1527 /* Calculate moments. */
1528 m = moments_create (MOMENT_KURTOSIS);
1529 for (f = ft->valid; f < ft->missing; f++)
1530 moments_pass_one (m, f->values[0].f, f->count);
1531 for (f = ft->valid; f < ft->missing; f++)
1532 moments_pass_two (m, f->values[0].f, f->count);
1533 moments_calculate (m, NULL, &d[FRQ_ST_MEAN], &d[FRQ_ST_VARIANCE],
1534 &d[FRQ_ST_SKEWNESS], &d[FRQ_ST_KURTOSIS]);
1535 moments_destroy (m);
1537 /* Formulae below are taken from _SPSS Statistical Algorithms_. */
1538 if (ft->n_valid > 0)
1540 d[FRQ_ST_MINIMUM] = ft->valid[0].values[0].f;
1541 d[FRQ_ST_MAXIMUM] = ft->valid[ft->n_valid - 1].values[0].f;
1542 d[FRQ_ST_RANGE] = d[FRQ_ST_MAXIMUM] - d[FRQ_ST_MINIMUM];
1546 d[FRQ_ST_MINIMUM] = SYSMIS;
1547 d[FRQ_ST_MAXIMUM] = SYSMIS;
1548 d[FRQ_ST_RANGE] = SYSMIS;
1550 d[FRQ_ST_MODE] = X_mode;
1551 d[FRQ_ST_SUM] = d[FRQ_ST_MEAN] * W;
1552 d[FRQ_ST_STDDEV] = sqrt (d[FRQ_ST_VARIANCE]);
1553 d[FRQ_ST_SEMEAN] = d[FRQ_ST_STDDEV] / sqrt (W);
1554 d[FRQ_ST_SESKEWNESS] = calc_seskew (W);
1555 d[FRQ_ST_SEKURTOSIS] = calc_sekurt (W);
1556 d[FRQ_ST_MEDIAN] = frq->median ? frq->median->value : SYSMIS;
1560 all_string_variables (const struct frq_proc *frq)
1562 for (size_t i = 0; i < frq->n_vars; i++)
1563 if (var_is_numeric (frq->vars[i].var))
1569 /* Displays a table of all the statistics requested. */
1571 dump_statistics (const struct frq_proc *frq, const struct variable *wv)
1573 if (all_string_variables (frq))
1576 struct pivot_table *table = pivot_table_create (N_("Statistics"));
1577 pivot_table_set_weight_var (table, wv);
1579 struct pivot_dimension *variables
1580 = pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Variables"));
1582 struct pivot_dimension *statistics = pivot_dimension_create (
1583 table, PIVOT_AXIS_ROW, N_("Statistics"));
1584 struct pivot_category *n = pivot_category_create_group (
1585 statistics->root, N_("N"));
1586 pivot_category_create_leaves (n,
1587 N_("Valid"), PIVOT_RC_COUNT,
1588 N_("Missing"), PIVOT_RC_COUNT);
1589 for (int i = 0; i < FRQ_ST_count; i++)
1590 if (frq->stats & BIT_INDEX (i))
1591 pivot_category_create_leaf (statistics->root,
1592 pivot_value_new_text (st_name[i]));
1593 struct pivot_category *percentiles = NULL;
1594 for (size_t i = 0; i < frq->n_percentiles; i++)
1596 const struct percentile *pc = &frq->percentiles[i];
1602 percentiles = pivot_category_create_group (
1603 statistics->root, N_("Percentiles"));
1604 pivot_category_create_leaf (percentiles, pivot_value_new_integer (
1608 for (size_t i = 0; i < frq->n_vars; i++)
1610 struct var_freqs *vf = &frq->vars[i];
1611 if (var_is_alpha (vf->var))
1614 const struct freq_tab *ft = &vf->tab;
1616 int var_idx = pivot_category_create_leaf (
1617 variables->root, pivot_value_new_variable (vf->var));
1620 pivot_table_put2 (table, var_idx, row++,
1621 pivot_value_new_number (ft->valid_cases));
1622 pivot_table_put2 (table, var_idx, row++,
1623 pivot_value_new_number (
1624 ft->total_cases - ft->valid_cases));
1626 double stat_values[FRQ_ST_count];
1627 calc_stats (frq, vf, stat_values);
1628 for (int j = 0; j < FRQ_ST_count; j++)
1630 if (!(frq->stats & BIT_INDEX (j)))
1633 union value v = { .f = vf->tab.n_valid ? stat_values[j] : SYSMIS };
1634 struct pivot_value *pv
1635 = (j == FRQ_ST_MODE || j == FRQ_ST_MINIMUM || j == FRQ_ST_MAXIMUM
1636 ? pivot_value_new_var_value (vf->var, &v)
1637 : pivot_value_new_number (v.f));
1638 pivot_table_put2 (table, var_idx, row++, pv);
1641 for (size_t j = 0; j < frq->n_percentiles; j++)
1643 const struct percentile *pc = &frq->percentiles[j];
1647 union value v = { .f = vf->tab.n_valid ? pc->value : SYSMIS };
1648 pivot_table_put2 (table, var_idx, row++,
1649 pivot_value_new_var_value (vf->var, &v));
1653 pivot_table_submit (table);