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/charts/barchart.h"
56 #include "output/charts/piechart.h"
57 #include "output/charts/plot-hist.h"
58 #include "output/pivot-table.h"
60 #include "gl/minmax.h"
61 #include "gl/xalloc.h"
64 #define _(msgid) gettext (msgid)
65 #define N_(msgid) msgid
67 /* Percentiles to calculate. */
71 double p; /* the %ile to be calculated */
72 double value; /* the %ile's value */
73 bool show; /* True to show this percentile in the statistics box. */
77 ptile_3way (const void *_p1, const void *_p2)
79 const struct percentile *p1 = _p1;
80 const struct percentile *p2 = _p2;
87 if (p1->show > p2->show)
90 return (p1->show < p2->show);
93 return (p1->p > p2->p);
117 /* Array indices for STATISTICS subcommand. */
137 /* Description of statistics. */
138 static const char *st_name[FRQ_ST_count] =
158 struct hmap data; /* Hash table for accumulating counts. */
159 struct freq *valid; /* Valid freqs. */
160 int n_valid; /* Number of total freqs. */
161 const struct dictionary *dict; /* Source of entries in the table. */
163 struct freq *missing; /* Missing freqs. */
164 int n_missing; /* Number of missing freqs. */
167 double total_cases; /* Sum of weights of all cases. */
168 double valid_cases; /* Sum of weights of valid cases. */
173 double x_min; /* X axis minimum value. */
174 double x_max; /* X axis maximum value. */
175 int y_scale; /* Y axis scale: FRQ_FREQ or FRQ_PERCENT. */
177 /* Histograms only. */
178 double y_max; /* Y axis maximum value. */
179 bool draw_normal; /* Whether to draw normal curve. */
181 /* Pie charts only. */
182 bool include_missing; /* Whether to include missing values. */
185 /* Per-variable frequency data. */
188 const struct variable *var;
190 /* Freqency table. */
191 struct freq_tab tab; /* Frequencies table to use. */
194 int n_groups; /* Number of groups. */
195 double *groups; /* Groups. */
198 double stat[FRQ_ST_count];
200 /* Variable attributes. */
208 struct var_freqs *vars;
211 /* Percentiles to calculate and possibly display. */
212 struct percentile *percentiles;
213 const struct percentile *median;
216 /* Frequency table display. */
217 long int max_categories; /* Maximum categories to show. */
218 int sort; /* FRQ_AVALUE or FRQ_DVALUE
219 or FRQ_AFREQ or FRQ_DFREQ. */
221 /* Statistics; number of statistics. */
225 /* Histogram and pie chart settings. */
226 struct frq_chart *hist, *pie, *bar;
232 struct freq_compare_aux
238 bool ascending_value;
241 static void calc_stats (const struct frq_proc *,
242 const struct var_freqs *, double d[FRQ_ST_count]);
244 static void do_piechart(const struct frq_chart *pie,
245 const struct variable *var,
246 const struct freq_tab *frq_tab);
248 static void do_barchart(const struct frq_chart *bar,
249 const struct variable **var,
250 const struct freq_tab *frq_tab);
252 static void dump_statistics (const struct frq_proc *frq,
253 const struct variable *wv);
256 compare_freq (const void *a_, const void *b_, const void *aux_)
258 const struct freq_compare_aux *aux = aux_;
259 const struct freq *a = a_;
260 const struct freq *b = b_;
262 if (aux->by_freq && a->count != b->count)
264 int cmp = a->count > b->count ? 1 : -1;
265 return aux->ascending_freq ? cmp : -cmp;
269 int cmp = value_compare_3way (a->values, b->values, aux->width);
270 return aux->ascending_value ? cmp : -cmp;
274 /* Create a gsl_histogram from a freq_tab */
275 static struct histogram *
276 freq_tab_to_hist (const struct frq_proc *frq, const struct freq_tab *ft,
277 const struct variable *var);
280 put_freq_row (struct pivot_table *table, int var_idx,
281 double frequency, double percent,
282 double valid_percent, double cum_percent)
284 double entries[] = { frequency, percent, valid_percent, cum_percent };
285 for (size_t i = 0; i < sizeof entries / sizeof *entries; i++)
286 if (entries[i] != SYSMIS)
287 pivot_table_put2 (table, i, var_idx,
288 pivot_value_new_number (entries[i]));
291 /* Displays a full frequency table for variable V. */
293 dump_freq_table (const struct var_freqs *vf, const struct variable *wv)
295 const struct freq_tab *ft = &vf->tab;
297 struct pivot_table *table = pivot_table_create__ (pivot_value_new_variable (
298 vf->var), "Frequencies");
299 pivot_table_set_weight_var (table, wv);
301 pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Statistics"),
302 N_("Frequency"), PIVOT_RC_COUNT,
303 N_("Percent"), PIVOT_RC_PERCENT,
304 N_("Valid Percent"), PIVOT_RC_PERCENT,
305 N_("Cumulative Percent"), PIVOT_RC_PERCENT);
307 struct pivot_dimension *variable = pivot_dimension_create__ (
308 table, PIVOT_AXIS_ROW, pivot_value_new_variable (vf->var));
310 double cum_freq = 0.0;
311 double cum_percent = 0.0;
312 struct pivot_category *valid = NULL;
313 for (const struct freq *f = ft->valid; f < ft->missing; f++)
315 cum_freq += f->count;
316 double valid_percent = f->count / ft->valid_cases * 100.0;
317 cum_percent += valid_percent;
320 valid = pivot_category_create_group (variable->root, N_("Valid"));
321 int var_idx = pivot_category_create_leaf (
322 valid, pivot_value_new_var_value (vf->var, &f->values[0]));
323 put_freq_row (table, var_idx, f->count,
324 f->count / ft->total_cases * 100.0,
325 valid_percent, cum_percent);
328 struct pivot_category *missing = NULL;
329 size_t n_categories = ft->n_valid + ft->n_missing;
330 for (const struct freq *f = ft->missing; f < &ft->valid[n_categories]; f++)
332 cum_freq += f->count;
335 missing = pivot_category_create_group (variable->root, N_("Missing"));
336 int var_idx = pivot_category_create_leaf (
337 missing, pivot_value_new_var_value (vf->var, &f->values[0]));
338 put_freq_row (table, var_idx, f->count,
339 f->count / ft->total_cases * 100.0, SYSMIS, SYSMIS);
342 int var_idx = pivot_category_create_leaf (
343 variable->root, pivot_value_new_text (N_("Total")));
344 put_freq_row (table, var_idx, cum_freq, cum_percent, SYSMIS, SYSMIS);
346 pivot_table_submit (table);
349 /* Statistical display. */
352 calc_percentile (double p, double valid_cases, double x1, double x2)
356 s = (settings_get_algorithm () != COMPATIBLE
357 ? modf ((valid_cases - 1) * p, &dummy)
358 : modf ((valid_cases + 1) * p - 1, &dummy));
360 return x1 + (x2 - x1) * s;
363 /* Calculates all of the percentiles for VF within FRQ. */
365 calc_percentiles (const struct frq_proc *frq, const struct var_freqs *vf)
367 const struct freq_tab *ft = &vf->tab;
368 double W = ft->valid_cases;
369 const struct freq *f;
370 int percentile_idx = 0;
373 for (f = ft->valid; f < ft->missing; f++)
376 for (; percentile_idx < frq->n_percentiles; percentile_idx++)
378 struct percentile *pc = &frq->percentiles[percentile_idx];
381 tp = (settings_get_algorithm () == ENHANCED
383 : (W + 1) * pc->p - 1);
388 if (tp + 1 < rank || f + 1 >= ft->missing)
389 pc->value = f->values[0].f;
391 pc->value = calc_percentile (pc->p, W, f->values[0].f, f[1].values[0].f);
394 for (; percentile_idx < frq->n_percentiles; percentile_idx++)
396 struct percentile *pc = &frq->percentiles[percentile_idx];
397 pc->value = (ft->n_valid > 0
398 ? ft->valid[ft->n_valid - 1].values[0].f
403 /* Returns true iff the value in struct freq F is non-missing
406 not_missing (const void *f_, const void *v_)
408 const struct freq *f = f_;
409 const struct variable *v = v_;
411 return !var_is_value_missing (v, f->values, MV_ANY);
415 /* Summarizes the frequency table data for variable V. */
417 postprocess_freq_tab (const struct frq_proc *frq, struct var_freqs *vf)
419 struct freq_tab *ft = &vf->tab;
420 struct freq_compare_aux aux;
422 struct freq *freqs, *f;
425 /* Extract data from hash table. */
426 count = hmap_count (&ft->data);
427 freqs = freq_hmap_extract (&ft->data);
429 /* Put data into ft. */
431 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, vf->var);
432 ft->missing = freqs + ft->n_valid;
433 ft->n_missing = count - ft->n_valid;
436 aux.by_freq = frq->sort == FRQ_AFREQ || frq->sort == FRQ_DFREQ;
437 aux.ascending_freq = frq->sort != FRQ_DFREQ;
438 aux.width = vf->width;
439 aux.ascending_value = frq->sort != FRQ_DVALUE;
440 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare_freq, &aux);
441 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare_freq, &aux);
443 /* Summary statistics. */
444 ft->valid_cases = 0.0;
445 for(i = 0 ; i < ft->n_valid ; ++i)
448 ft->valid_cases += f->count;
452 ft->total_cases = ft->valid_cases ;
453 for(i = 0 ; i < ft->n_missing ; ++i)
456 ft->total_cases += f->count;
461 /* Frees the frequency table for variable V. */
463 cleanup_freq_tab (struct var_freqs *vf)
465 free (vf->tab.valid);
466 freq_hmap_destroy (&vf->tab.data, vf->width);
469 /* Add data from case C to the frequency table. */
471 calc (struct frq_proc *frq, const struct ccase *c, const struct dataset *ds)
473 double weight = dict_get_case_weight (dataset_dict (ds), c, &frq->warn);
476 for (i = 0; i < frq->n_vars; i++)
478 struct var_freqs *vf = &frq->vars[i];
479 const union value *value = case_data (c, vf->var);
480 size_t hash = value_hash (value, vf->width, 0);
483 f = freq_hmap_search (&vf->tab.data, value, vf->width, hash);
485 f = freq_hmap_insert (&vf->tab.data, value, vf->width, hash);
491 /* Prepares each variable that is the target of FREQUENCIES by setting
492 up its hash table. */
494 precalc (struct frq_proc *frq, struct casereader *input, struct dataset *ds)
499 c = casereader_peek (input, 0);
502 output_split_file_values (ds, c);
506 for (i = 0; i < frq->n_vars; i++)
507 hmap_init (&frq->vars[i].tab.data);
510 /* Finishes up with the variables after frequencies have been
511 calculated. Displays statistics, percentiles, ... */
513 postcalc (struct frq_proc *frq, const struct dataset *ds)
515 const struct dictionary *dict = dataset_dict (ds);
516 const struct variable *wv = dict_get_weight (dict);
519 for (i = 0; i < frq->n_vars; i++)
521 struct var_freqs *vf = &frq->vars[i];
522 postprocess_freq_tab (frq, vf);
523 calc_percentiles (frq, vf);
527 dump_statistics (frq, wv);
529 for (i = 0; i < frq->n_vars; i++)
531 struct var_freqs *vf = &frq->vars[i];
533 /* Frequencies tables. */
534 if (vf->tab.n_valid + vf->tab.n_missing <= frq->max_categories)
535 dump_freq_table (vf, wv);
538 if (frq->hist && var_is_numeric (vf->var) && vf->tab.n_valid > 0)
540 double d[FRQ_ST_count];
541 struct histogram *histogram;
543 calc_stats (frq, vf, d);
545 histogram = freq_tab_to_hist (frq, &vf->tab, vf->var);
549 chart_submit (histogram_chart_create (
550 histogram->gsl_hist, var_to_string(vf->var),
554 frq->hist->draw_normal));
556 statistic_destroy (&histogram->parent);
561 do_piechart(frq->pie, vf->var, &vf->tab);
564 do_barchart(frq->bar, &vf->var, &vf->tab);
566 cleanup_freq_tab (vf);
571 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
575 const struct variable **vars = NULL;
577 bool sbc_barchart = false;
578 bool sbc_piechart = false;
579 bool sbc_histogram = false;
581 double pie_min = -DBL_MAX;
582 double pie_max = DBL_MAX;
583 bool pie_missing = true;
585 double bar_min = -DBL_MAX;
586 double bar_max = DBL_MAX;
587 bool bar_freq = true;
589 double hi_min = -DBL_MAX;
590 double hi_max = DBL_MAX;
591 int hi_scale = FRQ_FREQ;
592 int hi_freq = INT_MIN;
593 int hi_pcnt = INT_MIN;
594 int hi_norm = FRQ_NONORMAL;
596 frq.pool = pool_create ();
597 frq.sort = FRQ_AVALUE;
602 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
603 | BIT_INDEX (FRQ_ST_STDDEV)
604 | BIT_INDEX (FRQ_ST_MINIMUM)
605 | BIT_INDEX (FRQ_ST_MAXIMUM);
609 frq.max_categories = LONG_MAX;
611 frq.percentiles = NULL;
612 frq.n_percentiles = 0;
620 /* Accept an optional, completely pointless "/VARIABLES=" */
621 lex_match (lexer, T_SLASH);
622 if (lex_match_id (lexer, "VARIABLES"))
624 if (! lex_force_match (lexer, T_EQUALS))
628 if (!parse_variables_const (lexer, dataset_dict (ds),
634 frq.vars = xzalloc (frq.n_vars * sizeof (*frq.vars));
635 for (i = 0; i < frq.n_vars; ++i)
637 frq.vars[i].var = vars[i];
638 frq.vars[i].width = var_get_width (vars[i]);
641 while (lex_token (lexer) != T_ENDCMD)
643 lex_match (lexer, T_SLASH);
645 if (lex_match_id (lexer, "STATISTICS"))
647 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
648 | BIT_INDEX (FRQ_ST_STDDEV)
649 | BIT_INDEX (FRQ_ST_MINIMUM)
650 | BIT_INDEX (FRQ_ST_MAXIMUM);
654 if (lex_match (lexer, T_EQUALS))
660 while (lex_token (lexer) != T_ENDCMD
661 && lex_token (lexer) != T_SLASH)
663 if (lex_match_id (lexer, "DEFAULT"))
665 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
666 | BIT_INDEX (FRQ_ST_STDDEV)
667 | BIT_INDEX (FRQ_ST_MINIMUM)
668 | BIT_INDEX (FRQ_ST_MAXIMUM);
672 else if (lex_match_id (lexer, "MEAN"))
674 frq.stats |= BIT_INDEX (FRQ_ST_MEAN);
677 else if (lex_match_id (lexer, "SEMEAN"))
679 frq.stats |= BIT_INDEX (FRQ_ST_SEMEAN);
682 else if (lex_match_id (lexer, "MEDIAN"))
684 frq.stats |= BIT_INDEX (FRQ_ST_MEDIAN);
687 else if (lex_match_id (lexer, "MODE"))
689 frq.stats |= BIT_INDEX (FRQ_ST_MODE);
692 else if (lex_match_id (lexer, "STDDEV"))
694 frq.stats |= BIT_INDEX (FRQ_ST_STDDEV);
697 else if (lex_match_id (lexer, "VARIANCE"))
699 frq.stats |= BIT_INDEX (FRQ_ST_VARIANCE);
702 else if (lex_match_id (lexer, "KURTOSIS"))
704 frq.stats |= BIT_INDEX (FRQ_ST_KURTOSIS);
707 else if (lex_match_id (lexer, "SKEWNESS"))
709 frq.stats |= BIT_INDEX (FRQ_ST_SKEWNESS);
712 else if (lex_match_id (lexer, "RANGE"))
714 frq.stats |= BIT_INDEX (FRQ_ST_RANGE);
717 else if (lex_match_id (lexer, "MINIMUM"))
719 frq.stats |= BIT_INDEX (FRQ_ST_MINIMUM);
722 else if (lex_match_id (lexer, "MAXIMUM"))
724 frq.stats |= BIT_INDEX (FRQ_ST_MAXIMUM);
727 else if (lex_match_id (lexer, "SUM"))
729 frq.stats |= BIT_INDEX (FRQ_ST_SUM);
732 else if (lex_match_id (lexer, "SESKEWNESS"))
734 frq.stats |= BIT_INDEX (FRQ_ST_SESKEWNESS);
737 else if (lex_match_id (lexer, "SEKURTOSIS"))
739 frq.stats |= BIT_INDEX (FRQ_ST_SEKURTOSIS);
742 else if (lex_match_id (lexer, "NONE"))
747 else if (lex_match (lexer, T_ALL))
750 frq.n_stats = FRQ_ST_count;
754 lex_error (lexer, NULL);
759 else if (lex_match_id (lexer, "PERCENTILES"))
761 lex_match (lexer, T_EQUALS);
762 while (lex_token (lexer) != T_ENDCMD
763 && lex_token (lexer) != T_SLASH)
765 if (lex_force_num (lexer))
768 xrealloc (frq.percentiles,
769 (frq.n_percentiles + 1)
770 * sizeof (*frq.percentiles));
771 frq.percentiles[frq.n_percentiles].p = lex_number (lexer) / 100.0;
772 frq.percentiles[frq.n_percentiles].show = true;
778 lex_error (lexer, NULL);
781 lex_match (lexer, T_COMMA);
784 else if (lex_match_id (lexer, "FORMAT"))
786 lex_match (lexer, T_EQUALS);
787 while (lex_token (lexer) != T_ENDCMD
788 && lex_token (lexer) != T_SLASH)
790 if (lex_match_id (lexer, "TABLE"))
793 else if (lex_match_id (lexer, "NOTABLE"))
795 frq.max_categories = 0;
797 else if (lex_match_id (lexer, "LIMIT"))
799 if (!lex_force_match (lexer, T_LPAREN)
800 || !lex_force_int (lexer))
803 frq.max_categories = lex_integer (lexer);
806 if (!lex_force_match (lexer, T_RPAREN))
809 else if (lex_match_id (lexer, "AVALUE"))
811 frq.sort = FRQ_AVALUE;
813 else if (lex_match_id (lexer, "DVALUE"))
815 frq.sort = FRQ_DVALUE;
817 else if (lex_match_id (lexer, "AFREQ"))
819 frq.sort = FRQ_AFREQ;
821 else if (lex_match_id (lexer, "DFREQ"))
823 frq.sort = FRQ_DFREQ;
827 lex_error (lexer, NULL);
832 else if (lex_match_id (lexer, "NTILES"))
834 lex_match (lexer, T_EQUALS);
836 if (lex_force_int (lexer))
839 int n = lex_integer (lexer);
841 for (i = 0; i < n + 1; ++i)
844 xrealloc (frq.percentiles,
845 (frq.n_percentiles + 1)
846 * sizeof (*frq.percentiles));
847 frq.percentiles[frq.n_percentiles].p =
849 frq.percentiles[frq.n_percentiles].show = true;
856 lex_error (lexer, NULL);
860 else if (lex_match_id (lexer, "ALGORITHM"))
862 lex_match (lexer, T_EQUALS);
864 if (lex_match_id (lexer, "COMPATIBLE"))
866 settings_set_cmd_algorithm (COMPATIBLE);
868 else if (lex_match_id (lexer, "ENHANCED"))
870 settings_set_cmd_algorithm (ENHANCED);
874 lex_error (lexer, NULL);
878 else if (lex_match_id (lexer, "HISTOGRAM"))
880 lex_match (lexer, T_EQUALS);
881 sbc_histogram = true;
883 while (lex_token (lexer) != T_ENDCMD
884 && lex_token (lexer) != T_SLASH)
886 if (lex_match_id (lexer, "NORMAL"))
888 hi_norm = FRQ_NORMAL;
890 else if (lex_match_id (lexer, "NONORMAL"))
892 hi_norm = FRQ_NONORMAL;
894 else if (lex_match_id (lexer, "FREQ"))
897 if (lex_match (lexer, T_LPAREN))
899 if (lex_force_int (lexer))
901 hi_freq = lex_integer (lexer);
904 lex_error (lexer, _("Histogram frequency must be greater than zero."));
907 if (! lex_force_match (lexer, T_RPAREN))
912 else if (lex_match_id (lexer, "PERCENT"))
914 hi_scale = FRQ_PERCENT;
915 if (lex_match (lexer, T_LPAREN))
917 if (lex_force_int (lexer))
919 hi_pcnt = lex_integer (lexer);
922 lex_error (lexer, _("Histogram percentage must be greater than zero."));
925 if (! lex_force_match (lexer, T_RPAREN))
930 else if (lex_match_id (lexer, "MINIMUM"))
932 if (! lex_force_match (lexer, T_LPAREN))
934 if (lex_force_num (lexer))
936 hi_min = lex_number (lexer);
939 if (! lex_force_match (lexer, T_RPAREN))
942 else if (lex_match_id (lexer, "MAXIMUM"))
944 if (! lex_force_match (lexer, T_LPAREN))
946 if (lex_force_num (lexer))
948 hi_max = lex_number (lexer);
951 if (! lex_force_match (lexer, T_RPAREN))
956 lex_error (lexer, NULL);
961 else if (lex_match_id (lexer, "PIECHART"))
963 lex_match (lexer, T_EQUALS);
964 while (lex_token (lexer) != T_ENDCMD
965 && lex_token (lexer) != T_SLASH)
967 if (lex_match_id (lexer, "MINIMUM"))
969 if (! lex_force_match (lexer, T_LPAREN))
971 if (lex_force_num (lexer))
973 pie_min = lex_number (lexer);
976 if (! lex_force_match (lexer, T_RPAREN))
979 else if (lex_match_id (lexer, "MAXIMUM"))
981 if (! lex_force_match (lexer, T_LPAREN))
983 if (lex_force_num (lexer))
985 pie_max = lex_number (lexer);
988 if (! lex_force_match (lexer, T_RPAREN))
991 else if (lex_match_id (lexer, "MISSING"))
995 else if (lex_match_id (lexer, "NOMISSING"))
1001 lex_error (lexer, NULL);
1005 sbc_piechart = true;
1007 else if (lex_match_id (lexer, "BARCHART"))
1009 lex_match (lexer, T_EQUALS);
1010 while (lex_token (lexer) != T_ENDCMD
1011 && lex_token (lexer) != T_SLASH)
1013 if (lex_match_id (lexer, "MINIMUM"))
1015 if (! lex_force_match (lexer, T_LPAREN))
1017 if (lex_force_num (lexer))
1019 bar_min = lex_number (lexer);
1022 if (! lex_force_match (lexer, T_RPAREN))
1025 else if (lex_match_id (lexer, "MAXIMUM"))
1027 if (! lex_force_match (lexer, T_LPAREN))
1029 if (lex_force_num (lexer))
1031 bar_max = lex_number (lexer);
1034 if (! lex_force_match (lexer, T_RPAREN))
1037 else if (lex_match_id (lexer, "FREQ"))
1039 if (lex_match (lexer, T_LPAREN))
1041 if (lex_force_num (lexer))
1046 if (! lex_force_match (lexer, T_RPAREN))
1051 else if (lex_match_id (lexer, "PERCENT"))
1053 if (lex_match (lexer, T_LPAREN))
1055 if (lex_force_num (lexer))
1060 if (! lex_force_match (lexer, T_RPAREN))
1067 lex_error (lexer, NULL);
1071 sbc_barchart = true;
1073 else if (lex_match_id (lexer, "MISSING"))
1075 lex_match (lexer, T_EQUALS);
1077 while (lex_token (lexer) != T_ENDCMD
1078 && lex_token (lexer) != T_SLASH)
1080 if (lex_match_id (lexer, "EXCLUDE"))
1083 else if (lex_match_id (lexer, "INCLUDE"))
1088 lex_error (lexer, NULL);
1093 else if (lex_match_id (lexer, "ORDER"))
1095 lex_match (lexer, T_EQUALS);
1096 if (!lex_match_id (lexer, "ANALYSIS"))
1097 lex_match_id (lexer, "VARIABLE");
1101 lex_error (lexer, NULL);
1106 if (frq.stats & BIT_INDEX (FRQ_ST_MEDIAN))
1109 xrealloc (frq.percentiles,
1110 (frq.n_percentiles + 1)
1111 * sizeof (*frq.percentiles));
1113 frq.percentiles[frq.n_percentiles].p = 0.50;
1114 frq.percentiles[frq.n_percentiles].show = false;
1116 frq.n_percentiles++;
1120 /* Figure out which charts the user requested. */
1125 struct frq_chart *hist;
1127 hist = frq.hist = xmalloc (sizeof *frq.hist);
1128 hist->x_min = hi_min;
1129 hist->x_max = hi_max;
1130 hist->y_scale = hi_scale;
1131 hist->y_max = hi_scale == FRQ_FREQ ? hi_freq : hi_pcnt;
1132 hist->draw_normal = hi_norm != FRQ_NONORMAL;
1133 hist->include_missing = false;
1135 if (hist->x_min != SYSMIS && hist->x_max != SYSMIS
1136 && hist->x_min >= hist->x_max)
1138 msg (SE, _("%s for histogram must be greater than or equal to %s, "
1139 "but %s was specified as %.15g and %s as %.15g. "
1140 "%s and %s will be ignored."),
1145 hist->x_min = hist->x_max = SYSMIS;
1149 xrealloc (frq.percentiles,
1150 (frq.n_percentiles + 2)
1151 * sizeof (*frq.percentiles));
1153 frq.percentiles[frq.n_percentiles].p = 0.25;
1154 frq.percentiles[frq.n_percentiles].show = false;
1156 frq.percentiles[frq.n_percentiles + 1].p = 0.75;
1157 frq.percentiles[frq.n_percentiles + 1].show = false;
1159 frq.n_percentiles+=2;
1164 frq.bar = xmalloc (sizeof *frq.bar);
1165 frq.bar->x_min = bar_min;
1166 frq.bar->x_max = bar_max;
1167 frq.bar->include_missing = false;
1168 frq.bar->y_scale = bar_freq ? FRQ_FREQ : FRQ_PERCENT;
1173 struct frq_chart *pie;
1175 pie = frq.pie = xmalloc (sizeof *frq.pie);
1176 pie->x_min = pie_min;
1177 pie->x_max = pie_max;
1178 pie->include_missing = pie_missing;
1180 if (pie->x_min != SYSMIS && pie->x_max != SYSMIS
1181 && pie->x_min >= pie->x_max)
1183 msg (SE, _("%s for pie chart must be greater than or equal to %s, "
1184 "but %s was specified as %.15g and %s as %.15g. "
1185 "%s and %s will be ignored."),
1190 pie->x_min = pie->x_max = SYSMIS;
1197 double previous_p = -1;
1198 qsort (frq.percentiles, frq.n_percentiles,
1199 sizeof (*frq.percentiles),
1202 for (i = o = 0; i < frq.n_percentiles; ++i)
1204 if (frq.percentiles[i].p != previous_p)
1206 frq.percentiles[o].p = frq.percentiles[i].p;
1207 frq.percentiles[o].show = frq.percentiles[i].show;
1210 else if (frq.percentiles[i].show &&
1211 !frq.percentiles[o].show)
1213 frq.percentiles[o].show = true;
1215 previous_p = frq.percentiles[i].p;
1218 frq.n_percentiles = o;
1221 for (i = 0; i < frq.n_percentiles; i++)
1222 if (frq.percentiles[i].p == 0.5)
1224 frq.median = &frq.percentiles[i];
1230 struct casegrouper *grouper;
1231 struct casereader *group;
1234 grouper = casegrouper_create_splits (proc_open (ds), dataset_dict (ds));
1235 while (casegrouper_get_next_group (grouper, &group))
1238 precalc (&frq, group, ds);
1240 for (; (c = casereader_read (group)) != NULL; case_unref (c))
1242 postcalc (&frq, ds);
1243 casereader_destroy (group);
1245 ok = casegrouper_destroy (grouper);
1246 ok = proc_commit (ds) && ok;
1255 free (frq.percentiles);
1256 pool_destroy (frq.pool);
1267 free (frq.percentiles);
1268 pool_destroy (frq.pool);
1274 calculate_iqr (const struct frq_proc *frq)
1280 /* This cannot work unless the 25th and 75th percentile are calculated */
1281 assert (frq->n_percentiles >= 2);
1282 for (i = 0; i < frq->n_percentiles; i++)
1284 struct percentile *pc = &frq->percentiles[i];
1286 if (fabs (0.25 - pc->p) < DBL_EPSILON)
1288 else if (fabs (0.75 - pc->p) < DBL_EPSILON)
1292 return q1 == SYSMIS || q3 == SYSMIS ? SYSMIS : q3 - q1;
1296 chart_includes_value (const struct frq_chart *chart,
1297 const struct variable *var,
1298 const union value *value)
1300 if (!chart->include_missing && var_is_value_missing (var, value, MV_ANY))
1303 if (var_is_numeric (var)
1304 && ((chart->x_min != SYSMIS && value->f < chart->x_min)
1305 || (chart->x_max != SYSMIS && value->f > chart->x_max)))
1311 /* Create a gsl_histogram from a freq_tab */
1312 static struct histogram *
1313 freq_tab_to_hist (const struct frq_proc *frq, const struct freq_tab *ft,
1314 const struct variable *var)
1316 double x_min, x_max, valid_freq;
1319 struct histogram *histogram;
1322 /* Find out the extremes of the x value, within the range to be included in
1323 the histogram, and sum the total frequency of those values. */
1327 for (i = 0; i < ft->n_valid; i++)
1329 const struct freq *f = &ft->valid[i];
1330 if (chart_includes_value (frq->hist, var, f->values))
1332 x_min = MIN (x_min, f->values[0].f);
1333 x_max = MAX (x_max, f->values[0].f);
1334 valid_freq += f->count;
1338 if (valid_freq <= 0)
1341 iqr = calculate_iqr (frq);
1344 /* Freedman-Diaconis' choice of bin width. */
1345 bin_width = 2 * iqr / pow (valid_freq, 1.0 / 3.0);
1349 bin_width = (x_max - x_min) / (1 + log2 (valid_freq));
1351 histogram = histogram_create (bin_width, x_min, x_max);
1353 if (histogram == NULL)
1356 for (i = 0; i < ft->n_valid; i++)
1358 const struct freq *f = &ft->valid[i];
1359 if (chart_includes_value (frq->hist, var, f->values))
1360 histogram_add (histogram, f->values[0].f, f->count);
1367 /* Allocate an array of struct freqs and fill them from the data in FRQ_TAB,
1368 according to the parameters of CATCHART
1369 N_SLICES will contain the number of slices allocated.
1370 The caller is responsible for freeing slices
1372 static struct freq *
1373 pick_cat_counts (const struct frq_chart *catchart,
1374 const struct freq_tab *frq_tab,
1379 struct freq *slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1381 for (i = 0; i < frq_tab->n_valid; i++)
1383 const struct freq *f = &frq_tab->valid[i];
1384 if (f->count > catchart->x_max)
1387 if (f->count < catchart->x_min)
1390 slices[n_slices] = *f;
1395 if (catchart->include_missing)
1397 for (i = 0; i < frq_tab->n_missing; i++)
1399 const struct freq *f = &frq_tab->missing[i];
1400 slices[n_slices].count += f->count;
1403 slices[n_slices].values[0] = f->values[0];
1406 if (frq_tab->n_missing > 0)
1410 *n_slicesp = n_slices;
1415 /* Allocate an array of struct freqs and fill them from the data in FRQ_TAB,
1416 according to the parameters of CATCHART
1417 N_SLICES will contain the number of slices allocated.
1418 The caller is responsible for freeing slices
1420 static struct freq **
1421 pick_cat_counts_ptr (const struct frq_chart *catchart,
1422 const struct freq_tab *frq_tab,
1427 struct freq **slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1429 for (i = 0; i < frq_tab->n_valid; i++)
1431 struct freq *f = &frq_tab->valid[i];
1432 if (f->count > catchart->x_max)
1435 if (f->count < catchart->x_min)
1438 slices[n_slices] = f;
1443 if (catchart->include_missing)
1445 for (i = 0; i < frq_tab->n_missing; i++)
1447 const struct freq *f = &frq_tab->missing[i];
1450 slices[n_slices] = xmalloc (sizeof (struct freq));
1451 slices[n_slices]->values[0] = f->values[0];
1454 slices[n_slices]->count += f->count;
1459 *n_slicesp = n_slices;
1466 do_piechart(const struct frq_chart *pie, const struct variable *var,
1467 const struct freq_tab *frq_tab)
1470 struct freq *slices = pick_cat_counts (pie, frq_tab, &n_slices);
1473 msg (SW, _("Omitting pie chart for %s, which has only %d unique values."),
1474 var_get_name (var), n_slices);
1475 else if (n_slices > 50)
1476 msg (SW, _("Omitting pie chart for %s, which has over 50 unique values."),
1477 var_get_name (var));
1479 chart_submit (piechart_create (var, slices, n_slices));
1486 do_barchart(const struct frq_chart *bar, const struct variable **var,
1487 const struct freq_tab *frq_tab)
1490 struct freq **slices = pick_cat_counts_ptr (bar, frq_tab, &n_slices);
1492 chart_submit (barchart_create (var, 1,
1493 (bar->y_scale == FRQ_FREQ) ? _("Count") : _("Percent"),
1494 (bar->y_scale == FRQ_PERCENT),
1500 /* Calculates all the pertinent statistics for VF, putting them in array
1503 calc_stats (const struct frq_proc *frq, const struct var_freqs *vf,
1504 double d[FRQ_ST_count])
1506 const struct freq_tab *ft = &vf->tab;
1507 double W = ft->valid_cases;
1508 const struct freq *f;
1510 int most_often = -1;
1511 double X_mode = SYSMIS;
1513 /* Calculate the mode. */
1514 for (f = ft->valid; f < ft->missing; f++)
1516 if (most_often < f->count)
1518 most_often = f->count;
1519 X_mode = f->values[0].f;
1521 else if (most_often == f->count)
1523 /* A duplicate mode is undefined.
1524 FIXME: keep track of *all* the modes. */
1529 /* Calculate moments. */
1530 m = moments_create (MOMENT_KURTOSIS);
1531 for (f = ft->valid; f < ft->missing; f++)
1532 moments_pass_one (m, f->values[0].f, f->count);
1533 for (f = ft->valid; f < ft->missing; f++)
1534 moments_pass_two (m, f->values[0].f, f->count);
1535 moments_calculate (m, NULL, &d[FRQ_ST_MEAN], &d[FRQ_ST_VARIANCE],
1536 &d[FRQ_ST_SKEWNESS], &d[FRQ_ST_KURTOSIS]);
1537 moments_destroy (m);
1539 /* Formulae below are taken from _SPSS Statistical Algorithms_. */
1540 if (ft->n_valid > 0)
1542 d[FRQ_ST_MINIMUM] = ft->valid[0].values[0].f;
1543 d[FRQ_ST_MAXIMUM] = ft->valid[ft->n_valid - 1].values[0].f;
1544 d[FRQ_ST_RANGE] = d[FRQ_ST_MAXIMUM] - d[FRQ_ST_MINIMUM];
1548 d[FRQ_ST_MINIMUM] = SYSMIS;
1549 d[FRQ_ST_MAXIMUM] = SYSMIS;
1550 d[FRQ_ST_RANGE] = SYSMIS;
1552 d[FRQ_ST_MODE] = X_mode;
1553 d[FRQ_ST_SUM] = d[FRQ_ST_MEAN] * W;
1554 d[FRQ_ST_STDDEV] = sqrt (d[FRQ_ST_VARIANCE]);
1555 d[FRQ_ST_SEMEAN] = d[FRQ_ST_STDDEV] / sqrt (W);
1556 d[FRQ_ST_SESKEWNESS] = calc_seskew (W);
1557 d[FRQ_ST_SEKURTOSIS] = calc_sekurt (W);
1558 d[FRQ_ST_MEDIAN] = frq->median ? frq->median->value : SYSMIS;
1562 all_string_variables (const struct frq_proc *frq)
1564 for (size_t i = 0; i < frq->n_vars; i++)
1565 if (var_is_numeric (frq->vars[i].var))
1571 /* Displays a table of all the statistics requested. */
1573 dump_statistics (const struct frq_proc *frq, const struct variable *wv)
1575 if (all_string_variables (frq))
1578 struct pivot_table *table = pivot_table_create (N_("Statistics"));
1579 pivot_table_set_weight_var (table, wv);
1581 struct pivot_dimension *variables
1582 = pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Variables"));
1584 struct pivot_dimension *statistics = pivot_dimension_create (
1585 table, PIVOT_AXIS_ROW, N_("Statistics"));
1586 struct pivot_category *n = pivot_category_create_group (
1587 statistics->root, N_("N"));
1588 pivot_category_create_leaves (n,
1589 N_("Valid"), PIVOT_RC_COUNT,
1590 N_("Missing"), PIVOT_RC_COUNT);
1591 for (int i = 0; i < FRQ_ST_count; i++)
1592 if (frq->stats & BIT_INDEX (i))
1593 pivot_category_create_leaf (statistics->root,
1594 pivot_value_new_text (st_name[i]));
1595 struct pivot_category *percentiles = NULL;
1596 for (size_t i = 0; i < frq->n_percentiles; i++)
1598 const struct percentile *pc = &frq->percentiles[i];
1604 percentiles = pivot_category_create_group (
1605 statistics->root, N_("Percentiles"));
1606 pivot_category_create_leaf (percentiles, pivot_value_new_integer (
1610 for (size_t i = 0; i < frq->n_vars; i++)
1612 struct var_freqs *vf = &frq->vars[i];
1613 if (var_is_alpha (vf->var))
1616 const struct freq_tab *ft = &vf->tab;
1618 int var_idx = pivot_category_create_leaf (
1619 variables->root, pivot_value_new_variable (vf->var));
1622 pivot_table_put2 (table, var_idx, row++,
1623 pivot_value_new_number (ft->valid_cases));
1624 pivot_table_put2 (table, var_idx, row++,
1625 pivot_value_new_number (
1626 ft->total_cases - ft->valid_cases));
1628 double stat_values[FRQ_ST_count];
1629 calc_stats (frq, vf, stat_values);
1630 for (int j = 0; j < FRQ_ST_count; j++)
1632 if (!(frq->stats & BIT_INDEX (j)))
1635 union value v = { .f = vf->tab.n_valid ? stat_values[j] : SYSMIS };
1636 struct pivot_value *pv
1637 = (j == FRQ_ST_MODE || j == FRQ_ST_MINIMUM || j == FRQ_ST_MAXIMUM
1638 ? pivot_value_new_var_value (vf->var, &v)
1639 : pivot_value_new_number (v.f));
1640 pivot_table_put2 (table, var_idx, row++, pv);
1643 for (size_t j = 0; j < frq->n_percentiles; j++)
1645 const struct percentile *pc = &frq->percentiles[j];
1649 union value v = { .f = vf->tab.n_valid ? pc->value : SYSMIS };
1650 pivot_table_put2 (table, var_idx, row++,
1651 pivot_value_new_var_value (vf->var, &v));
1655 pivot_table_submit (table);