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 double stat[FRQ_ST_count];
196 /* Variable attributes. */
204 struct var_freqs *vars;
207 /* Percentiles to calculate and possibly display. */
208 struct percentile *percentiles;
209 const struct percentile *median;
212 /* Frequency table display. */
213 long int max_categories; /* Maximum categories to show. */
214 int sort; /* FRQ_AVALUE or FRQ_DVALUE
215 or FRQ_AFREQ or FRQ_DFREQ. */
217 /* Statistics; number of statistics. */
221 /* Histogram and pie chart settings. */
222 struct frq_chart *hist, *pie, *bar;
228 struct freq_compare_aux
234 bool ascending_value;
237 static void calc_stats (const struct frq_proc *,
238 const struct var_freqs *, double d[FRQ_ST_count]);
240 static void do_piechart(const struct frq_chart *pie,
241 const struct variable *var,
242 const struct freq_tab *frq_tab);
244 static void do_barchart(const struct frq_chart *bar,
245 const struct variable **var,
246 const struct freq_tab *frq_tab);
248 static void dump_statistics (const struct frq_proc *frq,
249 const struct variable *wv);
252 compare_freq (const void *a_, const void *b_, const void *aux_)
254 const struct freq_compare_aux *aux = aux_;
255 const struct freq *a = a_;
256 const struct freq *b = b_;
258 if (aux->by_freq && a->count != b->count)
260 int cmp = a->count > b->count ? 1 : -1;
261 return aux->ascending_freq ? cmp : -cmp;
265 int cmp = value_compare_3way (a->values, b->values, aux->width);
266 return aux->ascending_value ? cmp : -cmp;
270 /* Create a gsl_histogram from a freq_tab */
271 static struct histogram *
272 freq_tab_to_hist (const struct frq_proc *frq, const struct freq_tab *ft,
273 const struct variable *var);
276 put_freq_row (struct pivot_table *table, int var_idx,
277 double frequency, double percent,
278 double valid_percent, double cum_percent)
280 double entries[] = { frequency, percent, valid_percent, cum_percent };
281 for (size_t i = 0; i < sizeof entries / sizeof *entries; i++)
282 if (entries[i] != SYSMIS)
283 pivot_table_put2 (table, i, var_idx,
284 pivot_value_new_number (entries[i]));
287 /* Displays a full frequency table for variable V. */
289 dump_freq_table (const struct var_freqs *vf, const struct variable *wv)
291 const struct freq_tab *ft = &vf->tab;
293 struct pivot_table *table = pivot_table_create__ (pivot_value_new_variable (
294 vf->var), "Frequencies");
295 pivot_table_set_weight_var (table, wv);
297 pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Statistics"),
298 N_("Frequency"), PIVOT_RC_COUNT,
299 N_("Percent"), PIVOT_RC_PERCENT,
300 N_("Valid Percent"), PIVOT_RC_PERCENT,
301 N_("Cumulative Percent"), PIVOT_RC_PERCENT);
303 struct pivot_dimension *variable = pivot_dimension_create__ (
304 table, PIVOT_AXIS_ROW, pivot_value_new_variable (vf->var));
306 double cum_freq = 0.0;
307 double cum_percent = 0.0;
308 struct pivot_category *valid = NULL;
309 for (const struct freq *f = ft->valid; f < ft->missing; f++)
311 cum_freq += f->count;
312 double valid_percent = f->count / ft->valid_cases * 100.0;
313 cum_percent += valid_percent;
316 valid = pivot_category_create_group (variable->root, N_("Valid"));
317 int var_idx = pivot_category_create_leaf (
318 valid, pivot_value_new_var_value (vf->var, &f->values[0]));
319 put_freq_row (table, var_idx, f->count,
320 f->count / ft->total_cases * 100.0,
321 valid_percent, cum_percent);
324 struct pivot_category *missing = NULL;
325 size_t n_categories = ft->n_valid + ft->n_missing;
326 for (const struct freq *f = ft->missing; f < &ft->valid[n_categories]; f++)
328 cum_freq += f->count;
331 missing = pivot_category_create_group (variable->root, N_("Missing"));
332 int var_idx = pivot_category_create_leaf (
333 missing, pivot_value_new_var_value (vf->var, &f->values[0]));
334 put_freq_row (table, var_idx, f->count,
335 f->count / ft->total_cases * 100.0, SYSMIS, SYSMIS);
338 int var_idx = pivot_category_create_leaf (
339 variable->root, pivot_value_new_text (N_("Total")));
340 put_freq_row (table, var_idx, cum_freq, cum_percent, SYSMIS, SYSMIS);
342 pivot_table_submit (table);
345 /* Statistical display. */
348 calc_percentile (double p, double valid_cases, double x1, double x2)
352 s = (settings_get_algorithm () != COMPATIBLE
353 ? modf ((valid_cases - 1) * p, &dummy)
354 : modf ((valid_cases + 1) * p - 1, &dummy));
356 return x1 + (x2 - x1) * s;
359 /* Calculates all of the percentiles for VF within FRQ. */
361 calc_percentiles (const struct frq_proc *frq, const struct var_freqs *vf)
363 const struct freq_tab *ft = &vf->tab;
364 double W = ft->valid_cases;
365 const struct freq *f;
366 int percentile_idx = 0;
369 for (f = ft->valid; f < ft->missing; f++)
372 for (; percentile_idx < frq->n_percentiles; percentile_idx++)
374 struct percentile *pc = &frq->percentiles[percentile_idx];
377 tp = (settings_get_algorithm () == ENHANCED
379 : (W + 1) * pc->p - 1);
384 if (tp + 1 < rank || f + 1 >= ft->missing)
385 pc->value = f->values[0].f;
387 pc->value = calc_percentile (pc->p, W, f->values[0].f, f[1].values[0].f);
390 for (; percentile_idx < frq->n_percentiles; percentile_idx++)
392 struct percentile *pc = &frq->percentiles[percentile_idx];
393 pc->value = (ft->n_valid > 0
394 ? ft->valid[ft->n_valid - 1].values[0].f
399 /* Returns true iff the value in struct freq F is non-missing
402 not_missing (const void *f_, const void *v_)
404 const struct freq *f = f_;
405 const struct variable *v = v_;
407 return !var_is_value_missing (v, f->values);
411 /* Summarizes the frequency table data for variable V. */
413 postprocess_freq_tab (const struct frq_proc *frq, struct var_freqs *vf)
415 struct freq_tab *ft = &vf->tab;
416 struct freq_compare_aux aux;
418 struct freq *freqs, *f;
421 /* Extract data from hash table. */
422 count = hmap_count (&ft->data);
423 freqs = freq_hmap_extract (&ft->data);
425 /* Put data into ft. */
427 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, vf->var);
428 ft->missing = freqs + ft->n_valid;
429 ft->n_missing = count - ft->n_valid;
432 aux.by_freq = frq->sort == FRQ_AFREQ || frq->sort == FRQ_DFREQ;
433 aux.ascending_freq = frq->sort != FRQ_DFREQ;
434 aux.width = vf->width;
435 aux.ascending_value = frq->sort != FRQ_DVALUE;
436 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare_freq, &aux);
437 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare_freq, &aux);
439 /* Summary statistics. */
440 ft->valid_cases = 0.0;
441 for(i = 0 ; i < ft->n_valid ; ++i)
444 ft->valid_cases += f->count;
448 ft->total_cases = ft->valid_cases ;
449 for(i = 0 ; i < ft->n_missing ; ++i)
452 ft->total_cases += f->count;
457 /* Frees the frequency table for variable V. */
459 cleanup_freq_tab (struct var_freqs *vf)
461 free (vf->tab.valid);
462 freq_hmap_destroy (&vf->tab.data, vf->width);
465 /* Add data from case C to the frequency table. */
467 calc (struct frq_proc *frq, const struct ccase *c, const struct dataset *ds)
469 double weight = dict_get_case_weight (dataset_dict (ds), c, &frq->warn);
472 for (i = 0; i < frq->n_vars; i++)
474 struct var_freqs *vf = &frq->vars[i];
475 const union value *value = case_data (c, vf->var);
476 size_t hash = value_hash (value, vf->width, 0);
479 f = freq_hmap_search (&vf->tab.data, value, vf->width, hash);
481 f = freq_hmap_insert (&vf->tab.data, value, vf->width, hash);
487 /* Prepares each variable that is the target of FREQUENCIES by setting
488 up its hash table. */
490 precalc (struct frq_proc *frq, struct casereader *input, struct dataset *ds)
495 c = casereader_peek (input, 0);
498 output_split_file_values (ds, c);
502 for (i = 0; i < frq->n_vars; i++)
503 hmap_init (&frq->vars[i].tab.data);
506 /* Finishes up with the variables after frequencies have been
507 calculated. Displays statistics, percentiles, ... */
509 postcalc (struct frq_proc *frq, const struct dataset *ds)
511 const struct dictionary *dict = dataset_dict (ds);
512 const struct variable *wv = dict_get_weight (dict);
515 for (i = 0; i < frq->n_vars; i++)
517 struct var_freqs *vf = &frq->vars[i];
518 postprocess_freq_tab (frq, vf);
519 calc_percentiles (frq, vf);
523 dump_statistics (frq, wv);
525 for (i = 0; i < frq->n_vars; i++)
527 struct var_freqs *vf = &frq->vars[i];
529 /* Frequencies tables. */
530 if (vf->tab.n_valid + vf->tab.n_missing <= frq->max_categories)
531 dump_freq_table (vf, wv);
534 if (frq->hist && var_is_numeric (vf->var) && vf->tab.n_valid > 0)
536 double d[FRQ_ST_count];
537 struct histogram *histogram;
539 calc_stats (frq, vf, d);
541 histogram = freq_tab_to_hist (frq, &vf->tab, vf->var);
545 chart_submit (histogram_chart_create (
546 histogram->gsl_hist, var_to_string(vf->var),
550 frq->hist->draw_normal));
552 statistic_destroy (&histogram->parent);
557 do_piechart(frq->pie, vf->var, &vf->tab);
560 do_barchart(frq->bar, &vf->var, &vf->tab);
562 cleanup_freq_tab (vf);
567 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
571 const struct variable **vars = NULL;
573 bool sbc_barchart = false;
574 bool sbc_piechart = false;
575 bool sbc_histogram = false;
577 double pie_min = -DBL_MAX;
578 double pie_max = DBL_MAX;
579 bool pie_missing = true;
581 double bar_min = -DBL_MAX;
582 double bar_max = DBL_MAX;
583 bool bar_freq = true;
585 double hi_min = -DBL_MAX;
586 double hi_max = DBL_MAX;
587 int hi_scale = FRQ_FREQ;
588 int hi_freq = INT_MIN;
589 int hi_pcnt = INT_MIN;
590 int hi_norm = FRQ_NONORMAL;
592 frq.pool = pool_create ();
593 frq.sort = FRQ_AVALUE;
598 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
599 | BIT_INDEX (FRQ_ST_STDDEV)
600 | BIT_INDEX (FRQ_ST_MINIMUM)
601 | BIT_INDEX (FRQ_ST_MAXIMUM);
605 frq.max_categories = LONG_MAX;
607 frq.percentiles = NULL;
608 frq.n_percentiles = 0;
616 /* Accept an optional, completely pointless "/VARIABLES=" */
617 lex_match (lexer, T_SLASH);
618 if (lex_match_id (lexer, "VARIABLES"))
620 if (! lex_force_match (lexer, T_EQUALS))
624 if (!parse_variables_const (lexer, dataset_dict (ds),
630 frq.vars = xcalloc (frq.n_vars, sizeof (*frq.vars));
631 for (i = 0; i < frq.n_vars; ++i)
633 frq.vars[i].var = vars[i];
634 frq.vars[i].width = var_get_width (vars[i]);
637 while (lex_token (lexer) != T_ENDCMD)
639 lex_match (lexer, T_SLASH);
641 if (lex_match_id (lexer, "STATISTICS"))
643 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
644 | BIT_INDEX (FRQ_ST_STDDEV)
645 | BIT_INDEX (FRQ_ST_MINIMUM)
646 | BIT_INDEX (FRQ_ST_MAXIMUM);
650 if (lex_match (lexer, T_EQUALS))
656 while (lex_token (lexer) != T_ENDCMD
657 && lex_token (lexer) != T_SLASH)
659 if (lex_match_id (lexer, "DEFAULT"))
661 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
662 | BIT_INDEX (FRQ_ST_STDDEV)
663 | BIT_INDEX (FRQ_ST_MINIMUM)
664 | BIT_INDEX (FRQ_ST_MAXIMUM);
668 else if (lex_match_id (lexer, "MEAN"))
670 frq.stats |= BIT_INDEX (FRQ_ST_MEAN);
673 else if (lex_match_id (lexer, "SEMEAN"))
675 frq.stats |= BIT_INDEX (FRQ_ST_SEMEAN);
678 else if (lex_match_id (lexer, "MEDIAN"))
680 frq.stats |= BIT_INDEX (FRQ_ST_MEDIAN);
683 else if (lex_match_id (lexer, "MODE"))
685 frq.stats |= BIT_INDEX (FRQ_ST_MODE);
688 else if (lex_match_id (lexer, "STDDEV"))
690 frq.stats |= BIT_INDEX (FRQ_ST_STDDEV);
693 else if (lex_match_id (lexer, "VARIANCE"))
695 frq.stats |= BIT_INDEX (FRQ_ST_VARIANCE);
698 else if (lex_match_id (lexer, "KURTOSIS"))
700 frq.stats |= BIT_INDEX (FRQ_ST_KURTOSIS);
703 else if (lex_match_id (lexer, "SKEWNESS"))
705 frq.stats |= BIT_INDEX (FRQ_ST_SKEWNESS);
708 else if (lex_match_id (lexer, "RANGE"))
710 frq.stats |= BIT_INDEX (FRQ_ST_RANGE);
713 else if (lex_match_id (lexer, "MINIMUM"))
715 frq.stats |= BIT_INDEX (FRQ_ST_MINIMUM);
718 else if (lex_match_id (lexer, "MAXIMUM"))
720 frq.stats |= BIT_INDEX (FRQ_ST_MAXIMUM);
723 else if (lex_match_id (lexer, "SUM"))
725 frq.stats |= BIT_INDEX (FRQ_ST_SUM);
728 else if (lex_match_id (lexer, "SESKEWNESS"))
730 frq.stats |= BIT_INDEX (FRQ_ST_SESKEWNESS);
733 else if (lex_match_id (lexer, "SEKURTOSIS"))
735 frq.stats |= BIT_INDEX (FRQ_ST_SEKURTOSIS);
738 else if (lex_match_id (lexer, "NONE"))
743 else if (lex_match (lexer, T_ALL))
746 frq.n_stats = FRQ_ST_count;
750 lex_error (lexer, NULL);
755 else if (lex_match_id (lexer, "PERCENTILES"))
757 lex_match (lexer, T_EQUALS);
758 while (lex_token (lexer) != T_ENDCMD
759 && lex_token (lexer) != T_SLASH)
761 if (lex_force_num (lexer))
764 xrealloc (frq.percentiles,
765 (frq.n_percentiles + 1)
766 * sizeof (*frq.percentiles));
767 frq.percentiles[frq.n_percentiles].p = lex_number (lexer) / 100.0;
768 frq.percentiles[frq.n_percentiles].show = true;
774 lex_error (lexer, NULL);
777 lex_match (lexer, T_COMMA);
780 else if (lex_match_id (lexer, "FORMAT"))
782 lex_match (lexer, T_EQUALS);
783 while (lex_token (lexer) != T_ENDCMD
784 && lex_token (lexer) != T_SLASH)
786 if (lex_match_id (lexer, "TABLE"))
789 else if (lex_match_id (lexer, "NOTABLE"))
791 frq.max_categories = 0;
793 else if (lex_match_id (lexer, "LIMIT"))
795 if (!lex_force_match (lexer, T_LPAREN)
796 || !lex_force_int_range (lexer, "LIMIT", 0, INT_MAX))
799 frq.max_categories = lex_integer (lexer);
802 if (!lex_force_match (lexer, T_RPAREN))
805 else if (lex_match_id (lexer, "AVALUE"))
807 frq.sort = FRQ_AVALUE;
809 else if (lex_match_id (lexer, "DVALUE"))
811 frq.sort = FRQ_DVALUE;
813 else if (lex_match_id (lexer, "AFREQ"))
815 frq.sort = FRQ_AFREQ;
817 else if (lex_match_id (lexer, "DFREQ"))
819 frq.sort = FRQ_DFREQ;
823 lex_error (lexer, NULL);
828 else if (lex_match_id (lexer, "NTILES"))
830 lex_match (lexer, T_EQUALS);
832 if (lex_force_int_range (lexer, "NTILES", 0, INT_MAX))
834 int n = lex_integer (lexer);
836 for (int i = 0; i < n + 1; ++i)
839 xrealloc (frq.percentiles,
840 (frq.n_percentiles + 1)
841 * sizeof (*frq.percentiles));
842 frq.percentiles[frq.n_percentiles].p =
844 frq.percentiles[frq.n_percentiles].show = true;
851 lex_error (lexer, NULL);
855 else if (lex_match_id (lexer, "ALGORITHM"))
857 lex_match (lexer, T_EQUALS);
859 if (lex_match_id (lexer, "COMPATIBLE"))
861 settings_set_cmd_algorithm (COMPATIBLE);
863 else if (lex_match_id (lexer, "ENHANCED"))
865 settings_set_cmd_algorithm (ENHANCED);
869 lex_error (lexer, NULL);
873 else if (lex_match_id (lexer, "HISTOGRAM"))
875 lex_match (lexer, T_EQUALS);
876 sbc_histogram = true;
878 while (lex_token (lexer) != T_ENDCMD
879 && lex_token (lexer) != T_SLASH)
881 if (lex_match_id (lexer, "NORMAL"))
883 hi_norm = FRQ_NORMAL;
885 else if (lex_match_id (lexer, "NONORMAL"))
887 hi_norm = FRQ_NONORMAL;
889 else if (lex_match_id (lexer, "FREQ"))
892 if (lex_match (lexer, T_LPAREN))
894 if (lex_force_int_range (lexer, "FREQ", 1, INT_MAX))
896 hi_freq = lex_integer (lexer);
898 if (! lex_force_match (lexer, T_RPAREN))
903 else if (lex_match_id (lexer, "PERCENT"))
905 hi_scale = FRQ_PERCENT;
906 if (lex_match (lexer, T_LPAREN))
908 if (lex_force_int_range (lexer, "PERCENT", 1, INT_MAX))
910 hi_pcnt = lex_integer (lexer);
912 if (! lex_force_match (lexer, T_RPAREN))
917 else if (lex_match_id (lexer, "MINIMUM"))
919 if (! lex_force_match (lexer, T_LPAREN))
921 if (lex_force_num (lexer))
923 hi_min = lex_number (lexer);
926 if (! lex_force_match (lexer, T_RPAREN))
929 else if (lex_match_id (lexer, "MAXIMUM"))
931 if (! lex_force_match (lexer, T_LPAREN))
933 if (lex_force_num (lexer))
935 hi_max = lex_number (lexer);
938 if (! lex_force_match (lexer, T_RPAREN))
943 lex_error (lexer, NULL);
948 else if (lex_match_id (lexer, "PIECHART"))
950 lex_match (lexer, T_EQUALS);
951 while (lex_token (lexer) != T_ENDCMD
952 && lex_token (lexer) != T_SLASH)
954 if (lex_match_id (lexer, "MINIMUM"))
956 if (! lex_force_match (lexer, T_LPAREN))
958 if (lex_force_num (lexer))
960 pie_min = lex_number (lexer);
963 if (! lex_force_match (lexer, T_RPAREN))
966 else if (lex_match_id (lexer, "MAXIMUM"))
968 if (! lex_force_match (lexer, T_LPAREN))
970 if (lex_force_num (lexer))
972 pie_max = lex_number (lexer);
975 if (! lex_force_match (lexer, T_RPAREN))
978 else if (lex_match_id (lexer, "MISSING"))
982 else if (lex_match_id (lexer, "NOMISSING"))
988 lex_error (lexer, NULL);
994 else if (lex_match_id (lexer, "BARCHART"))
996 lex_match (lexer, T_EQUALS);
997 while (lex_token (lexer) != T_ENDCMD
998 && lex_token (lexer) != T_SLASH)
1000 if (lex_match_id (lexer, "MINIMUM"))
1002 if (! lex_force_match (lexer, T_LPAREN))
1004 if (lex_force_num (lexer))
1006 bar_min = lex_number (lexer);
1009 if (! lex_force_match (lexer, T_RPAREN))
1012 else if (lex_match_id (lexer, "MAXIMUM"))
1014 if (! lex_force_match (lexer, T_LPAREN))
1016 if (lex_force_num (lexer))
1018 bar_max = lex_number (lexer);
1021 if (! lex_force_match (lexer, T_RPAREN))
1024 else if (lex_match_id (lexer, "FREQ"))
1026 if (lex_match (lexer, T_LPAREN))
1028 if (lex_force_num (lexer))
1033 if (! lex_force_match (lexer, T_RPAREN))
1038 else if (lex_match_id (lexer, "PERCENT"))
1040 if (lex_match (lexer, T_LPAREN))
1042 if (lex_force_num (lexer))
1047 if (! lex_force_match (lexer, T_RPAREN))
1054 lex_error (lexer, NULL);
1058 sbc_barchart = true;
1060 else if (lex_match_id (lexer, "MISSING"))
1062 lex_match (lexer, T_EQUALS);
1064 while (lex_token (lexer) != T_ENDCMD
1065 && lex_token (lexer) != T_SLASH)
1067 if (lex_match_id (lexer, "EXCLUDE"))
1070 else if (lex_match_id (lexer, "INCLUDE"))
1075 lex_error (lexer, NULL);
1080 else if (lex_match_id (lexer, "ORDER"))
1082 lex_match (lexer, T_EQUALS);
1083 if (!lex_match_id (lexer, "ANALYSIS"))
1084 lex_match_id (lexer, "VARIABLE");
1088 lex_error (lexer, NULL);
1093 if (frq.stats & BIT_INDEX (FRQ_ST_MEDIAN))
1096 xrealloc (frq.percentiles,
1097 (frq.n_percentiles + 1)
1098 * sizeof (*frq.percentiles));
1100 frq.percentiles[frq.n_percentiles].p = 0.50;
1101 frq.percentiles[frq.n_percentiles].show = false;
1103 frq.n_percentiles++;
1107 /* Figure out which charts the user requested. */
1112 struct frq_chart *hist;
1114 hist = frq.hist = xmalloc (sizeof *frq.hist);
1115 hist->x_min = hi_min;
1116 hist->x_max = hi_max;
1117 hist->y_scale = hi_scale;
1118 hist->y_max = hi_scale == FRQ_FREQ ? hi_freq : hi_pcnt;
1119 hist->draw_normal = hi_norm != FRQ_NONORMAL;
1120 hist->include_missing = false;
1122 if (hist->x_min != SYSMIS && hist->x_max != SYSMIS
1123 && hist->x_min >= hist->x_max)
1125 msg (SE, _("%s for histogram must be greater than or equal to %s, "
1126 "but %s was specified as %.15g and %s as %.15g. "
1127 "%s and %s will be ignored."),
1132 hist->x_min = hist->x_max = SYSMIS;
1136 xrealloc (frq.percentiles,
1137 (frq.n_percentiles + 2)
1138 * sizeof (*frq.percentiles));
1140 frq.percentiles[frq.n_percentiles].p = 0.25;
1141 frq.percentiles[frq.n_percentiles].show = false;
1143 frq.percentiles[frq.n_percentiles + 1].p = 0.75;
1144 frq.percentiles[frq.n_percentiles + 1].show = false;
1146 frq.n_percentiles+=2;
1151 frq.bar = xmalloc (sizeof *frq.bar);
1152 frq.bar->x_min = bar_min;
1153 frq.bar->x_max = bar_max;
1154 frq.bar->include_missing = false;
1155 frq.bar->y_scale = bar_freq ? FRQ_FREQ : FRQ_PERCENT;
1160 struct frq_chart *pie;
1162 pie = frq.pie = xmalloc (sizeof *frq.pie);
1163 pie->x_min = pie_min;
1164 pie->x_max = pie_max;
1165 pie->include_missing = pie_missing;
1167 if (pie->x_min != SYSMIS && pie->x_max != SYSMIS
1168 && pie->x_min >= pie->x_max)
1170 msg (SE, _("%s for pie chart must be greater than or equal to %s, "
1171 "but %s was specified as %.15g and %s as %.15g. "
1172 "%s and %s will be ignored."),
1177 pie->x_min = pie->x_max = SYSMIS;
1184 double previous_p = -1;
1185 qsort (frq.percentiles, frq.n_percentiles,
1186 sizeof (*frq.percentiles),
1189 for (i = o = 0; i < frq.n_percentiles; ++i)
1191 if (frq.percentiles[i].p != previous_p)
1193 frq.percentiles[o].p = frq.percentiles[i].p;
1194 frq.percentiles[o].show = frq.percentiles[i].show;
1197 else if (frq.percentiles[i].show &&
1198 !frq.percentiles[o].show)
1200 frq.percentiles[o].show = true;
1202 previous_p = frq.percentiles[i].p;
1205 frq.n_percentiles = o;
1208 for (i = 0; i < frq.n_percentiles; i++)
1209 if (frq.percentiles[i].p == 0.5)
1211 frq.median = &frq.percentiles[i];
1217 struct casegrouper *grouper;
1218 struct casereader *group;
1221 grouper = casegrouper_create_splits (proc_open (ds), dataset_dict (ds));
1222 while (casegrouper_get_next_group (grouper, &group))
1225 precalc (&frq, group, ds);
1227 for (; (c = casereader_read (group)) != NULL; case_unref (c))
1229 postcalc (&frq, ds);
1230 casereader_destroy (group);
1232 ok = casegrouper_destroy (grouper);
1233 ok = proc_commit (ds) && ok;
1242 free (frq.percentiles);
1243 pool_destroy (frq.pool);
1254 free (frq.percentiles);
1255 pool_destroy (frq.pool);
1261 calculate_iqr (const struct frq_proc *frq)
1267 /* This cannot work unless the 25th and 75th percentile are calculated */
1268 assert (frq->n_percentiles >= 2);
1269 for (i = 0; i < frq->n_percentiles; i++)
1271 struct percentile *pc = &frq->percentiles[i];
1273 if (fabs (0.25 - pc->p) < DBL_EPSILON)
1275 else if (fabs (0.75 - pc->p) < DBL_EPSILON)
1279 return q1 == SYSMIS || q3 == SYSMIS ? SYSMIS : q3 - q1;
1283 chart_includes_value (const struct frq_chart *chart,
1284 const struct variable *var,
1285 const union value *value)
1287 if (!chart->include_missing && var_is_value_missing (var, value))
1290 if (var_is_numeric (var)
1291 && ((chart->x_min != SYSMIS && value->f < chart->x_min)
1292 || (chart->x_max != SYSMIS && value->f > chart->x_max)))
1298 /* Create a gsl_histogram from a freq_tab */
1299 static struct histogram *
1300 freq_tab_to_hist (const struct frq_proc *frq, const struct freq_tab *ft,
1301 const struct variable *var)
1303 double x_min, x_max, valid_freq;
1306 struct histogram *histogram;
1309 /* Find out the extremes of the x value, within the range to be included in
1310 the histogram, and sum the total frequency of those values. */
1314 for (i = 0; i < ft->n_valid; i++)
1316 const struct freq *f = &ft->valid[i];
1317 if (chart_includes_value (frq->hist, var, f->values))
1319 x_min = MIN (x_min, f->values[0].f);
1320 x_max = MAX (x_max, f->values[0].f);
1321 valid_freq += f->count;
1325 if (valid_freq <= 0)
1328 iqr = calculate_iqr (frq);
1331 /* Freedman-Diaconis' choice of bin width. */
1332 bin_width = 2 * iqr / pow (valid_freq, 1.0 / 3.0);
1336 bin_width = (x_max - x_min) / (1 + log2 (valid_freq));
1338 histogram = histogram_create (bin_width, x_min, x_max);
1340 if (histogram == NULL)
1343 for (i = 0; i < ft->n_valid; i++)
1345 const struct freq *f = &ft->valid[i];
1346 if (chart_includes_value (frq->hist, var, f->values))
1347 histogram_add (histogram, f->values[0].f, f->count);
1354 /* Allocate an array of struct freqs and fill them from the data in FRQ_TAB,
1355 according to the parameters of CATCHART
1356 N_SLICES will contain the number of slices allocated.
1357 The caller is responsible for freeing slices
1359 static struct freq *
1360 pick_cat_counts (const struct frq_chart *catchart,
1361 const struct freq_tab *frq_tab,
1366 struct freq *slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1368 for (i = 0; i < frq_tab->n_valid; i++)
1370 const struct freq *f = &frq_tab->valid[i];
1371 if (f->count > catchart->x_max)
1374 if (f->count < catchart->x_min)
1377 slices[n_slices] = *f;
1382 if (catchart->include_missing)
1384 for (i = 0; i < frq_tab->n_missing; i++)
1386 const struct freq *f = &frq_tab->missing[i];
1387 slices[n_slices].count += f->count;
1390 slices[n_slices].values[0] = f->values[0];
1393 if (frq_tab->n_missing > 0)
1397 *n_slicesp = n_slices;
1402 /* Allocate an array of struct freqs and fill them from the data in FRQ_TAB,
1403 according to the parameters of CATCHART
1404 N_SLICES will contain the number of slices allocated.
1405 The caller is responsible for freeing slices
1407 static struct freq **
1408 pick_cat_counts_ptr (const struct frq_chart *catchart,
1409 const struct freq_tab *frq_tab,
1414 struct freq **slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1416 for (i = 0; i < frq_tab->n_valid; i++)
1418 struct freq *f = &frq_tab->valid[i];
1419 if (f->count > catchart->x_max)
1422 if (f->count < catchart->x_min)
1425 slices[n_slices] = f;
1430 if (catchart->include_missing)
1432 for (i = 0; i < frq_tab->n_missing; i++)
1434 const struct freq *f = &frq_tab->missing[i];
1437 slices[n_slices] = xmalloc (sizeof (struct freq));
1438 slices[n_slices]->values[0] = f->values[0];
1441 slices[n_slices]->count += f->count;
1446 *n_slicesp = n_slices;
1453 do_piechart(const struct frq_chart *pie, const struct variable *var,
1454 const struct freq_tab *frq_tab)
1457 struct freq *slices = pick_cat_counts (pie, frq_tab, &n_slices);
1460 msg (SW, _("Omitting pie chart for %s, which has only %d unique values."),
1461 var_get_name (var), n_slices);
1462 else if (n_slices > 50)
1463 msg (SW, _("Omitting pie chart for %s, which has over 50 unique values."),
1464 var_get_name (var));
1466 chart_submit (piechart_create (var, slices, n_slices));
1473 do_barchart(const struct frq_chart *bar, const struct variable **var,
1474 const struct freq_tab *frq_tab)
1477 struct freq **slices = pick_cat_counts_ptr (bar, frq_tab, &n_slices);
1480 msg (SW, _("Omitting bar chart, which has no values."));
1482 chart_submit (barchart_create (var, 1,
1483 (bar->y_scale == FRQ_FREQ) ? _("Count") : _("Percent"),
1484 (bar->y_scale == FRQ_PERCENT),
1490 /* Calculates all the pertinent statistics for VF, putting them in array
1493 calc_stats (const struct frq_proc *frq, const struct var_freqs *vf,
1494 double d[FRQ_ST_count])
1496 const struct freq_tab *ft = &vf->tab;
1497 double W = ft->valid_cases;
1498 const struct freq *f;
1500 int most_often = -1;
1501 double X_mode = SYSMIS;
1503 /* Calculate the mode. */
1504 for (f = ft->valid; f < ft->missing; f++)
1506 if (most_often < f->count)
1508 most_often = f->count;
1509 X_mode = f->values[0].f;
1511 else if (most_often == f->count)
1513 /* A duplicate mode is undefined.
1514 FIXME: keep track of *all* the modes. */
1519 /* Calculate moments. */
1520 m = moments_create (MOMENT_KURTOSIS);
1521 for (f = ft->valid; f < ft->missing; f++)
1522 moments_pass_one (m, f->values[0].f, f->count);
1523 for (f = ft->valid; f < ft->missing; f++)
1524 moments_pass_two (m, f->values[0].f, f->count);
1525 moments_calculate (m, NULL, &d[FRQ_ST_MEAN], &d[FRQ_ST_VARIANCE],
1526 &d[FRQ_ST_SKEWNESS], &d[FRQ_ST_KURTOSIS]);
1527 moments_destroy (m);
1529 /* Formulae below are taken from _SPSS Statistical Algorithms_. */
1530 if (ft->n_valid > 0)
1532 d[FRQ_ST_MINIMUM] = ft->valid[0].values[0].f;
1533 d[FRQ_ST_MAXIMUM] = ft->valid[ft->n_valid - 1].values[0].f;
1534 d[FRQ_ST_RANGE] = d[FRQ_ST_MAXIMUM] - d[FRQ_ST_MINIMUM];
1538 d[FRQ_ST_MINIMUM] = SYSMIS;
1539 d[FRQ_ST_MAXIMUM] = SYSMIS;
1540 d[FRQ_ST_RANGE] = SYSMIS;
1542 d[FRQ_ST_MODE] = X_mode;
1543 d[FRQ_ST_SUM] = d[FRQ_ST_MEAN] * W;
1544 d[FRQ_ST_STDDEV] = sqrt (d[FRQ_ST_VARIANCE]);
1545 d[FRQ_ST_SEMEAN] = d[FRQ_ST_STDDEV] / sqrt (W);
1546 d[FRQ_ST_SESKEWNESS] = calc_seskew (W);
1547 d[FRQ_ST_SEKURTOSIS] = calc_sekurt (W);
1548 d[FRQ_ST_MEDIAN] = frq->median ? frq->median->value : SYSMIS;
1552 all_string_variables (const struct frq_proc *frq)
1554 for (size_t i = 0; i < frq->n_vars; i++)
1555 if (var_is_numeric (frq->vars[i].var))
1561 /* Displays a table of all the statistics requested. */
1563 dump_statistics (const struct frq_proc *frq, const struct variable *wv)
1565 if (all_string_variables (frq))
1568 struct pivot_table *table = pivot_table_create (N_("Statistics"));
1569 pivot_table_set_weight_var (table, wv);
1571 struct pivot_dimension *variables
1572 = pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Variables"));
1574 struct pivot_dimension *statistics = pivot_dimension_create (
1575 table, PIVOT_AXIS_ROW, N_("Statistics"));
1576 struct pivot_category *n = pivot_category_create_group (
1577 statistics->root, N_("N"));
1578 pivot_category_create_leaves (n,
1579 N_("Valid"), PIVOT_RC_COUNT,
1580 N_("Missing"), PIVOT_RC_COUNT);
1581 for (int i = 0; i < FRQ_ST_count; i++)
1582 if (frq->stats & BIT_INDEX (i))
1583 pivot_category_create_leaf (statistics->root,
1584 pivot_value_new_text (st_name[i]));
1585 struct pivot_category *percentiles = NULL;
1586 for (size_t i = 0; i < frq->n_percentiles; i++)
1588 const struct percentile *pc = &frq->percentiles[i];
1594 percentiles = pivot_category_create_group (
1595 statistics->root, N_("Percentiles"));
1596 pivot_category_create_leaf (percentiles, pivot_value_new_integer (
1600 for (size_t i = 0; i < frq->n_vars; i++)
1602 struct var_freqs *vf = &frq->vars[i];
1603 if (var_is_alpha (vf->var))
1606 const struct freq_tab *ft = &vf->tab;
1608 int var_idx = pivot_category_create_leaf (
1609 variables->root, pivot_value_new_variable (vf->var));
1612 pivot_table_put2 (table, var_idx, row++,
1613 pivot_value_new_number (ft->valid_cases));
1614 pivot_table_put2 (table, var_idx, row++,
1615 pivot_value_new_number (
1616 ft->total_cases - ft->valid_cases));
1618 double stat_values[FRQ_ST_count];
1619 calc_stats (frq, vf, stat_values);
1620 for (int j = 0; j < FRQ_ST_count; j++)
1622 if (!(frq->stats & BIT_INDEX (j)))
1625 union value v = { .f = vf->tab.n_valid ? stat_values[j] : SYSMIS };
1626 struct pivot_value *pv
1627 = (j == FRQ_ST_MODE || j == FRQ_ST_MINIMUM || j == FRQ_ST_MAXIMUM
1628 ? pivot_value_new_var_value (vf->var, &v)
1629 : pivot_value_new_number (v.f));
1630 pivot_table_put2 (table, var_idx, row++, pv);
1633 for (size_t j = 0; j < frq->n_percentiles; j++)
1635 const struct percentile *pc = &frq->percentiles[j];
1639 union value v = { .f = vf->tab.n_valid ? pc->value : SYSMIS };
1640 pivot_table_put2 (table, var_idx, row++,
1641 pivot_value_new_var_value (vf->var, &v));
1645 pivot_table_submit (table);