2 PSPP - a program for statistical analysis.
3 Copyright (C) 1997-9, 2000, 2007, 2009, 2010, 2011, 2014 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/piechart.h"
57 #include "output/charts/plot-hist.h"
58 #include "output/tab.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 int n_percentiles, n_show_percentiles;
215 /* Frequency table display. */
216 int max_categories; /* Maximum categories to show. */
217 int sort; /* FRQ_AVALUE or FRQ_DVALUE
218 or FRQ_AFREQ or FRQ_DFREQ. */
220 /* Statistics; number of statistics. */
224 /* Histogram and pie chart settings. */
225 struct frq_chart *hist, *pie;
229 struct freq_compare_aux
235 bool ascending_value;
238 static void calc_stats (const struct var_freqs *vf, 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 dump_statistics (const struct frq_proc *frq,
245 const struct var_freqs *vf,
246 const struct variable *wv);
249 compare_freq (const void *a_, const void *b_, const void *aux_)
251 const struct freq_compare_aux *aux = aux_;
252 const struct freq *a = a_;
253 const struct freq *b = b_;
255 if (aux->by_freq && a->count != b->count)
257 int cmp = a->count > b->count ? 1 : -1;
258 return aux->ascending_freq ? cmp : -cmp;
262 int cmp = value_compare_3way (&a->value, &b->value, aux->width);
263 return aux->ascending_value ? cmp : -cmp;
267 /* Create a gsl_histogram from a freq_tab */
268 static struct histogram *
269 freq_tab_to_hist (const struct frq_proc *frq, const struct freq_tab *ft,
270 const struct variable *var);
273 /* Displays a full frequency table for variable V. */
275 dump_freq_table (const struct var_freqs *vf, const struct variable *wv)
277 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
278 const struct freq_tab *ft = &vf->tab;
283 double cum_total = 0.0;
284 double cum_freq = 0.0;
286 static const char *headings[] = {
295 n_categories = ft->n_valid + ft->n_missing;
296 t = tab_create (6, n_categories + 2);
297 tab_set_format (t, RC_WEIGHT, wfmt);
298 tab_headers (t, 0, 0, 1, 0);
300 for (x = 0; x < 6; x++)
301 tab_text (t, x, 0, TAB_CENTER | TAT_TITLE, gettext (headings[x]));
304 for (f = ft->valid; f < ft->missing; f++)
307 double percent, valid_percent;
309 cum_freq += f->count;
311 percent = f->count / ft->total_cases * 100.0;
312 valid_percent = f->count / ft->valid_cases * 100.0;
313 cum_total += valid_percent;
315 label = var_lookup_value_label (vf->var, &f->value);
317 tab_text (t, 0, r, TAB_LEFT, label);
319 tab_value (t, 1, r, TAB_NONE, &f->value, vf->var, NULL);
320 tab_double (t, 2, r, TAB_NONE, f->count, NULL, RC_WEIGHT);
321 tab_double (t, 3, r, TAB_NONE, percent, NULL, RC_OTHER);
322 tab_double (t, 4, r, TAB_NONE, valid_percent, NULL, RC_OTHER);
323 tab_double (t, 5, r, TAB_NONE, cum_total, NULL, RC_OTHER);
326 for (; f < &ft->valid[n_categories]; f++)
330 cum_freq += f->count;
332 label = var_lookup_value_label (vf->var, &f->value);
334 tab_text (t, 0, r, TAB_LEFT, label);
336 tab_value (t, 1, r, TAB_NONE, &f->value, vf->var, NULL);
337 tab_double (t, 2, r, TAB_NONE, f->count, NULL, RC_WEIGHT);
338 tab_double (t, 3, r, TAB_NONE,
339 f->count / ft->total_cases * 100.0, NULL, RC_OTHER);
340 tab_text (t, 4, r, TAB_NONE, _("Missing"));
344 tab_box (t, TAL_1, TAL_1, -1, TAL_1, 0, 0, 5, r);
345 tab_hline (t, TAL_2, 0, 5, 1);
346 tab_hline (t, TAL_2, 0, 5, r);
347 tab_joint_text (t, 0, r, 1, r, TAB_RIGHT | TAT_TITLE, _("Total"));
348 tab_vline (t, TAL_0, 1, r, r);
349 tab_double (t, 2, r, TAB_NONE, cum_freq, NULL, RC_WEIGHT);
350 tab_double (t, 3, r, TAB_NONE, 100.0, &F_5_1, RC_OTHER);
351 tab_double (t, 4, r, TAB_NONE, 100.0, &F_5_1, RC_OTHER);
353 tab_title (t, "%s", var_to_string (vf->var));
357 /* Statistical display. */
360 calc_percentile (double p, double valid_cases, double x1, double x2)
364 s = (settings_get_algorithm () != COMPATIBLE
365 ? modf ((valid_cases - 1) * p, &dummy)
366 : modf ((valid_cases + 1) * p - 1, &dummy));
368 return x1 + (x2 - x1) * s;
371 /* Calculates all of the percentiles for VF within FRQ. */
373 calc_percentiles (const struct frq_proc *frq, const struct var_freqs *vf)
375 const struct freq_tab *ft = &vf->tab;
376 double W = ft->valid_cases;
377 const struct freq *f;
378 int percentile_idx = 0;
381 for (f = ft->valid; f < ft->missing; f++)
384 for (; percentile_idx < frq->n_percentiles; percentile_idx++)
386 struct percentile *pc = &frq->percentiles[percentile_idx];
389 tp = (settings_get_algorithm () == ENHANCED
391 : (W + 1) * pc->p - 1);
396 if (tp + 1 < rank || f + 1 >= ft->missing)
397 pc->value = f->value.f;
399 pc->value = calc_percentile (pc->p, W, f->value.f, f[1].value.f);
402 for (; percentile_idx < frq->n_percentiles; percentile_idx++)
404 struct percentile *pc = &frq->percentiles[percentile_idx];
405 pc->value = ft->valid[ft->n_valid - 1].value.f;
409 /* Returns true iff the value in struct freq F is non-missing
412 not_missing (const void *f_, const void *v_)
414 const struct freq *f = f_;
415 const struct variable *v = v_;
417 return !var_is_value_missing (v, &f->value, MV_ANY);
421 /* Summarizes the frequency table data for variable V. */
423 postprocess_freq_tab (const struct frq_proc *frq, struct var_freqs *vf)
425 struct freq_tab *ft = &vf->tab;
426 struct freq_compare_aux aux;
428 struct freq *freqs, *f;
431 /* Extract data from hash table. */
432 count = hmap_count (&ft->data);
433 freqs = freq_hmap_extract (&ft->data);
435 /* Put data into ft. */
437 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, vf->var);
438 ft->missing = freqs + ft->n_valid;
439 ft->n_missing = count - ft->n_valid;
442 aux.by_freq = frq->sort == FRQ_AFREQ || frq->sort == FRQ_DFREQ;
443 aux.ascending_freq = frq->sort != FRQ_DFREQ;
444 aux.width = vf->width;
445 aux.ascending_value = frq->sort != FRQ_DVALUE;
446 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare_freq, &aux);
447 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare_freq, &aux);
449 /* Summary statistics. */
450 ft->valid_cases = 0.0;
451 for(i = 0 ; i < ft->n_valid ; ++i )
454 ft->valid_cases += f->count;
458 ft->total_cases = ft->valid_cases ;
459 for(i = 0 ; i < ft->n_missing ; ++i )
462 ft->total_cases += f->count;
467 /* Frees the frequency table for variable V. */
469 cleanup_freq_tab (struct var_freqs *vf)
471 free (vf->tab.valid);
472 freq_hmap_destroy (&vf->tab.data, vf->width);
475 /* Add data from case C to the frequency table. */
477 calc (struct frq_proc *frq, const struct ccase *c, const struct dataset *ds)
479 double weight = dict_get_case_weight (dataset_dict (ds), c, NULL);
482 for (i = 0; i < frq->n_vars; i++)
484 struct var_freqs *vf = &frq->vars[i];
485 const union value *value = case_data (c, vf->var);
486 size_t hash = value_hash (value, vf->width, 0);
489 f = freq_hmap_search (&vf->tab.data, value, vf->width, hash);
491 f = freq_hmap_insert (&vf->tab.data, value, vf->width, hash);
497 /* Prepares each variable that is the target of FREQUENCIES by setting
498 up its hash table. */
500 precalc (struct frq_proc *frq, struct casereader *input, struct dataset *ds)
505 c = casereader_peek (input, 0);
508 output_split_file_values (ds, c);
512 for (i = 0; i < frq->n_vars; i++)
513 hmap_init (&frq->vars[i].tab.data);
516 /* Finishes up with the variables after frequencies have been
517 calculated. Displays statistics, percentiles, ... */
519 postcalc (struct frq_proc *frq, const struct dataset *ds)
521 const struct dictionary *dict = dataset_dict (ds);
522 const struct variable *wv = dict_get_weight (dict);
525 for (i = 0; i < frq->n_vars; i++)
527 struct var_freqs *vf = &frq->vars[i];
529 postprocess_freq_tab (frq, vf);
531 /* Frequencies tables. */
532 if (vf->tab.n_valid + vf->tab.n_missing <= frq->max_categories)
533 dump_freq_table (vf, wv);
535 calc_percentiles (frq, vf);
539 dump_statistics (frq, vf, wv);
541 if (frq->hist && var_is_numeric (vf->var) && vf->tab.n_valid > 0)
543 double d[FRQ_ST_count];
544 struct histogram *histogram;
548 histogram = freq_tab_to_hist (frq, &vf->tab, vf->var);
552 chart_item_submit (histogram_chart_create (
553 histogram->gsl_hist, var_to_string(vf->var),
557 frq->hist->draw_normal));
559 statistic_destroy (&histogram->parent);
564 do_piechart(frq->pie, vf->var, &vf->tab);
566 cleanup_freq_tab (vf);
571 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
575 const struct variable **vars;
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 = false;
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 = INT_MAX;
607 frq.percentiles = NULL;
608 frq.n_percentiles = 0;
609 frq.n_show_percentiles = 0;
615 /* Accept an optional, completely pointless "/VARIABLES=" */
616 lex_match (lexer, T_SLASH);
617 if (lex_match_id (lexer, "VARIABLES"))
619 if (! lex_force_match (lexer, T_EQUALS) )
623 if (!parse_variables_const (lexer, dataset_dict (ds),
629 frq.vars = xzalloc (frq.n_vars * sizeof (*frq.vars));
630 for (i = 0; i < frq.n_vars; ++i)
632 frq.vars[i].var = vars[i];
633 frq.vars[i].width = var_get_width (vars[i]);
636 while (lex_token (lexer) != T_ENDCMD)
638 lex_match (lexer, T_SLASH);
640 if (lex_match_id (lexer, "STATISTICS"))
642 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
643 | BIT_INDEX (FRQ_ST_STDDEV)
644 | BIT_INDEX (FRQ_ST_MINIMUM)
645 | BIT_INDEX (FRQ_ST_MAXIMUM);
649 if (lex_match (lexer, T_EQUALS))
655 while (lex_token (lexer) != T_ENDCMD
656 && lex_token (lexer) != T_SLASH)
658 if (lex_match_id (lexer, "DEFAULT"))
660 frq.stats = BIT_INDEX (FRQ_ST_MEAN)
661 | BIT_INDEX (FRQ_ST_STDDEV)
662 | BIT_INDEX (FRQ_ST_MINIMUM)
663 | BIT_INDEX (FRQ_ST_MAXIMUM);
667 else if (lex_match_id (lexer, "MEAN"))
669 frq.stats |= BIT_INDEX (FRQ_ST_MEAN);
672 else if (lex_match_id (lexer, "SEMEAN"))
674 frq.stats |= BIT_INDEX (FRQ_ST_SEMEAN);
677 else if (lex_match_id (lexer, "MEDIAN"))
679 frq.stats |= BIT_INDEX (FRQ_ST_MEDIAN);
682 else if (lex_match_id (lexer, "MODE"))
684 frq.stats |= BIT_INDEX (FRQ_ST_MODE);
687 else if (lex_match_id (lexer, "STDDEV"))
689 frq.stats |= BIT_INDEX (FRQ_ST_STDDEV);
692 else if (lex_match_id (lexer, "VARIANCE"))
694 frq.stats |= BIT_INDEX (FRQ_ST_MEAN);
697 else if (lex_match_id (lexer, "KURTOSIS"))
699 frq.stats |= BIT_INDEX (FRQ_ST_KURTOSIS);
702 else if (lex_match_id (lexer, "SKEWNESS"))
704 frq.stats |= BIT_INDEX (FRQ_ST_SKEWNESS);
707 else if (lex_match_id (lexer, "RANGE"))
709 frq.stats |= BIT_INDEX (FRQ_ST_RANGE);
712 else if (lex_match_id (lexer, "MINIMUM"))
714 frq.stats |= BIT_INDEX (FRQ_ST_MINIMUM);
717 else if (lex_match_id (lexer, "MAXIMUM"))
719 frq.stats |= BIT_INDEX (FRQ_ST_MAXIMUM);
722 else if (lex_match_id (lexer, "SUM"))
724 frq.stats |= BIT_INDEX (FRQ_ST_SUM);
727 else if (lex_match_id (lexer, "SESKEWNESS"))
729 frq.stats |= BIT_INDEX (FRQ_ST_SESKEWNESS);
732 else if (lex_match_id (lexer, "SEKURTOSIS"))
734 frq.stats |= BIT_INDEX (FRQ_ST_SEKURTOSIS);
737 else if (lex_match_id (lexer, "NONE"))
742 else if (lex_match (lexer, T_ALL))
745 frq.n_stats = FRQ_ST_count;
749 lex_error (lexer, NULL);
754 else if (lex_match_id (lexer, "PERCENTILES"))
756 lex_match (lexer, T_EQUALS);
757 while (lex_token (lexer) != T_ENDCMD
758 && lex_token (lexer) != T_SLASH)
760 if (lex_force_num (lexer))
763 xrealloc (frq.percentiles,
764 (frq.n_percentiles + 1)
765 * sizeof (*frq.percentiles));
766 frq.percentiles[frq.n_percentiles].p = lex_number (lexer) / 100.0;
767 frq.percentiles[frq.n_percentiles].show = true;
770 frq.n_show_percentiles++;
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"))
790 else if (lex_match_id (lexer, "NOTABLE"))
792 frq.max_categories = 0;
794 else if (lex_match_id (lexer, "AVALUE"))
796 frq.sort = FRQ_AVALUE;
798 else if (lex_match_id (lexer, "DVALUE"))
800 frq.sort = FRQ_DVALUE;
802 else if (lex_match_id (lexer, "AFREQ"))
804 frq.sort = FRQ_AFREQ;
806 else if (lex_match_id (lexer, "DFREQ"))
808 frq.sort = FRQ_DFREQ;
812 lex_error (lexer, NULL);
817 else if (lex_match_id (lexer, "NTILES"))
819 lex_match (lexer, T_EQUALS);
821 if (lex_force_int (lexer))
824 int n = lex_integer (lexer);
826 for (i = 0; i < n + 1; ++i)
829 xrealloc (frq.percentiles,
830 (frq.n_percentiles + 1)
831 * sizeof (*frq.percentiles));
832 frq.percentiles[frq.n_percentiles].p =
834 frq.percentiles[frq.n_percentiles].show = true;
837 frq.n_show_percentiles++;
842 lex_error (lexer, NULL);
846 else if (lex_match_id (lexer, "ALGORITHM"))
848 lex_match (lexer, T_EQUALS);
850 if (lex_match_id (lexer, "COMPATIBLE"))
852 settings_set_cmd_algorithm (COMPATIBLE);
854 else if (lex_match_id (lexer, "ENHANCED"))
856 settings_set_cmd_algorithm (ENHANCED);
860 lex_error (lexer, NULL);
864 else if (lex_match_id (lexer, "HISTOGRAM"))
866 lex_match (lexer, T_EQUALS);
867 sbc_histogram = true;
869 while (lex_token (lexer) != T_ENDCMD
870 && lex_token (lexer) != T_SLASH)
872 if (lex_match_id (lexer, "NORMAL"))
874 hi_norm = FRQ_NORMAL;
876 else if (lex_match_id (lexer, "NONORMAL"))
878 hi_norm = FRQ_NONORMAL;
880 else if (lex_match_id (lexer, "FREQ"))
883 if (lex_match (lexer, T_LPAREN))
885 if (lex_force_int (lexer))
887 hi_freq = lex_integer (lexer);
890 lex_error (lexer, _("Histogram frequency must be greater than zero."));
893 lex_force_match (lexer, T_RPAREN);
897 else if (lex_match_id (lexer, "PERCENT"))
899 hi_scale = FRQ_PERCENT;
900 if (lex_match (lexer, T_LPAREN))
902 if (lex_force_int (lexer))
904 hi_pcnt = lex_integer (lexer);
907 lex_error (lexer, _("Histogram percentage must be greater than zero."));
910 lex_force_match (lexer, T_RPAREN);
914 else if (lex_match_id (lexer, "MINIMUM"))
916 lex_force_match (lexer, T_LPAREN);
917 if (lex_force_num (lexer))
919 hi_min = lex_number (lexer);
922 lex_force_match (lexer, T_RPAREN);
924 else if (lex_match_id (lexer, "MAXIMUM"))
926 lex_force_match (lexer, T_LPAREN);
927 if (lex_force_num (lexer))
929 hi_max = lex_number (lexer);
932 lex_force_match (lexer, T_RPAREN);
936 lex_error (lexer, NULL);
941 else if (lex_match_id (lexer, "PIECHART"))
943 lex_match (lexer, T_EQUALS);
944 while (lex_token (lexer) != T_ENDCMD
945 && lex_token (lexer) != T_SLASH)
947 if (lex_match_id (lexer, "MINIMUM"))
949 lex_force_match (lexer, T_LPAREN);
950 if (lex_force_num (lexer))
952 pie_min = lex_number (lexer);
955 lex_force_match (lexer, T_RPAREN);
957 else if (lex_match_id (lexer, "MAXIMUM"))
959 lex_force_match (lexer, T_LPAREN);
960 if (lex_force_num (lexer))
962 pie_max = lex_number (lexer);
965 lex_force_match (lexer, T_RPAREN);
967 else if (lex_match_id (lexer, "MISSING"))
971 else if (lex_match_id (lexer, "NOMISSING"))
977 lex_error (lexer, NULL);
983 else if (lex_match_id (lexer, "MISSING"))
985 lex_match (lexer, T_EQUALS);
987 while (lex_token (lexer) != T_ENDCMD
988 && lex_token (lexer) != T_SLASH)
990 if (lex_match_id (lexer, "EXCLUDE"))
993 else if (lex_match_id (lexer, "INCLUDE"))
998 lex_error (lexer, NULL);
1005 lex_error (lexer, NULL);
1010 if (frq.stats & BIT_INDEX (FRQ_ST_MEDIAN))
1013 xrealloc (frq.percentiles,
1014 (frq.n_percentiles + 1)
1015 * sizeof (*frq.percentiles));
1017 frq.percentiles[frq.n_percentiles].p = 0.50;
1018 frq.percentiles[frq.n_percentiles].show = true;
1020 frq.n_percentiles++;
1024 /* Figure out which charts the user requested. */
1028 msg (SW, _("Bar charts are not implemented."));
1032 struct frq_chart *hist;
1034 hist = frq.hist = xmalloc (sizeof *frq.hist);
1035 hist->x_min = hi_min;
1036 hist->x_max = hi_max;
1037 hist->y_scale = hi_scale;
1038 hist->y_max = hi_scale == FRQ_FREQ ? hi_freq : hi_pcnt;
1039 hist->draw_normal = hi_norm != FRQ_NONORMAL;
1040 hist->include_missing = false;
1042 if (hist->x_min != SYSMIS && hist->x_max != SYSMIS
1043 && hist->x_min >= hist->x_max)
1045 msg (SE, _("%s for histogram must be greater than or equal to %s, "
1046 "but %s was specified as %.15g and %s as %.15g. "
1047 "%s and %s will be ignored."),
1052 hist->x_min = hist->x_max = SYSMIS;
1056 xrealloc (frq.percentiles,
1057 (frq.n_percentiles + 2)
1058 * sizeof (*frq.percentiles));
1060 frq.percentiles[frq.n_percentiles].p = 0.25;
1061 frq.percentiles[frq.n_percentiles].show = false;
1063 frq.percentiles[frq.n_percentiles + 1].p = 0.75;
1064 frq.percentiles[frq.n_percentiles + 1].show = false;
1066 frq.n_percentiles+=2;
1071 struct frq_chart *pie;
1073 pie = frq.pie = xmalloc (sizeof *frq.pie);
1074 pie->x_min = pie_min;
1075 pie->x_max = pie_max;
1076 pie->include_missing = pie_missing;
1078 if (pie->x_min != SYSMIS && pie->x_max != SYSMIS
1079 && pie->x_min >= pie->x_max)
1081 msg (SE, _("%s for pie chart must be greater than or equal to %s, "
1082 "but %s was specified as %.15g and %s as %.15g. "
1083 "%s and %s will be ignored."),
1088 pie->x_min = pie->x_max = SYSMIS;
1095 double previous_p = -1;
1096 qsort (frq.percentiles, frq.n_percentiles,
1097 sizeof (*frq.percentiles),
1100 frq.n_show_percentiles = 0;
1101 for (i = o = 0; i < frq.n_percentiles; ++i)
1103 frq.percentiles[o].p = frq.percentiles[i].p;
1105 if (frq.percentiles[i].show)
1106 frq.percentiles[o].show = true;
1108 if (frq.percentiles[i].p != previous_p)
1110 if (frq.percentiles[i].show)
1111 frq.n_show_percentiles++;
1116 previous_p = frq.percentiles[i].p;
1119 frq.n_percentiles = o;
1123 struct casegrouper *grouper;
1124 struct casereader *group;
1127 grouper = casegrouper_create_splits (proc_open (ds), dataset_dict (ds));
1128 while (casegrouper_get_next_group (grouper, &group))
1131 precalc (&frq, group, ds);
1132 for (; (c = casereader_read (group)) != NULL; case_unref (c))
1134 postcalc (&frq, ds);
1136 ok = casegrouper_destroy (grouper);
1137 ok = proc_commit (ds) && ok;
1149 calculate_iqr (const struct frq_proc *frq)
1155 /* This cannot work unless the 25th and 75th percentile are calculated */
1156 assert (frq->n_percentiles >= 2);
1157 for (i = 0; i < frq->n_percentiles; i++)
1159 struct percentile *pc = &frq->percentiles[i];
1161 if (fabs (0.25 - pc->p) < DBL_EPSILON)
1163 else if (fabs (0.75 - pc->p) < DBL_EPSILON)
1167 return q1 == SYSMIS || q3 == SYSMIS ? SYSMIS : q3 - q1;
1171 chart_includes_value (const struct frq_chart *chart,
1172 const struct variable *var,
1173 const union value *value)
1175 if (!chart->include_missing && var_is_value_missing (var, value, MV_ANY))
1178 if (var_is_numeric (var)
1179 && ((chart->x_min != SYSMIS && value->f < chart->x_min)
1180 || (chart->x_max != SYSMIS && value->f > chart->x_max)))
1186 /* Create a gsl_histogram from a freq_tab */
1187 static struct histogram *
1188 freq_tab_to_hist (const struct frq_proc *frq, const struct freq_tab *ft,
1189 const struct variable *var)
1191 double x_min, x_max, valid_freq;
1194 struct histogram *histogram;
1197 /* Find out the extremes of the x value, within the range to be included in
1198 the histogram, and sum the total frequency of those values. */
1202 for (i = 0; i < ft->n_valid; i++)
1204 const struct freq *f = &ft->valid[i];
1205 if (chart_includes_value (frq->hist, var, &f->value))
1207 x_min = MIN (x_min, f->value.f);
1208 x_max = MAX (x_max, f->value.f);
1209 valid_freq += f->count;
1214 iqr = calculate_iqr (frq);
1217 /* Freedman-Diaconis' choice of bin width. */
1218 bin_width = 2 * iqr / pow (valid_freq, 1.0 / 3.0);
1222 bin_width = (x_max - x_min) / (1 + log2 (valid_freq));
1224 histogram = histogram_create (bin_width, x_min, x_max);
1226 if ( histogram == NULL)
1229 for (i = 0; i < ft->n_valid; i++)
1231 const struct freq *f = &ft->valid[i];
1232 if (chart_includes_value (frq->hist, var, &f->value))
1233 histogram_add (histogram, f->value.f, f->count);
1240 add_slice (const struct frq_chart *pie, const struct freq *freq,
1241 const struct variable *var, struct slice *slice)
1243 if (chart_includes_value (pie, var, &freq->value))
1245 ds_init_empty (&slice->label);
1246 var_append_value_name (var, &freq->value, &slice->label);
1247 slice->magnitude = freq->count;
1254 /* Allocate an array of slices and fill them from the data in frq_tab
1255 n_slices will contain the number of slices allocated.
1256 The caller is responsible for freeing slices
1258 static struct slice *
1259 freq_tab_to_slice_array(const struct frq_chart *pie,
1260 const struct freq_tab *frq_tab,
1261 const struct variable *var,
1264 struct slice *slices;
1268 slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1271 for (i = 0; i < frq_tab->n_valid; i++)
1272 n_slices += add_slice (pie, &frq_tab->valid[i], var, &slices[n_slices]);
1273 for (i = 0; i < frq_tab->n_missing; i++)
1274 n_slices += add_slice (pie, &frq_tab->missing[i], var, &slices[n_slices]);
1276 *n_slicesp = n_slices;
1282 do_piechart(const struct frq_chart *pie, const struct variable *var,
1283 const struct freq_tab *frq_tab)
1285 struct slice *slices;
1288 slices = freq_tab_to_slice_array (pie, frq_tab, var, &n_slices);
1291 msg (SW, _("Omitting pie chart for %s, which has only %d unique values."),
1292 var_get_name (var), n_slices);
1293 else if (n_slices > 50)
1294 msg (SW, _("Omitting pie chart for %s, which has over 50 unique values."),
1295 var_get_name (var));
1297 chart_item_submit (piechart_create (var_to_string(var), slices, n_slices));
1299 for (i = 0; i < n_slices; i++)
1300 ds_destroy (&slices[i].label);
1304 /* Calculates all the pertinent statistics for VF, putting them in array
1307 calc_stats (const struct var_freqs *vf, double d[FRQ_ST_count])
1309 const struct freq_tab *ft = &vf->tab;
1310 double W = ft->valid_cases;
1311 const struct freq *f;
1313 int most_often = -1;
1314 double X_mode = SYSMIS;
1316 /* Calculate the mode. */
1317 for (f = ft->valid; f < ft->missing; f++)
1319 if (most_often < f->count)
1321 most_often = f->count;
1322 X_mode = f->value.f;
1324 else if (most_often == f->count)
1326 /* A duplicate mode is undefined.
1327 FIXME: keep track of *all* the modes. */
1332 /* Calculate moments. */
1333 m = moments_create (MOMENT_KURTOSIS);
1334 for (f = ft->valid; f < ft->missing; f++)
1335 moments_pass_one (m, f->value.f, f->count);
1336 for (f = ft->valid; f < ft->missing; f++)
1337 moments_pass_two (m, f->value.f, f->count);
1338 moments_calculate (m, NULL, &d[FRQ_ST_MEAN], &d[FRQ_ST_VARIANCE],
1339 &d[FRQ_ST_SKEWNESS], &d[FRQ_ST_KURTOSIS]);
1340 moments_destroy (m);
1342 /* Formulae below are taken from _SPSS Statistical Algorithms_. */
1343 d[FRQ_ST_MINIMUM] = ft->valid[0].value.f;
1344 d[FRQ_ST_MAXIMUM] = ft->valid[ft->n_valid - 1].value.f;
1345 d[FRQ_ST_MODE] = X_mode;
1346 d[FRQ_ST_RANGE] = d[FRQ_ST_MAXIMUM] - d[FRQ_ST_MINIMUM];
1347 d[FRQ_ST_SUM] = d[FRQ_ST_MEAN] * W;
1348 d[FRQ_ST_STDDEV] = sqrt (d[FRQ_ST_VARIANCE]);
1349 d[FRQ_ST_SEMEAN] = d[FRQ_ST_STDDEV] / sqrt (W);
1350 d[FRQ_ST_SESKEWNESS] = calc_seskew (W);
1351 d[FRQ_ST_SEKURTOSIS] = calc_sekurt (W);
1354 /* Displays a table of all the statistics requested for variable V. */
1356 dump_statistics (const struct frq_proc *frq, const struct var_freqs *vf,
1357 const struct variable *wv)
1359 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1360 const struct freq_tab *ft = &vf->tab;
1361 double stat_value[FRQ_ST_count];
1362 struct tab_table *t;
1363 int i, r = 2; /* N missing and N valid are always dumped */
1365 if (var_is_alpha (vf->var))
1368 calc_stats (vf, stat_value);
1370 t = tab_create (3, ((frq->stats & BIT_INDEX (FRQ_ST_MEDIAN)) ? frq->n_stats - 1 : frq->n_stats)
1371 + frq->n_show_percentiles + 2);
1373 tab_set_format (t, RC_WEIGHT, wfmt);
1374 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1376 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1377 tab_vline (t, TAL_GAP , 1, 0, tab_nr(t) - 1 ) ;
1379 for (i = 0; i < FRQ_ST_count; i++)
1381 if (FRQ_ST_MEDIAN == i)
1384 if (frq->stats & BIT_INDEX (i))
1386 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1387 gettext (st_name[i]));
1389 if (vf->tab.n_valid <= 0 && r >= 2)
1390 tab_text (t, 2, r, 0, ".");
1392 tab_double (t, 2, r, TAB_NONE, stat_value[i], NULL, RC_OTHER);
1397 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1398 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1399 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1401 tab_double (t, 2, 0, TAB_NONE, ft->valid_cases, NULL, RC_WEIGHT);
1402 tab_double (t, 2, 1, TAB_NONE, ft->total_cases - ft->valid_cases, NULL, RC_WEIGHT);
1404 for (i = 0; i < frq->n_percentiles; i++)
1406 const struct percentile *pc = &frq->percentiles[i];
1413 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1416 if (vf->tab.n_valid <= 0)
1418 tab_text (t, 2, r, 0, ".");
1424 tab_text (t, 1, r, TAB_LEFT, _("50 (Median)"));
1426 tab_double (t, 1, r, TAB_LEFT, pc->p * 100, NULL, RC_INTEGER);
1427 tab_double (t, 2, r, TAB_NONE, pc->value,
1428 var_get_print_format (vf->var), RC_OTHER);
1433 tab_title (t, "%s", var_to_string (vf->var));