1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2007, 2009, 2010 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <http://www.gnu.org/licenses/>. */
21 #include <gsl/gsl_histogram.h>
23 #include "data/case.h"
24 #include "data/casegrouper.h"
25 #include "data/casereader.h"
26 #include "data/dictionary.h"
27 #include "data/format.h"
28 #include "data/procedure.h"
29 #include "data/settings.h"
30 #include "data/value-labels.h"
31 #include "data/variable.h"
32 #include "language/command.h"
33 #include "language/dictionary/split-file.h"
34 #include "language/lexer/lexer.h"
35 #include "language/stats/freq.h"
36 #include "libpspp/array.h"
37 #include "libpspp/bit-vector.h"
38 #include "libpspp/compiler.h"
39 #include "libpspp/hmap.h"
40 #include "libpspp/message.h"
41 #include "libpspp/misc.h"
42 #include "libpspp/pool.h"
43 #include "libpspp/str.h"
44 #include "math/histogram.h"
45 #include "math/moments.h"
46 #include "output/chart-item.h"
47 #include "output/charts/piechart.h"
48 #include "output/charts/plot-hist.h"
49 #include "output/tab.h"
51 #include "gl/minmax.h"
52 #include "gl/xalloc.h"
55 #define _(msgid) gettext (msgid)
56 #define N_(msgid) msgid
63 +format=table:limit(n:limit,"%s>0")/notable/!table,
64 sort:!avalue/dvalue/afreq/dfreq;
65 missing=miss:include/!exclude;
66 barchart(ba_)=:minimum(d:min),
68 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0");
69 piechart(pie_)=:minimum(d:min),
71 missing:missing/!nomissing,
73 histogram(hi_)=:minimum(d:min),
75 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
76 norm:!nonormal/normal;
79 +percentiles = double list;
80 +statistics[st_]=mean,semean,median,mode,stddev,variance,
81 kurtosis,skewness,range,minimum,maximum,sum,
82 default,seskewness,sekurtosis,all,none.
90 FRQ_MEAN, FRQ_SEMEAN, FRQ_MEDIAN, FRQ_MODE, FRQ_STDDEV, FRQ_VARIANCE,
91 FRQ_KURT, FRQ_SEKURT, FRQ_SKEW, FRQ_SESKEW, FRQ_RANGE, FRQ_MIN, FRQ_MAX,
95 /* Description of a statistic. */
98 int st_indx; /* Index into a_statistics[]. */
99 const char *s10; /* Identifying string. */
102 /* Table of statistics, indexed by dsc_*. */
103 static const struct frq_info st_name[FRQ_N_STATS + 1] =
105 {FRQ_ST_MEAN, N_("Mean")},
106 {FRQ_ST_SEMEAN, N_("S.E. Mean")},
107 {FRQ_ST_MEDIAN, N_("Median")},
108 {FRQ_ST_MODE, N_("Mode")},
109 {FRQ_ST_STDDEV, N_("Std Dev")},
110 {FRQ_ST_VARIANCE, N_("Variance")},
111 {FRQ_ST_KURTOSIS, N_("Kurtosis")},
112 {FRQ_ST_SEKURTOSIS, N_("S.E. Kurt")},
113 {FRQ_ST_SKEWNESS, N_("Skewness")},
114 {FRQ_ST_SESKEWNESS, N_("S.E. Skew")},
115 {FRQ_ST_RANGE, N_("Range")},
116 {FRQ_ST_MINIMUM, N_("Minimum")},
117 {FRQ_ST_MAXIMUM, N_("Maximum")},
118 {FRQ_ST_SUM, N_("Sum")},
122 /* Percentiles to calculate. */
126 double p; /* the %ile to be calculated */
127 double value; /* the %ile's value */
128 double x1; /* The datum value <= the percentile */
129 double x2; /* The datum value >= the percentile */
131 int flag2; /* Set to 1 if this percentile value has been found */
132 bool show; /* True to show this percentile in the statistics box. */
135 /* Groups of statistics. */
137 #define FRQ_DEFAULT \
138 (BI (FRQ_MEAN) | BI (FRQ_STDDEV) | BI (FRQ_MIN) | BI (FRQ_MAX))
140 (BI (FRQ_SUM) | BI(FRQ_MIN) | BI(FRQ_MAX) \
141 | BI(FRQ_MEAN) | BI(FRQ_SEMEAN) | BI(FRQ_STDDEV) \
142 | BI(FRQ_VARIANCE) | BI(FRQ_KURT) | BI(FRQ_SEKURT) \
143 | BI(FRQ_SKEW) | BI(FRQ_SESKEW) | BI(FRQ_RANGE) \
144 | BI(FRQ_RANGE) | BI(FRQ_MODE) | BI(FRQ_MEDIAN))
148 double x_min; /* X axis minimum value. */
149 double x_max; /* X axis maximum value. */
150 int y_scale; /* Y axis scale: FRQ_FREQ or FRQ_PERCENT. */
152 /* Histograms only. */
153 double y_max; /* Y axis maximum value. */
154 bool draw_normal; /* Whether to draw normal curve. */
156 /* Pie charts only. */
157 bool include_missing; /* Whether to include missing values. */
160 /* Frequency tables. */
162 /* Entire frequency table. */
165 struct hmap data; /* Hash table for accumulating counts. */
166 struct freq *valid; /* Valid freqs. */
167 int n_valid; /* Number of total freqs. */
168 const struct dictionary *dict; /* Source of entries in the table. */
170 struct freq *missing; /* Missing freqs. */
171 int n_missing; /* Number of missing freqs. */
174 double total_cases; /* Sum of weights of all cases. */
175 double valid_cases; /* Sum of weights of valid cases. */
178 /* Per-variable frequency data. */
181 struct variable *var;
183 /* Freqency table. */
184 struct freq_tab tab; /* Frequencies table to use. */
187 int n_groups; /* Number of groups. */
188 double *groups; /* Groups. */
191 double stat[FRQ_N_STATS];
193 /* Variable attributes. */
195 struct fmt_spec print;
202 struct var_freqs *vars;
205 /* Percentiles to calculate and possibly display. */
206 struct percentile *percentiles;
207 int n_percentiles, n_show_percentiles;
209 /* Frequency table display. */
210 int max_categories; /* Maximum categories to show. */
211 int sort; /* FRQ_AVALUE or FRQ_DVALUE
212 or FRQ_ACOUNT or FRQ_DCOUNT. */
214 /* Statistics; number of statistics. */
218 /* Histogram and pie chart settings. */
219 struct frq_chart *hist, *pie;
222 static void determine_charts (struct frq_proc *,
223 const struct cmd_frequencies *);
225 static void calc_stats (const struct frq_proc *, const struct var_freqs *,
226 double d[FRQ_N_STATS]);
228 static void precalc (struct frq_proc *, struct casereader *, struct dataset *);
229 static void calc (struct frq_proc *, const struct ccase *,
230 const struct dataset *);
231 static void postcalc (struct frq_proc *, const struct dataset *);
233 static void postprocess_freq_tab (const struct frq_proc *, struct var_freqs *);
234 static void dump_freq_table (const struct var_freqs *,
235 const struct variable *weight_var);
236 static void dump_statistics (const struct frq_proc *, const struct var_freqs *,
237 const struct variable *weight_var);
238 static void cleanup_freq_tab (struct var_freqs *);
240 static void add_percentile (struct frq_proc *, double x, bool show,
241 size_t *allocated_percentiles);
243 static void do_piechart(const struct frq_chart *, const struct variable *,
244 const struct freq_tab *);
246 struct histogram *freq_tab_to_hist(const struct frq_proc *,
247 const struct freq_tab *,
248 const struct variable *);
250 /* Parser and outline. */
253 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
255 struct cmd_frequencies cmd;
257 struct casegrouper *grouper;
258 struct casereader *input, *group;
259 size_t allocated_percentiles;
263 frq.pool = pool_create ();
268 frq.percentiles = NULL;
269 frq.n_percentiles = 0;
270 frq.n_show_percentiles = 0;
275 allocated_percentiles = 0;
277 if (!parse_frequencies (lexer, ds, &cmd, &frq))
279 pool_destroy (frq.pool);
283 /* Figure out when to show frequency tables. */
284 frq.max_categories = (cmd.table == FRQ_NOTABLE ? -1
285 : cmd.table == FRQ_TABLE ? INT_MAX
289 /* Figure out statistics to calculate. */
291 if (cmd.a_statistics[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
292 frq.stats |= FRQ_DEFAULT;
293 if (cmd.a_statistics[FRQ_ST_ALL])
294 frq.stats |= FRQ_ALL;
295 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
296 frq.stats &= ~BIT_INDEX (FRQ_MEDIAN);
297 for (i = 0; i < FRQ_N_STATS; i++)
298 if (cmd.a_statistics[st_name[i].st_indx])
299 frq.stats |= BIT_INDEX (i);
300 if (frq.stats & FRQ_KURT)
301 frq.stats |= BIT_INDEX (FRQ_SEKURT);
302 if (frq.stats & FRQ_SKEW)
303 frq.stats |= BIT_INDEX (FRQ_SESKEW);
305 /* Calculate n_stats. */
307 for (i = 0; i < FRQ_N_STATS; i++)
308 if ((frq.stats & BIT_INDEX (i)))
312 determine_charts (&frq, &cmd);
313 if (cmd.sbc_histogram || cmd.sbc_piechart || cmd.sbc_ntiles)
314 cmd.sort = FRQ_AVALUE;
316 /* Work out what percentiles need to be calculated */
317 if ( cmd.sbc_percentiles )
319 for ( i = 0 ; i < MAXLISTS ; ++i )
322 subc_list_double *ptl_list = &cmd.dl_percentiles[i];
323 for ( pl = 0 ; pl < subc_list_double_count(ptl_list); ++pl)
324 add_percentile (&frq, subc_list_double_at(ptl_list, pl) / 100.0,
325 true, &allocated_percentiles);
328 if ( cmd.sbc_ntiles )
330 for ( i = 0 ; i < cmd.sbc_ntiles ; ++i )
333 for (j = 0; j <= cmd.n_ntiles[i]; ++j )
334 add_percentile (&frq, j / (double) cmd.n_ntiles[i], true,
335 &allocated_percentiles);
338 if (frq.stats & BIT_INDEX (FRQ_MEDIAN))
340 /* Treat the median as the 50% percentile.
341 We output it in the percentiles table as "50 (Median)." */
342 add_percentile (&frq, 0.5, true, &allocated_percentiles);
343 frq.stats &= ~BIT_INDEX (FRQ_MEDIAN);
346 if (cmd.sbc_histogram)
348 add_percentile (&frq, 0.25, false, &allocated_percentiles);
349 add_percentile (&frq, 0.75, false, &allocated_percentiles);
353 input = casereader_create_filter_weight (proc_open (ds), dataset_dict (ds),
355 grouper = casegrouper_create_splits (input, dataset_dict (ds));
356 for (; casegrouper_get_next_group (grouper, &group);
357 casereader_destroy (group))
361 precalc (&frq, group, ds);
362 for (; (c = casereader_read (group)) != NULL; case_unref (c))
366 ok = casegrouper_destroy (grouper);
367 ok = proc_commit (ds) && ok;
369 free_frequencies(&cmd);
371 pool_destroy (frq.pool);
373 free (frq.percentiles);
377 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
380 /* Figure out which charts the user requested. */
382 determine_charts (struct frq_proc *frq, const struct cmd_frequencies *cmd)
384 if (cmd->sbc_barchart)
385 msg (SW, _("Bar charts are not implemented."));
387 if (cmd->sbc_histogram)
389 struct frq_chart *hist;
391 hist = frq->hist = xmalloc (sizeof *frq->hist);
392 hist->x_min = cmd->hi_min;
393 hist->x_max = cmd->hi_max;
394 hist->y_scale = cmd->hi_scale;
395 hist->y_max = cmd->hi_scale == FRQ_FREQ ? cmd->hi_freq : cmd->hi_pcnt;
396 hist->draw_normal = cmd->hi_norm != FRQ_NONORMAL;
397 hist->include_missing = false;
399 if (hist->x_min != SYSMIS && hist->x_max != SYSMIS
400 && hist->x_min >= hist->x_max)
402 msg (SE, _("MAX for histogram must be greater than or equal to MIN, "
403 "but MIN was specified as %.15g and MAX as %.15g. "
404 "MIN and MAX will be ignored."),
405 hist->x_min, hist->x_max);
406 hist->x_min = hist->x_max = SYSMIS;
410 if (cmd->sbc_piechart)
412 struct frq_chart *pie;
414 pie = frq->pie = xmalloc (sizeof *frq->pie);
415 pie->x_min = cmd->pie_min;
416 pie->x_max = cmd->pie_max;
417 pie->y_scale = cmd->pie_scale;
418 pie->include_missing = cmd->pie_missing == FRQ_MISSING;
420 if (pie->x_min != SYSMIS && pie->x_max != SYSMIS
421 && pie->x_min >= pie->x_max)
423 msg (SE, _("MAX for pie chart must be greater than or equal to MIN, "
424 "but MIN was specified as %.15g and MAX as %.15g. "
425 "MIN and MAX will be ignored."), pie->x_min, pie->x_max);
426 pie->x_min = pie->x_max = SYSMIS;
431 /* Add data from case C to the frequency table. */
433 calc (struct frq_proc *frq, const struct ccase *c, const struct dataset *ds)
435 double weight = dict_get_case_weight (dataset_dict (ds), c, NULL);
438 for (i = 0; i < frq->n_vars; i++)
440 struct var_freqs *vf = &frq->vars[i];
441 const union value *value = case_data (c, vf->var);
442 size_t hash = value_hash (value, vf->width, 0);
445 f = freq_hmap_search (&vf->tab.data, value, vf->width, hash);
447 f = freq_hmap_insert (&vf->tab.data, value, vf->width, hash);
453 /* Prepares each variable that is the target of FREQUENCIES by setting
454 up its hash table. */
456 precalc (struct frq_proc *frq, struct casereader *input, struct dataset *ds)
461 c = casereader_peek (input, 0);
464 output_split_file_values (ds, c);
468 for (i = 0; i < frq->n_vars; i++)
469 hmap_init (&frq->vars[i].tab.data);
472 /* Finishes up with the variables after frequencies have been
473 calculated. Displays statistics, percentiles, ... */
475 postcalc (struct frq_proc *frq, const struct dataset *ds)
477 const struct dictionary *dict = dataset_dict (ds);
478 const struct variable *wv = dict_get_weight (dict);
481 for (i = 0; i < frq->n_vars; i++)
483 struct var_freqs *vf = &frq->vars[i];
485 postprocess_freq_tab (frq, vf);
487 /* Frequencies tables. */
488 if (vf->tab.n_valid + vf->tab.n_missing <= frq->max_categories)
489 dump_freq_table (vf, wv);
493 dump_statistics (frq, vf, wv);
495 if (frq->hist && var_is_numeric (vf->var) && vf->tab.n_valid > 0)
497 double d[FRQ_N_STATS];
498 struct histogram *histogram;
500 calc_stats (frq, vf, d);
502 histogram = freq_tab_to_hist (frq, &vf->tab, vf->var);
504 chart_item_submit (histogram_chart_create (
505 histogram->gsl_hist, var_to_string(vf->var),
509 frq->hist->draw_normal));
511 statistic_destroy (&histogram->parent);
515 do_piechart(frq->pie, vf->var, &vf->tab);
517 cleanup_freq_tab (vf);
522 /* Returns true iff the value in struct freq F is non-missing
525 not_missing (const void *f_, const void *v_)
527 const struct freq *f = f_;
528 const struct variable *v = v_;
530 return !var_is_value_missing (v, &f->value, MV_ANY);
533 struct freq_compare_aux
539 bool ascending_value;
543 compare_freq (const void *a_, const void *b_, const void *aux_)
545 const struct freq_compare_aux *aux = aux_;
546 const struct freq *a = a_;
547 const struct freq *b = b_;
549 if (aux->by_freq && a->count != b->count)
551 int cmp = a->count > b->count ? 1 : -1;
552 return aux->ascending_freq ? cmp : -cmp;
556 int cmp = value_compare_3way (&a->value, &b->value, aux->width);
557 return aux->ascending_value ? cmp : -cmp;
560 /* Summarizes the frequency table data for variable V. */
562 postprocess_freq_tab (const struct frq_proc *frq, struct var_freqs *vf)
564 struct freq_tab *ft = &vf->tab;
565 struct freq_compare_aux aux;
567 struct freq *freqs, *f;
570 /* Extract data from hash table. */
571 count = hmap_count (&ft->data);
572 freqs = freq_hmap_extract (&ft->data);
574 /* Put data into ft. */
576 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, vf->var);
577 ft->missing = freqs + ft->n_valid;
578 ft->n_missing = count - ft->n_valid;
581 aux.by_freq = frq->sort == FRQ_AFREQ || frq->sort == FRQ_DFREQ;
582 aux.ascending_freq = frq->sort != FRQ_DFREQ;
583 aux.width = vf->width;
584 aux.ascending_value = frq->sort != FRQ_DVALUE;
585 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare_freq, &aux);
586 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare_freq, &aux);
588 /* Summary statistics. */
589 ft->valid_cases = 0.0;
590 for(i = 0 ; i < ft->n_valid ; ++i )
593 ft->valid_cases += f->count;
597 ft->total_cases = ft->valid_cases ;
598 for(i = 0 ; i < ft->n_missing ; ++i )
601 ft->total_cases += f->count;
606 /* Frees the frequency table for variable V. */
608 cleanup_freq_tab (struct var_freqs *vf)
610 free (vf->tab.valid);
611 freq_hmap_destroy (&vf->tab.data, vf->width);
614 /* Parses the VARIABLES subcommand. */
616 frq_custom_variables (struct lexer *lexer, struct dataset *ds,
617 struct cmd_frequencies *cmd UNUSED, void *frq_ UNUSED)
619 struct frq_proc *frq = frq_;
620 struct variable **vars;
624 lex_match (lexer, '=');
625 if (lex_token (lexer) != T_ALL
626 && (lex_token (lexer) != T_ID
627 || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) == NULL))
630 /* Get list of current variables, to avoid duplicates. */
631 vars = xmalloc (frq->n_vars * sizeof *vars);
632 n_vars = frq->n_vars;
633 for (i = 0; i < frq->n_vars; i++)
634 vars[i] = frq->vars[i].var;
636 if (!parse_variables (lexer, dataset_dict (ds), &vars, &n_vars,
637 PV_APPEND | PV_NO_SCRATCH))
640 frq->vars = xrealloc (frq->vars, n_vars * sizeof *frq->vars);
641 for (i = frq->n_vars; i < n_vars; i++)
643 struct variable *var = vars[i];
644 struct var_freqs *vf = &frq->vars[i];
647 vf->tab.valid = vf->tab.missing = NULL;
648 vf->tab.dict = dataset_dict (ds);
651 vf->width = var_get_width (var);
652 vf->print = *var_get_print_format (var);
654 frq->n_vars = n_vars;
661 /* Parses the GROUPED subcommand, setting the n_grouped, grouped
662 fields of specified variables. */
664 frq_custom_grouped (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *frq_ UNUSED)
666 struct frq_proc *frq = frq_;
668 lex_match (lexer, '=');
669 if ((lex_token (lexer) == T_ID && dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
670 || lex_token (lexer) == T_ID)
675 /* Max, current size of list; list itself. */
681 const struct variable **v;
683 if (!parse_variables_const (lexer, dataset_dict (ds), &v, &n,
684 PV_NO_DUPLICATE | PV_NUMERIC))
686 if (lex_match (lexer, '('))
690 while (lex_integer (lexer))
695 dl = pool_nrealloc (frq->pool, dl, ml, sizeof *dl);
697 dl[nl++] = lex_tokval (lexer);
699 lex_match (lexer, ',');
701 /* Note that nl might still be 0 and dl might still be
702 NULL. That's okay. */
703 if (!lex_match (lexer, ')'))
706 msg (SE, _("`)' expected after GROUPED interval list."));
716 for (i = 0; i < n; i++)
720 for (j = 0; j < frq->n_vars; j++)
722 struct var_freqs *vf = &frq->vars[j];
725 if (vf->groups != NULL)
726 msg (SE, _("Variables %s specified multiple times on "
727 "GROUPED subcommand."), var_get_name (v[i]));
736 msg (SE, _("Variables %s specified on GROUPED but not on "
737 "VARIABLES."), var_get_name (v[i]));
743 if (!lex_match (lexer, '/'))
745 if ((lex_token (lexer) != T_ID || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
746 && lex_token (lexer) != T_ALL)
748 lex_put_back (lexer, '/');
756 /* Adds X to the list of percentiles, keeping the list in proper
757 order. If SHOW is true, the percentile will be shown in the statistics
758 box, otherwise it will be hidden. */
760 add_percentile (struct frq_proc *frq, double x, bool show,
761 size_t *allocated_percentiles)
765 /* Do nothing if it's already in the list */
766 for (i = 0; i < frq->n_percentiles; i++)
768 struct percentile *pc = &frq->percentiles[i];
770 if ( fabs(x - pc->p) < DBL_EPSILON )
772 if (show && !pc->show)
774 frq->n_show_percentiles++;
784 if (frq->n_percentiles >= *allocated_percentiles)
785 frq->percentiles = x2nrealloc (frq->percentiles, allocated_percentiles,
786 sizeof *frq->percentiles);
787 insert_element (frq->percentiles, frq->n_percentiles,
788 sizeof *frq->percentiles, i);
789 frq->percentiles[i].p = x;
790 frq->percentiles[i].show = show;
791 frq->n_percentiles++;
793 frq->n_show_percentiles++;
796 /* Comparison functions. */
799 /* Frequency table display. */
801 /* Displays a full frequency table for variable V. */
803 dump_freq_table (const struct var_freqs *vf, const struct variable *wv)
805 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
806 const struct freq_tab *ft = &vf->tab;
811 double cum_total = 0.0;
812 double cum_freq = 0.0;
814 static const char *headings[] = {
823 n_categories = ft->n_valid + ft->n_missing;
824 t = tab_create (6, n_categories + 2);
825 tab_headers (t, 0, 0, 1, 0);
827 for (x = 0; x < 6; x++)
828 tab_text (t, x, 0, TAB_CENTER | TAT_TITLE, gettext (headings[x]));
831 for (f = ft->valid; f < ft->missing; f++)
834 double percent, valid_percent;
836 cum_freq += f->count;
838 percent = f->count / ft->total_cases * 100.0;
839 valid_percent = f->count / ft->valid_cases * 100.0;
840 cum_total += valid_percent;
842 label = var_lookup_value_label (vf->var, &f->value);
844 tab_text (t, 0, r, TAB_LEFT, label);
846 tab_value (t, 1, r, TAB_NONE, &f->value, ft->dict, &vf->print);
847 tab_double (t, 2, r, TAB_NONE, f->count, wfmt);
848 tab_double (t, 3, r, TAB_NONE, percent, NULL);
849 tab_double (t, 4, r, TAB_NONE, valid_percent, NULL);
850 tab_double (t, 5, r, TAB_NONE, cum_total, NULL);
853 for (; f < &ft->valid[n_categories]; f++)
857 cum_freq += f->count;
859 label = var_lookup_value_label (vf->var, &f->value);
861 tab_text (t, 0, r, TAB_LEFT, label);
863 tab_value (t, 1, r, TAB_NONE, &f->value, ft->dict, &vf->print);
864 tab_double (t, 2, r, TAB_NONE, f->count, wfmt);
865 tab_double (t, 3, r, TAB_NONE,
866 f->count / ft->total_cases * 100.0, NULL);
867 tab_text (t, 4, r, TAB_NONE, _("Missing"));
871 tab_box (t, TAL_1, TAL_1, -1, TAL_1, 0, 0, 5, r);
872 tab_hline (t, TAL_2, 0, 5, 1);
873 tab_hline (t, TAL_2, 0, 5, r);
874 tab_joint_text (t, 0, r, 1, r, TAB_RIGHT | TAT_TITLE, _("Total"));
875 tab_vline (t, TAL_0, 1, r, r);
876 tab_double (t, 2, r, TAB_NONE, cum_freq, wfmt);
877 tab_fixed (t, 3, r, TAB_NONE, 100.0, 5, 1);
878 tab_fixed (t, 4, r, TAB_NONE, 100.0, 5, 1);
880 tab_title (t, "%s", var_to_string (vf->var));
884 /* Statistical display. */
886 /* Calculates all the pertinent statistics for variable V, putting them in
889 calc_stats (const struct frq_proc *frq,
890 const struct var_freqs *vf, double d[FRQ_N_STATS])
892 const struct freq_tab *ft = &vf->tab;
893 double W = ft->valid_cases;
904 /* Calculate percentiles. */
906 assert (ft->n_valid > 0);
908 for (i = 0; i < frq->n_percentiles; i++)
910 struct percentile *pc = &frq->percentiles[i];
918 for (idx = 0; idx < ft->n_valid; ++idx)
922 for (i = 0; i < frq->n_percentiles; i++)
924 struct percentile *pc = &frq->percentiles[i];
927 if ( pc->flag2 ) continue ;
929 if ( settings_get_algorithm () != COMPATIBLE )
931 (ft->valid_cases - 1) * pc->p;
934 (ft->valid_cases + 1) * pc->p - 1;
946 if ( f->count > 1 && rank - (f->count - 1) > tp )
948 pc->x2 = pc->x1 = f->value.f;
959 prev_value = f->value.f;
962 for (i = 0; i < frq->n_percentiles; i++)
964 struct percentile *pc = &frq->percentiles[i];
966 /* Catches the case when p == 100% */
968 pc->x1 = pc->x2 = f->value.f;
971 printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
972 i,pc->p,pc->x1,pc->x2);
976 for (i = 0; i < frq->n_percentiles; i++)
978 struct percentile *pc = &frq->percentiles[i];
982 if ( settings_get_algorithm () != COMPATIBLE )
984 s = modf((ft->valid_cases - 1) * pc->p , &dummy);
988 s = modf((ft->valid_cases + 1) * pc->p -1, &dummy);
991 pc->value = pc->x1 + (pc->x2 - pc->x1) * s ;
995 /* Calculate the mode. */
998 for (f = ft->valid; f < ft->missing; f++)
1000 if (most_often < f->count)
1002 most_often = f->count;
1003 X_mode = f->value.f;
1005 else if (most_often == f->count)
1007 /* A duplicate mode is undefined.
1008 FIXME: keep track of *all* the modes. */
1013 /* Calculate moments. */
1014 m = moments_create (MOMENT_KURTOSIS);
1015 for (f = ft->valid; f < ft->missing; f++)
1016 moments_pass_one (m, f->value.f, f->count);
1017 for (f = ft->valid; f < ft->missing; f++)
1018 moments_pass_two (m, f->value.f, f->count);
1019 moments_calculate (m, NULL, &d[FRQ_MEAN], &d[FRQ_VARIANCE],
1020 &d[FRQ_SKEW], &d[FRQ_KURT]);
1021 moments_destroy (m);
1023 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1024 d[FRQ_MIN] = ft->valid[0].value.f;
1025 d[FRQ_MAX] = ft->valid[ft->n_valid - 1].value.f;
1026 d[FRQ_MODE] = X_mode;
1027 d[FRQ_RANGE] = d[FRQ_MAX] - d[FRQ_MIN];
1028 d[FRQ_SUM] = d[FRQ_MEAN] * W;
1029 d[FRQ_STDDEV] = sqrt (d[FRQ_VARIANCE]);
1030 d[FRQ_SEMEAN] = d[FRQ_STDDEV] / sqrt (W);
1031 d[FRQ_SESKEW] = calc_seskew (W);
1032 d[FRQ_SEKURT] = calc_sekurt (W);
1035 /* Displays a table of all the statistics requested for variable V. */
1037 dump_statistics (const struct frq_proc *frq, const struct var_freqs *vf,
1038 const struct variable *wv)
1040 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1041 const struct freq_tab *ft = &vf->tab;
1042 double stat_value[FRQ_N_STATS];
1043 struct tab_table *t;
1046 if (var_is_alpha (vf->var))
1049 if (ft->n_valid == 0)
1051 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1052 var_get_name (vf->var));
1055 calc_stats (frq, vf, stat_value);
1057 t = tab_create (3, frq->n_stats + frq->n_show_percentiles + 2);
1059 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1062 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1063 tab_vline (t, TAL_GAP , 1, 0, tab_nr(t) - 1 ) ;
1065 r=2; /* N missing and N valid are always dumped */
1067 for (i = 0; i < FRQ_N_STATS; i++)
1068 if (frq->stats & BIT_INDEX (i))
1070 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1071 gettext (st_name[i].s10));
1072 tab_double (t, 2, r, TAB_NONE, stat_value[i], NULL);
1076 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1077 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1078 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1080 tab_double (t, 2, 0, TAB_NONE, ft->valid_cases, wfmt);
1081 tab_double (t, 2, 1, TAB_NONE, ft->total_cases - ft->valid_cases, wfmt);
1083 for (i = 0; i < frq->n_percentiles; i++, r++)
1085 struct percentile *pc = &frq->percentiles[i];
1092 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1096 tab_text (t, 1, r, TAB_LEFT, _("50 (Median)"));
1098 tab_fixed (t, 1, r, TAB_LEFT, pc->p * 100, 3, 0);
1099 tab_double (t, 2, r, TAB_NONE, pc->value,
1100 var_get_print_format (vf->var));
1103 tab_title (t, "%s", var_to_string (vf->var));
1109 calculate_iqr (const struct frq_proc *frq)
1115 for (i = 0; i < frq->n_percentiles; i++)
1117 struct percentile *pc = &frq->percentiles[i];
1119 if (fabs (0.25 - pc->p) < DBL_EPSILON)
1121 else if (fabs (0.75 - pc->p) < DBL_EPSILON)
1125 return q1 == SYSMIS || q3 == SYSMIS ? SYSMIS : q3 - q1;
1129 chart_includes_value (const struct frq_chart *chart,
1130 const struct variable *var,
1131 const union value *value)
1133 if (!chart->include_missing && var_is_value_missing (var, value, MV_ANY))
1136 if (var_is_numeric (var)
1137 && ((chart->x_min != SYSMIS && value->f < chart->x_min)
1138 || (chart->x_max != SYSMIS && value->f > chart->x_max)))
1144 /* Create a gsl_histogram from a freq_tab */
1146 freq_tab_to_hist (const struct frq_proc *frq, const struct freq_tab *ft,
1147 const struct variable *var)
1149 double x_min, x_max, valid_freq;
1152 struct histogram *histogram;
1156 /* Find out the extremes of the x value, within the range to be included in
1157 the histogram, and sum the total frequency of those values. */
1161 for (i = 0; i < ft->n_valid; i++)
1163 const struct freq *f = &ft->valid[i];
1164 if (chart_includes_value (frq->hist, var, &f->value))
1166 x_min = MIN (x_min, f->value.f);
1167 x_max = MAX (x_max, f->value.f);
1168 valid_freq += f->count;
1172 /* Freedman-Diaconis' choice of bin width. */
1173 iqr = calculate_iqr (frq);
1176 double bin_width = 2 * iqr / pow (valid_freq, 1.0 / 3.0);
1177 bins = (x_max - x_min) / bin_width;
1180 else if (bins > 400)
1186 histogram = histogram_create (bins, x_min, x_max);
1187 for (i = 0; i < ft->n_valid; i++)
1189 const struct freq *f = &ft->valid[i];
1190 if (chart_includes_value (frq->hist, var, &f->value))
1191 histogram_add (histogram, f->value.f, f->count);
1198 add_slice (const struct frq_chart *pie, const struct freq *freq,
1199 const struct variable *var, struct slice *slice)
1201 if (chart_includes_value (pie, var, &freq->value))
1203 ds_init_empty (&slice->label);
1204 var_append_value_name (var, &freq->value, &slice->label);
1205 slice->magnitude = freq->count;
1212 /* Allocate an array of slices and fill them from the data in frq_tab
1213 n_slices will contain the number of slices allocated.
1214 The caller is responsible for freeing slices
1216 static struct slice *
1217 freq_tab_to_slice_array(const struct frq_chart *pie,
1218 const struct freq_tab *frq_tab,
1219 const struct variable *var,
1222 struct slice *slices;
1226 slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1229 for (i = 0; i < frq_tab->n_valid; i++)
1230 n_slices += add_slice (pie, &frq_tab->valid[i], var, &slices[n_slices]);
1231 for (i = 0; i < frq_tab->n_missing; i++)
1232 n_slices += add_slice (pie, &frq_tab->missing[i], var, &slices[n_slices]);
1234 *n_slicesp = n_slices;
1242 do_piechart(const struct frq_chart *pie, const struct variable *var,
1243 const struct freq_tab *frq_tab)
1245 struct slice *slices;
1248 slices = freq_tab_to_slice_array (pie, frq_tab, var, &n_slices);
1251 msg (SW, _("Omitting pie chart for %s, which has only %d unique values."),
1252 var_get_name (var), n_slices);
1253 else if (n_slices > 50)
1254 msg (SW, _("Omitting pie chart for %s, which has over 50 unique values."),
1255 var_get_name (var));
1257 chart_item_submit (piechart_create (var_to_string(var), slices, n_slices));
1259 for (i = 0; i < n_slices; i++)
1260 ds_destroy (&slices[i].label);