1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2007, 2009, 2010, 2011 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/dataset.h"
27 #include "data/dictionary.h"
28 #include "data/format.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 "math/chart-geometry.h"
48 #include "output/chart-item.h"
49 #include "output/charts/piechart.h"
50 #include "output/charts/plot-hist.h"
51 #include "output/tab.h"
53 #include "gl/minmax.h"
54 #include "gl/xalloc.h"
57 #define _(msgid) gettext (msgid)
58 #define N_(msgid) msgid
65 +format=table:limit(n:limit,"%s>0")/notable/!table,
66 sort:!avalue/dvalue/afreq/dfreq;
67 missing=miss:include/!exclude;
68 barchart(ba_)=:minimum(d:min),
70 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0");
71 piechart(pie_)=:minimum(d:min),
73 missing:missing/!nomissing,
75 histogram(hi_)=:minimum(d:min),
77 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
78 norm:!nonormal/normal;
81 +percentiles = double list;
82 +statistics[st_]=mean,semean,median,mode,stddev,variance,
83 kurtosis,skewness,range,minimum,maximum,sum,
84 default,seskewness,sekurtosis,all,none.
92 FRQ_MEAN, FRQ_SEMEAN, FRQ_MEDIAN, FRQ_MODE, FRQ_STDDEV, FRQ_VARIANCE,
93 FRQ_KURT, FRQ_SEKURT, FRQ_SKEW, FRQ_SESKEW, FRQ_RANGE, FRQ_MIN, FRQ_MAX,
97 /* Description of a statistic. */
100 int st_indx; /* Index into a_statistics[]. */
101 const char *s10; /* Identifying string. */
104 /* Table of statistics, indexed by dsc_*. */
105 static const struct frq_info st_name[FRQ_N_STATS + 1] =
107 {FRQ_ST_MEAN, N_("Mean")},
108 {FRQ_ST_SEMEAN, N_("S.E. Mean")},
109 {FRQ_ST_MEDIAN, N_("Median")},
110 {FRQ_ST_MODE, N_("Mode")},
111 {FRQ_ST_STDDEV, N_("Std Dev")},
112 {FRQ_ST_VARIANCE, N_("Variance")},
113 {FRQ_ST_KURTOSIS, N_("Kurtosis")},
114 {FRQ_ST_SEKURTOSIS, N_("S.E. Kurt")},
115 {FRQ_ST_SKEWNESS, N_("Skewness")},
116 {FRQ_ST_SESKEWNESS, N_("S.E. Skew")},
117 {FRQ_ST_RANGE, N_("Range")},
118 {FRQ_ST_MINIMUM, N_("Minimum")},
119 {FRQ_ST_MAXIMUM, N_("Maximum")},
120 {FRQ_ST_SUM, N_("Sum")},
124 /* Percentiles to calculate. */
128 double p; /* the %ile to be calculated */
129 double value; /* the %ile's value */
130 bool show; /* True to show this percentile in the statistics box. */
133 /* Groups of statistics. */
135 #define FRQ_DEFAULT \
136 (BI (FRQ_MEAN) | BI (FRQ_STDDEV) | BI (FRQ_MIN) | BI (FRQ_MAX))
138 (BI (FRQ_SUM) | BI(FRQ_MIN) | BI(FRQ_MAX) \
139 | BI(FRQ_MEAN) | BI(FRQ_SEMEAN) | BI(FRQ_STDDEV) \
140 | BI(FRQ_VARIANCE) | BI(FRQ_KURT) | BI(FRQ_SEKURT) \
141 | BI(FRQ_SKEW) | BI(FRQ_SESKEW) | BI(FRQ_RANGE) \
142 | BI(FRQ_RANGE) | BI(FRQ_MODE) | BI(FRQ_MEDIAN))
146 double x_min; /* X axis minimum value. */
147 double x_max; /* X axis maximum value. */
148 int y_scale; /* Y axis scale: FRQ_FREQ or FRQ_PERCENT. */
150 /* Histograms only. */
151 double y_max; /* Y axis maximum value. */
152 bool draw_normal; /* Whether to draw normal curve. */
154 /* Pie charts only. */
155 bool include_missing; /* Whether to include missing values. */
158 /* Frequency tables. */
160 /* Entire frequency table. */
163 struct hmap data; /* Hash table for accumulating counts. */
164 struct freq *valid; /* Valid freqs. */
165 int n_valid; /* Number of total freqs. */
166 const struct dictionary *dict; /* Source of entries in the table. */
168 struct freq *missing; /* Missing freqs. */
169 int n_missing; /* Number of missing freqs. */
172 double total_cases; /* Sum of weights of all cases. */
173 double valid_cases; /* Sum of weights of valid cases. */
176 /* Per-variable frequency data. */
179 struct variable *var;
181 /* Freqency table. */
182 struct freq_tab tab; /* Frequencies table to use. */
185 int n_groups; /* Number of groups. */
186 double *groups; /* Groups. */
189 double stat[FRQ_N_STATS];
191 /* Variable attributes. */
199 struct var_freqs *vars;
202 /* Percentiles to calculate and possibly display. */
203 struct percentile *percentiles;
204 int n_percentiles, n_show_percentiles;
206 /* Frequency table display. */
207 int max_categories; /* Maximum categories to show. */
208 int sort; /* FRQ_AVALUE or FRQ_DVALUE
209 or FRQ_ACOUNT or FRQ_DCOUNT. */
211 /* Statistics; number of statistics. */
215 /* Histogram and pie chart settings. */
216 struct frq_chart *hist, *pie;
219 static void determine_charts (struct frq_proc *,
220 const struct cmd_frequencies *);
222 static void calc_stats (const struct var_freqs *, double d[FRQ_N_STATS]);
223 static void calc_percentiles (const struct frq_proc *,
224 const struct var_freqs *);
226 static void precalc (struct frq_proc *, struct casereader *, struct dataset *);
227 static void calc (struct frq_proc *, const struct ccase *,
228 const struct dataset *);
229 static void postcalc (struct frq_proc *, const struct dataset *);
231 static void postprocess_freq_tab (const struct frq_proc *, struct var_freqs *);
232 static void dump_freq_table (const struct var_freqs *,
233 const struct variable *weight_var);
234 static void dump_statistics (const struct frq_proc *, const struct var_freqs *,
235 const struct variable *weight_var);
236 static void cleanup_freq_tab (struct var_freqs *);
238 static void add_percentile (struct frq_proc *, double x, bool show,
239 size_t *allocated_percentiles);
241 static void do_piechart(const struct frq_chart *, const struct variable *,
242 const struct freq_tab *);
244 struct histogram *freq_tab_to_hist(const struct frq_proc *,
245 const struct freq_tab *,
246 const struct variable *);
248 /* Parser and outline. */
251 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
253 struct cmd_frequencies cmd;
255 struct casegrouper *grouper;
256 struct casereader *input, *group;
257 size_t allocated_percentiles;
261 frq.pool = pool_create ();
266 frq.percentiles = NULL;
267 frq.n_percentiles = 0;
268 frq.n_show_percentiles = 0;
273 allocated_percentiles = 0;
275 if (!parse_frequencies (lexer, ds, &cmd, &frq))
277 pool_destroy (frq.pool);
281 /* Figure out when to show frequency tables. */
282 frq.max_categories = (cmd.table == FRQ_NOTABLE ? -1
283 : cmd.table == FRQ_TABLE ? INT_MAX
287 /* Figure out statistics to calculate. */
289 if (cmd.a_statistics[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
290 frq.stats |= FRQ_DEFAULT;
291 if (cmd.a_statistics[FRQ_ST_ALL])
292 frq.stats |= FRQ_ALL;
293 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
294 frq.stats &= ~BIT_INDEX (FRQ_MEDIAN);
295 for (i = 0; i < FRQ_N_STATS; i++)
296 if (cmd.a_statistics[st_name[i].st_indx])
297 frq.stats |= BIT_INDEX (i);
298 if (frq.stats & FRQ_KURT)
299 frq.stats |= BIT_INDEX (FRQ_SEKURT);
300 if (frq.stats & FRQ_SKEW)
301 frq.stats |= BIT_INDEX (FRQ_SESKEW);
303 /* Calculate n_stats. */
305 for (i = 0; i < FRQ_N_STATS; i++)
306 if ((frq.stats & BIT_INDEX (i)))
310 determine_charts (&frq, &cmd);
311 if (cmd.sbc_histogram || cmd.sbc_piechart || cmd.sbc_ntiles)
312 cmd.sort = FRQ_AVALUE;
314 /* Work out what percentiles need to be calculated */
315 if ( cmd.sbc_percentiles )
317 for ( i = 0 ; i < MAXLISTS ; ++i )
320 subc_list_double *ptl_list = &cmd.dl_percentiles[i];
321 for ( pl = 0 ; pl < subc_list_double_count(ptl_list); ++pl)
322 add_percentile (&frq, subc_list_double_at(ptl_list, pl) / 100.0,
323 true, &allocated_percentiles);
326 if ( cmd.sbc_ntiles )
328 for ( i = 0 ; i < cmd.sbc_ntiles ; ++i )
331 for (j = 0; j <= cmd.n_ntiles[i]; ++j )
332 add_percentile (&frq, j / (double) cmd.n_ntiles[i], true,
333 &allocated_percentiles);
336 if (frq.stats & BIT_INDEX (FRQ_MEDIAN))
338 /* Treat the median as the 50% percentile.
339 We output it in the percentiles table as "50 (Median)." */
340 add_percentile (&frq, 0.5, true, &allocated_percentiles);
341 frq.stats &= ~BIT_INDEX (FRQ_MEDIAN);
344 if (cmd.sbc_histogram)
346 add_percentile (&frq, 0.25, false, &allocated_percentiles);
347 add_percentile (&frq, 0.75, false, &allocated_percentiles);
351 input = casereader_create_filter_weight (proc_open (ds), dataset_dict (ds),
353 grouper = casegrouper_create_splits (input, dataset_dict (ds));
354 for (; casegrouper_get_next_group (grouper, &group);
355 casereader_destroy (group))
359 precalc (&frq, group, ds);
360 for (; (c = casereader_read (group)) != NULL; case_unref (c))
364 ok = casegrouper_destroy (grouper);
365 ok = proc_commit (ds) && ok;
367 free_frequencies(&cmd);
369 pool_destroy (frq.pool);
371 free (frq.percentiles);
375 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
378 /* Figure out which charts the user requested. */
380 determine_charts (struct frq_proc *frq, const struct cmd_frequencies *cmd)
382 if (cmd->sbc_barchart)
383 msg (SW, _("Bar charts are not implemented."));
385 if (cmd->sbc_histogram)
387 struct frq_chart *hist;
389 hist = frq->hist = xmalloc (sizeof *frq->hist);
390 hist->x_min = cmd->hi_min;
391 hist->x_max = cmd->hi_max;
392 hist->y_scale = cmd->hi_scale;
393 hist->y_max = cmd->hi_scale == FRQ_FREQ ? cmd->hi_freq : cmd->hi_pcnt;
394 hist->draw_normal = cmd->hi_norm != FRQ_NONORMAL;
395 hist->include_missing = false;
397 if (hist->x_min != SYSMIS && hist->x_max != SYSMIS
398 && hist->x_min >= hist->x_max)
400 msg (SE, _("%s for histogram must be greater than or equal to %s, "
401 "but %s was specified as %.15g and %s as %.15g. "
402 "%s and %s will be ignored."),
407 hist->x_min = hist->x_max = SYSMIS;
411 if (cmd->sbc_piechart)
413 struct frq_chart *pie;
415 pie = frq->pie = xmalloc (sizeof *frq->pie);
416 pie->x_min = cmd->pie_min;
417 pie->x_max = cmd->pie_max;
418 pie->y_scale = cmd->pie_scale;
419 pie->include_missing = cmd->pie_missing == FRQ_MISSING;
421 if (pie->x_min != SYSMIS && pie->x_max != SYSMIS
422 && pie->x_min >= pie->x_max)
424 msg (SE, _("%s for pie chart must be greater than or equal to %s, "
425 "but %s was specified as %.15g and %s as %.15g. "
426 "%s and %s will be ignored."),
431 pie->x_min = pie->x_max = SYSMIS;
436 /* Add data from case C to the frequency table. */
438 calc (struct frq_proc *frq, const struct ccase *c, const struct dataset *ds)
440 double weight = dict_get_case_weight (dataset_dict (ds), c, NULL);
443 for (i = 0; i < frq->n_vars; i++)
445 struct var_freqs *vf = &frq->vars[i];
446 const union value *value = case_data (c, vf->var);
447 size_t hash = value_hash (value, vf->width, 0);
450 f = freq_hmap_search (&vf->tab.data, value, vf->width, hash);
452 f = freq_hmap_insert (&vf->tab.data, value, vf->width, hash);
458 /* Prepares each variable that is the target of FREQUENCIES by setting
459 up its hash table. */
461 precalc (struct frq_proc *frq, struct casereader *input, struct dataset *ds)
466 c = casereader_peek (input, 0);
469 output_split_file_values (ds, c);
473 for (i = 0; i < frq->n_vars; i++)
474 hmap_init (&frq->vars[i].tab.data);
477 /* Finishes up with the variables after frequencies have been
478 calculated. Displays statistics, percentiles, ... */
480 postcalc (struct frq_proc *frq, const struct dataset *ds)
482 const struct dictionary *dict = dataset_dict (ds);
483 const struct variable *wv = dict_get_weight (dict);
486 for (i = 0; i < frq->n_vars; i++)
488 struct var_freqs *vf = &frq->vars[i];
490 postprocess_freq_tab (frq, vf);
492 /* Frequencies tables. */
493 if (vf->tab.n_valid + vf->tab.n_missing <= frq->max_categories)
494 dump_freq_table (vf, wv);
498 dump_statistics (frq, vf, wv);
500 if (frq->hist && var_is_numeric (vf->var) && vf->tab.n_valid > 0)
502 double d[FRQ_N_STATS];
503 struct histogram *histogram;
507 histogram = freq_tab_to_hist (frq, &vf->tab, vf->var);
511 chart_item_submit (histogram_chart_create (
512 histogram->gsl_hist, var_to_string(vf->var),
516 frq->hist->draw_normal));
518 statistic_destroy (&histogram->parent);
523 do_piechart(frq->pie, vf->var, &vf->tab);
525 cleanup_freq_tab (vf);
530 /* Returns true iff the value in struct freq F is non-missing
533 not_missing (const void *f_, const void *v_)
535 const struct freq *f = f_;
536 const struct variable *v = v_;
538 return !var_is_value_missing (v, &f->value, MV_ANY);
541 struct freq_compare_aux
547 bool ascending_value;
551 compare_freq (const void *a_, const void *b_, const void *aux_)
553 const struct freq_compare_aux *aux = aux_;
554 const struct freq *a = a_;
555 const struct freq *b = b_;
557 if (aux->by_freq && a->count != b->count)
559 int cmp = a->count > b->count ? 1 : -1;
560 return aux->ascending_freq ? cmp : -cmp;
564 int cmp = value_compare_3way (&a->value, &b->value, aux->width);
565 return aux->ascending_value ? cmp : -cmp;
568 /* Summarizes the frequency table data for variable V. */
570 postprocess_freq_tab (const struct frq_proc *frq, struct var_freqs *vf)
572 struct freq_tab *ft = &vf->tab;
573 struct freq_compare_aux aux;
575 struct freq *freqs, *f;
578 /* Extract data from hash table. */
579 count = hmap_count (&ft->data);
580 freqs = freq_hmap_extract (&ft->data);
582 /* Put data into ft. */
584 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, vf->var);
585 ft->missing = freqs + ft->n_valid;
586 ft->n_missing = count - ft->n_valid;
589 aux.by_freq = frq->sort == FRQ_AFREQ || frq->sort == FRQ_DFREQ;
590 aux.ascending_freq = frq->sort != FRQ_DFREQ;
591 aux.width = vf->width;
592 aux.ascending_value = frq->sort != FRQ_DVALUE;
593 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare_freq, &aux);
594 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare_freq, &aux);
596 /* Summary statistics. */
597 ft->valid_cases = 0.0;
598 for(i = 0 ; i < ft->n_valid ; ++i )
601 ft->valid_cases += f->count;
605 ft->total_cases = ft->valid_cases ;
606 for(i = 0 ; i < ft->n_missing ; ++i )
609 ft->total_cases += f->count;
614 /* Frees the frequency table for variable V. */
616 cleanup_freq_tab (struct var_freqs *vf)
618 free (vf->tab.valid);
619 freq_hmap_destroy (&vf->tab.data, vf->width);
622 /* Parses the VARIABLES subcommand. */
624 frq_custom_variables (struct lexer *lexer, struct dataset *ds,
625 struct cmd_frequencies *cmd UNUSED, void *frq_ UNUSED)
627 struct frq_proc *frq = frq_;
628 struct variable **vars;
632 lex_match (lexer, T_EQUALS);
633 if (lex_token (lexer) != T_ALL
634 && (lex_token (lexer) != T_ID
635 || dict_lookup_var (dataset_dict (ds), lex_tokcstr (lexer)) == NULL))
638 /* Get list of current variables, to avoid duplicates. */
639 vars = xmalloc (frq->n_vars * sizeof *vars);
640 n_vars = frq->n_vars;
641 for (i = 0; i < frq->n_vars; i++)
642 vars[i] = frq->vars[i].var;
644 if (!parse_variables (lexer, dataset_dict (ds), &vars, &n_vars,
645 PV_APPEND | PV_NO_SCRATCH))
648 frq->vars = xrealloc (frq->vars, n_vars * sizeof *frq->vars);
649 for (i = frq->n_vars; i < n_vars; i++)
651 struct variable *var = vars[i];
652 struct var_freqs *vf = &frq->vars[i];
655 vf->tab.valid = vf->tab.missing = NULL;
656 vf->tab.dict = dataset_dict (ds);
659 vf->width = var_get_width (var);
661 frq->n_vars = n_vars;
668 /* Parses the GROUPED subcommand, setting the n_grouped, grouped
669 fields of specified variables. */
671 frq_custom_grouped (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *frq_ UNUSED)
673 struct frq_proc *frq = frq_;
675 lex_match (lexer, T_EQUALS);
676 if ((lex_token (lexer) == T_ID
677 && dict_lookup_var (dataset_dict (ds), lex_tokcstr (lexer)) != NULL)
678 || lex_token (lexer) == T_ID)
683 /* Max, current size of list; list itself. */
689 const struct variable **v;
691 if (!parse_variables_const (lexer, dataset_dict (ds), &v, &n,
692 PV_NO_DUPLICATE | PV_NUMERIC))
694 if (lex_match (lexer, T_LPAREN))
698 while (lex_integer (lexer))
703 dl = pool_nrealloc (frq->pool, dl, ml, sizeof *dl);
705 dl[nl++] = lex_tokval (lexer);
707 lex_match (lexer, T_COMMA);
709 /* Note that nl might still be 0 and dl might still be
710 NULL. That's okay. */
711 if (!lex_match (lexer, T_RPAREN))
714 lex_error_expecting (lexer, "`)'", NULL_SENTINEL);
724 for (i = 0; i < n; i++)
728 for (j = 0; j < frq->n_vars; j++)
730 struct var_freqs *vf = &frq->vars[j];
733 if (vf->groups != NULL)
734 msg (SE, _("Variables %s specified multiple times on "
735 "%s subcommand."), var_get_name (v[i]), "GROUPED");
744 msg (SE, _("Variables %s specified on %s but not on "
745 "%s."), var_get_name (v[i]), "GROUPED", "VARIABLES");
751 if (lex_token (lexer) != T_SLASH)
754 if ((lex_next_token (lexer, 1) == T_ID
755 && dict_lookup_var (dataset_dict (ds),
756 lex_next_tokcstr (lexer, 1)))
757 || lex_next_token (lexer, 1) == T_ALL)
759 /* The token after the slash is a variable name. Keep parsing. */
764 /* The token after the slash must be the start of a new
765 subcommand. Let the caller see the slash. */
773 /* Adds X to the list of percentiles, keeping the list in proper
774 order. If SHOW is true, the percentile will be shown in the statistics
775 box, otherwise it will be hidden. */
777 add_percentile (struct frq_proc *frq, double x, bool show,
778 size_t *allocated_percentiles)
782 /* Do nothing if it's already in the list */
783 for (i = 0; i < frq->n_percentiles; i++)
785 struct percentile *pc = &frq->percentiles[i];
787 if ( fabs(x - pc->p) < DBL_EPSILON )
789 if (show && !pc->show)
791 frq->n_show_percentiles++;
801 if (frq->n_percentiles >= *allocated_percentiles)
802 frq->percentiles = x2nrealloc (frq->percentiles, allocated_percentiles,
803 sizeof *frq->percentiles);
804 insert_element (frq->percentiles, frq->n_percentiles,
805 sizeof *frq->percentiles, i);
806 frq->percentiles[i].p = x;
807 frq->percentiles[i].show = show;
808 frq->n_percentiles++;
810 frq->n_show_percentiles++;
813 /* Comparison functions. */
816 /* Frequency table display. */
818 /* Displays a full frequency table for variable V. */
820 dump_freq_table (const struct var_freqs *vf, const struct variable *wv)
822 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
823 const struct freq_tab *ft = &vf->tab;
828 double cum_total = 0.0;
829 double cum_freq = 0.0;
831 static const char *headings[] = {
840 n_categories = ft->n_valid + ft->n_missing;
841 t = tab_create (6, n_categories + 2);
842 tab_headers (t, 0, 0, 1, 0);
844 for (x = 0; x < 6; x++)
845 tab_text (t, x, 0, TAB_CENTER | TAT_TITLE, gettext (headings[x]));
848 for (f = ft->valid; f < ft->missing; f++)
851 double percent, valid_percent;
853 cum_freq += f->count;
855 percent = f->count / ft->total_cases * 100.0;
856 valid_percent = f->count / ft->valid_cases * 100.0;
857 cum_total += valid_percent;
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, vf->var, NULL);
864 tab_double (t, 2, r, TAB_NONE, f->count, wfmt);
865 tab_double (t, 3, r, TAB_NONE, percent, NULL);
866 tab_double (t, 4, r, TAB_NONE, valid_percent, NULL);
867 tab_double (t, 5, r, TAB_NONE, cum_total, NULL);
870 for (; f < &ft->valid[n_categories]; f++)
874 cum_freq += f->count;
876 label = var_lookup_value_label (vf->var, &f->value);
878 tab_text (t, 0, r, TAB_LEFT, label);
880 tab_value (t, 1, r, TAB_NONE, &f->value, vf->var, NULL);
881 tab_double (t, 2, r, TAB_NONE, f->count, wfmt);
882 tab_double (t, 3, r, TAB_NONE,
883 f->count / ft->total_cases * 100.0, NULL);
884 tab_text (t, 4, r, TAB_NONE, _("Missing"));
888 tab_box (t, TAL_1, TAL_1, -1, TAL_1, 0, 0, 5, r);
889 tab_hline (t, TAL_2, 0, 5, 1);
890 tab_hline (t, TAL_2, 0, 5, r);
891 tab_joint_text (t, 0, r, 1, r, TAB_RIGHT | TAT_TITLE, _("Total"));
892 tab_vline (t, TAL_0, 1, r, r);
893 tab_double (t, 2, r, TAB_NONE, cum_freq, wfmt);
894 tab_fixed (t, 3, r, TAB_NONE, 100.0, 5, 1);
895 tab_fixed (t, 4, r, TAB_NONE, 100.0, 5, 1);
897 tab_title (t, "%s", var_to_string (vf->var));
901 /* Statistical display. */
904 calc_percentile (double p, double valid_cases, double x1, double x2)
908 s = (settings_get_algorithm () != COMPATIBLE
909 ? modf ((valid_cases - 1) * p, &dummy)
910 : modf ((valid_cases + 1) * p - 1, &dummy));
912 return x1 + (x2 - x1) * s;
915 /* Calculates all of the percentiles for VF within FRQ. */
917 calc_percentiles (const struct frq_proc *frq, const struct var_freqs *vf)
919 const struct freq_tab *ft = &vf->tab;
920 double W = ft->valid_cases;
921 const struct freq *f;
925 assert (ft->n_valid > 0);
929 for (f = ft->valid; f < ft->missing; f++)
932 for (; percentile_idx < frq->n_percentiles; percentile_idx++)
934 struct percentile *pc = &frq->percentiles[percentile_idx];
937 tp = (settings_get_algorithm () == ENHANCED
939 : (W + 1) * pc->p - 1);
944 if (tp + 1 < rank || f + 1 >= ft->missing)
945 pc->value = f->value.f;
947 pc->value = calc_percentile (pc->p, W, f->value.f, f[1].value.f);
950 for (; percentile_idx < frq->n_percentiles; percentile_idx++)
952 struct percentile *pc = &frq->percentiles[percentile_idx];
953 pc->value = ft->valid[ft->n_valid - 1].value.f;
957 /* Calculates all the pertinent statistics for VF, putting them in array
960 calc_stats (const struct var_freqs *vf, double d[FRQ_N_STATS])
962 const struct freq_tab *ft = &vf->tab;
963 double W = ft->valid_cases;
964 const struct freq *f;
969 assert (ft->n_valid > 0);
971 /* Calculate the mode. */
974 for (f = ft->valid; f < ft->missing; f++)
976 if (most_often < f->count)
978 most_often = f->count;
981 else if (most_often == f->count)
983 /* A duplicate mode is undefined.
984 FIXME: keep track of *all* the modes. */
989 /* Calculate moments. */
990 m = moments_create (MOMENT_KURTOSIS);
991 for (f = ft->valid; f < ft->missing; f++)
992 moments_pass_one (m, f->value.f, f->count);
993 for (f = ft->valid; f < ft->missing; f++)
994 moments_pass_two (m, f->value.f, f->count);
995 moments_calculate (m, NULL, &d[FRQ_MEAN], &d[FRQ_VARIANCE],
996 &d[FRQ_SKEW], &d[FRQ_KURT]);
999 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1000 d[FRQ_MIN] = ft->valid[0].value.f;
1001 d[FRQ_MAX] = ft->valid[ft->n_valid - 1].value.f;
1002 d[FRQ_MODE] = X_mode;
1003 d[FRQ_RANGE] = d[FRQ_MAX] - d[FRQ_MIN];
1004 d[FRQ_SUM] = d[FRQ_MEAN] * W;
1005 d[FRQ_STDDEV] = sqrt (d[FRQ_VARIANCE]);
1006 d[FRQ_SEMEAN] = d[FRQ_STDDEV] / sqrt (W);
1007 d[FRQ_SESKEW] = calc_seskew (W);
1008 d[FRQ_SEKURT] = calc_sekurt (W);
1011 /* Displays a table of all the statistics requested for variable V. */
1013 dump_statistics (const struct frq_proc *frq, const struct var_freqs *vf,
1014 const struct variable *wv)
1016 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1017 const struct freq_tab *ft = &vf->tab;
1018 double stat_value[FRQ_N_STATS];
1019 struct tab_table *t;
1022 if (var_is_alpha (vf->var))
1025 if (ft->n_valid == 0)
1027 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1028 var_get_name (vf->var));
1031 calc_stats (vf, stat_value);
1032 calc_percentiles (frq, vf);
1034 t = tab_create (3, frq->n_stats + frq->n_show_percentiles + 2);
1036 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1039 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1040 tab_vline (t, TAL_GAP , 1, 0, tab_nr(t) - 1 ) ;
1042 r=2; /* N missing and N valid are always dumped */
1044 for (i = 0; i < FRQ_N_STATS; i++)
1045 if (frq->stats & BIT_INDEX (i))
1047 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1048 gettext (st_name[i].s10));
1049 tab_double (t, 2, r, TAB_NONE, stat_value[i], NULL);
1053 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1054 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1055 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1057 tab_double (t, 2, 0, TAB_NONE, ft->valid_cases, wfmt);
1058 tab_double (t, 2, 1, TAB_NONE, ft->total_cases - ft->valid_cases, wfmt);
1060 for (i = 0; i < frq->n_percentiles; i++)
1062 struct percentile *pc = &frq->percentiles[i];
1069 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1073 tab_text (t, 1, r, TAB_LEFT, _("50 (Median)"));
1075 tab_fixed (t, 1, r, TAB_LEFT, pc->p * 100, 3, 0);
1076 tab_double (t, 2, r, TAB_NONE, pc->value,
1077 var_get_print_format (vf->var));
1081 tab_title (t, "%s", var_to_string (vf->var));
1087 calculate_iqr (const struct frq_proc *frq)
1093 for (i = 0; i < frq->n_percentiles; i++)
1095 struct percentile *pc = &frq->percentiles[i];
1097 if (fabs (0.25 - pc->p) < DBL_EPSILON)
1099 else if (fabs (0.75 - pc->p) < DBL_EPSILON)
1103 return q1 == SYSMIS || q3 == SYSMIS ? SYSMIS : q3 - q1;
1107 chart_includes_value (const struct frq_chart *chart,
1108 const struct variable *var,
1109 const union value *value)
1111 if (!chart->include_missing && var_is_value_missing (var, value, MV_ANY))
1114 if (var_is_numeric (var)
1115 && ((chart->x_min != SYSMIS && value->f < chart->x_min)
1116 || (chart->x_max != SYSMIS && value->f > chart->x_max)))
1122 /* Create a gsl_histogram from a freq_tab */
1124 freq_tab_to_hist (const struct frq_proc *frq, const struct freq_tab *ft,
1125 const struct variable *var)
1127 double x_min, x_max, valid_freq;
1130 struct histogram *histogram;
1133 /* Find out the extremes of the x value, within the range to be included in
1134 the histogram, and sum the total frequency of those values. */
1138 for (i = 0; i < ft->n_valid; i++)
1140 const struct freq *f = &ft->valid[i];
1141 if (chart_includes_value (frq->hist, var, &f->value))
1143 x_min = MIN (x_min, f->value.f);
1144 x_max = MAX (x_max, f->value.f);
1145 valid_freq += f->count;
1149 /* Freedman-Diaconis' choice of bin width. */
1150 iqr = calculate_iqr (frq);
1151 bin_width = 2 * iqr / pow (valid_freq, 1.0 / 3.0);
1153 histogram = histogram_create (bin_width, x_min, x_max);
1155 if ( histogram == NULL)
1158 for (i = 0; i < ft->n_valid; i++)
1160 const struct freq *f = &ft->valid[i];
1161 if (chart_includes_value (frq->hist, var, &f->value))
1162 histogram_add (histogram, f->value.f, f->count);
1169 add_slice (const struct frq_chart *pie, const struct freq *freq,
1170 const struct variable *var, struct slice *slice)
1172 if (chart_includes_value (pie, var, &freq->value))
1174 ds_init_empty (&slice->label);
1175 var_append_value_name (var, &freq->value, &slice->label);
1176 slice->magnitude = freq->count;
1183 /* Allocate an array of slices and fill them from the data in frq_tab
1184 n_slices will contain the number of slices allocated.
1185 The caller is responsible for freeing slices
1187 static struct slice *
1188 freq_tab_to_slice_array(const struct frq_chart *pie,
1189 const struct freq_tab *frq_tab,
1190 const struct variable *var,
1193 struct slice *slices;
1197 slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1200 for (i = 0; i < frq_tab->n_valid; i++)
1201 n_slices += add_slice (pie, &frq_tab->valid[i], var, &slices[n_slices]);
1202 for (i = 0; i < frq_tab->n_missing; i++)
1203 n_slices += add_slice (pie, &frq_tab->missing[i], var, &slices[n_slices]);
1205 *n_slicesp = n_slices;
1213 do_piechart(const struct frq_chart *pie, const struct variable *var,
1214 const struct freq_tab *frq_tab)
1216 struct slice *slices;
1219 slices = freq_tab_to_slice_array (pie, frq_tab, var, &n_slices);
1222 msg (SW, _("Omitting pie chart for %s, which has only %d unique values."),
1223 var_get_name (var), n_slices);
1224 else if (n_slices > 50)
1225 msg (SW, _("Omitting pie chart for %s, which has over 50 unique values."),
1226 var_get_name (var));
1228 chart_item_submit (piechart_create (var_to_string(var), slices, n_slices));
1230 for (i = 0; i < n_slices; i++)
1231 ds_destroy (&slices[i].label);