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 /* Parsed command. */
161 static struct cmd_frequencies cmd;
163 /* Frequency tables. */
165 /* Entire frequency table. */
168 struct hmap data; /* Hash table for accumulating counts. */
169 struct freq *valid; /* Valid freqs. */
170 int n_valid; /* Number of total freqs. */
171 const struct dictionary *dict; /* Source of entries in the table. */
173 struct freq *missing; /* Missing freqs. */
174 int n_missing; /* Number of missing freqs. */
177 double total_cases; /* Sum of weights of all cases. */
178 double valid_cases; /* Sum of weights of valid cases. */
181 /* Per-variable frequency data. */
184 struct variable *var;
186 /* Freqency table. */
187 struct freq_tab tab; /* Frequencies table to use. */
190 int n_groups; /* Number of groups. */
191 double *groups; /* Groups. */
194 double stat[FRQ_N_STATS];
196 /* Variable attributes. */
198 struct fmt_spec print;
205 struct var_freqs *vars;
208 struct percentile *percentiles;
209 int n_percentiles, n_show_percentiles;
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 *frq);
221 static void calc_stats (const struct frq_proc *, const struct var_freqs *,
222 double d[FRQ_N_STATS]);
224 static void precalc (struct frq_proc *, struct casereader *, struct dataset *);
225 static void calc (struct frq_proc *, const struct ccase *,
226 const struct dataset *);
227 static void postcalc (struct frq_proc *, const struct dataset *);
229 static void postprocess_freq_tab (struct var_freqs *);
230 static void dump_freq_table (const struct var_freqs *,
231 const struct variable *weight_var);
232 static void dump_statistics (const struct frq_proc *, const struct var_freqs *,
233 const struct variable *weight_var);
234 static void cleanup_freq_tab (struct var_freqs *);
236 static algo_compare_func compare_value_numeric_a, compare_value_alpha_a;
237 static algo_compare_func compare_value_numeric_d, compare_value_alpha_d;
238 static algo_compare_func compare_freq_numeric_a, compare_freq_alpha_a;
239 static algo_compare_func compare_freq_numeric_d, compare_freq_alpha_d;
241 static void add_percentile (struct frq_proc *, double x, bool show,
242 size_t *allocated_percentiles);
244 static void do_piechart(const struct frq_chart *, const struct variable *,
245 const struct freq_tab *);
247 struct histogram *freq_tab_to_hist(const struct frq_proc *,
248 const struct freq_tab *,
249 const struct variable *);
251 /* Parser and outline. */
254 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
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 statistics to calculate. */
285 if (cmd.a_statistics[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
286 frq.stats |= FRQ_DEFAULT;
287 if (cmd.a_statistics[FRQ_ST_ALL])
288 frq.stats |= FRQ_ALL;
289 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
290 frq.stats &= ~BIT_INDEX (FRQ_MEDIAN);
291 for (i = 0; i < FRQ_N_STATS; i++)
292 if (cmd.a_statistics[st_name[i].st_indx])
293 frq.stats |= BIT_INDEX (i);
294 if (frq.stats & FRQ_KURT)
295 frq.stats |= BIT_INDEX (FRQ_SEKURT);
296 if (frq.stats & FRQ_SKEW)
297 frq.stats |= BIT_INDEX (FRQ_SESKEW);
299 /* Calculate n_stats. */
301 for (i = 0; i < FRQ_N_STATS; i++)
302 if ((frq.stats & BIT_INDEX (i)))
306 determine_charts (&frq);
307 if (cmd.sbc_histogram || cmd.sbc_piechart || cmd.sbc_ntiles)
308 cmd.sort = FRQ_AVALUE;
310 /* Work out what percentiles need to be calculated */
311 if ( cmd.sbc_percentiles )
313 for ( i = 0 ; i < MAXLISTS ; ++i )
316 subc_list_double *ptl_list = &cmd.dl_percentiles[i];
317 for ( pl = 0 ; pl < subc_list_double_count(ptl_list); ++pl)
318 add_percentile (&frq, subc_list_double_at(ptl_list, pl) / 100.0,
319 true, &allocated_percentiles);
322 if ( cmd.sbc_ntiles )
324 for ( i = 0 ; i < cmd.sbc_ntiles ; ++i )
327 for (j = 0; j <= cmd.n_ntiles[i]; ++j )
328 add_percentile (&frq, j / (double) cmd.n_ntiles[i], true,
329 &allocated_percentiles);
332 if (frq.stats & BIT_INDEX (FRQ_MEDIAN))
334 /* Treat the median as the 50% percentile.
335 We output it in the percentiles table as "50 (Median)." */
336 add_percentile (&frq, 0.5, true, &allocated_percentiles);
337 frq.stats &= ~BIT_INDEX (FRQ_MEDIAN);
340 if (cmd.sbc_histogram)
342 add_percentile (&frq, 0.25, false, &allocated_percentiles);
343 add_percentile (&frq, 0.75, false, &allocated_percentiles);
347 input = casereader_create_filter_weight (proc_open (ds), dataset_dict (ds),
349 grouper = casegrouper_create_splits (input, dataset_dict (ds));
350 for (; casegrouper_get_next_group (grouper, &group);
351 casereader_destroy (group))
355 precalc (&frq, group, ds);
356 for (; (c = casereader_read (group)) != NULL; case_unref (c))
360 ok = casegrouper_destroy (grouper);
361 ok = proc_commit (ds) && ok;
363 free_frequencies(&cmd);
365 pool_destroy (frq.pool);
367 free (frq.percentiles);
371 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
374 /* Figure out which charts the user requested. */
376 determine_charts (struct frq_proc *frq)
378 if (cmd.sbc_barchart)
379 msg (SW, _("Bar charts are not implemented."));
381 if (cmd.sbc_histogram)
383 struct frq_chart *hist;
385 hist = frq->hist = xmalloc (sizeof *frq->hist);
386 hist->x_min = cmd.hi_min;
387 hist->x_max = cmd.hi_max;
388 hist->y_scale = cmd.hi_scale;
389 hist->y_max = cmd.hi_scale == FRQ_FREQ ? cmd.hi_freq : cmd.hi_pcnt;
390 hist->draw_normal = cmd.hi_norm != FRQ_NONORMAL;
391 hist->include_missing = false;
393 if (hist->x_min != SYSMIS && hist->x_max != SYSMIS
394 && hist->x_min >= hist->x_max)
396 msg (SE, _("MAX for histogram must be greater than or equal to MIN, "
397 "but MIN was specified as %.15g and MAX as %.15g. "
398 "MIN and MAX will be ignored."),
399 hist->x_min, hist->x_max);
400 hist->x_min = hist->x_max = SYSMIS;
404 if (cmd.sbc_piechart)
406 struct frq_chart *pie;
408 pie = frq->pie = xmalloc (sizeof *frq->pie);
409 pie->x_min = cmd.pie_min;
410 pie->x_max = cmd.pie_max;
411 pie->y_scale = cmd.pie_scale;
412 pie->include_missing = cmd.pie_missing == FRQ_MISSING;
414 if (pie->x_min != SYSMIS && pie->x_max != SYSMIS
415 && pie->x_min >= pie->x_max)
417 msg (SE, _("MAX for pie chart must be greater than or equal to MIN, "
418 "but MIN was specified as %.15g and MAX as %.15g. "
419 "MIN and MAX will be ignored."), pie->x_min, pie->x_max);
420 pie->x_min = pie->x_max = SYSMIS;
425 /* Add data from case C to the frequency table. */
427 calc (struct frq_proc *frq, const struct ccase *c, const struct dataset *ds)
429 double weight = dict_get_case_weight (dataset_dict (ds), c, NULL);
432 for (i = 0; i < frq->n_vars; i++)
434 struct var_freqs *vf = &frq->vars[i];
435 const union value *value = case_data (c, vf->var);
436 size_t hash = value_hash (value, vf->width, 0);
439 f = freq_hmap_search (&vf->tab.data, value, vf->width, hash);
441 f = freq_hmap_insert (&vf->tab.data, value, vf->width, hash);
447 /* Prepares each variable that is the target of FREQUENCIES by setting
448 up its hash table. */
450 precalc (struct frq_proc *frq, struct casereader *input, struct dataset *ds)
455 c = casereader_peek (input, 0);
458 output_split_file_values (ds, c);
462 for (i = 0; i < frq->n_vars; i++)
463 hmap_init (&frq->vars[i].tab.data);
466 /* Finishes up with the variables after frequencies have been
467 calculated. Displays statistics, percentiles, ... */
469 postcalc (struct frq_proc *frq, const struct dataset *ds)
471 const struct dictionary *dict = dataset_dict (ds);
472 const struct variable *wv = dict_get_weight (dict);
475 for (i = 0; i < frq->n_vars; i++)
477 struct var_freqs *vf = &frq->vars[i];
480 postprocess_freq_tab (vf);
482 /* Frequencies tables. */
483 n_categories = vf->tab.n_valid + vf->tab.n_missing;
484 if (cmd.table == FRQ_TABLE
485 || (cmd.table == FRQ_LIMIT && n_categories <= cmd.limit))
486 dump_freq_table (vf, wv);
490 dump_statistics (frq, vf, wv);
492 if (frq->hist && var_is_numeric (vf->var) && vf->tab.n_valid > 0)
494 double d[FRQ_N_STATS];
495 struct histogram *histogram;
497 calc_stats (frq, vf, d);
499 histogram = freq_tab_to_hist (frq, &vf->tab, vf->var);
501 chart_item_submit (histogram_chart_create (
502 histogram->gsl_hist, var_to_string(vf->var),
506 frq->hist->draw_normal));
508 statistic_destroy (&histogram->parent);
512 do_piechart(frq->pie, vf->var, &vf->tab);
514 cleanup_freq_tab (vf);
519 /* Returns the comparison function that should be used for
520 sorting a frequency table by FRQ_SORT using VAL_TYPE
522 static algo_compare_func *
523 get_freq_comparator (int frq_sort, enum val_type val_type)
525 bool is_numeric = val_type == VAL_NUMERIC;
529 return is_numeric ? compare_value_numeric_a : compare_value_alpha_a;
531 return is_numeric ? compare_value_numeric_d : compare_value_alpha_d;
533 return is_numeric ? compare_freq_numeric_a : compare_freq_alpha_a;
535 return is_numeric ? compare_freq_numeric_d : compare_freq_alpha_d;
541 /* Returns true iff the value in struct freq F is non-missing
544 not_missing (const void *f_, const void *v_)
546 const struct freq *f = f_;
547 const struct variable *v = v_;
549 return !var_is_value_missing (v, &f->value, MV_ANY);
552 /* Summarizes the frequency table data for variable V. */
554 postprocess_freq_tab (struct var_freqs *vf)
556 struct freq_tab *ft = &vf->tab;
557 algo_compare_func *compare;
559 struct freq *freqs, *f;
562 /* Extract data from hash table. */
563 count = hmap_count (&ft->data);
564 freqs = freq_hmap_extract (&ft->data);
566 /* Put data into ft. */
568 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, vf->var);
569 ft->missing = freqs + ft->n_valid;
570 ft->n_missing = count - ft->n_valid;
573 compare = get_freq_comparator (cmd.sort, var_get_type (vf->var));
574 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare, vf);
575 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, vf);
577 /* Summary statistics. */
578 ft->valid_cases = 0.0;
579 for(i = 0 ; i < ft->n_valid ; ++i )
582 ft->valid_cases += f->count;
586 ft->total_cases = ft->valid_cases ;
587 for(i = 0 ; i < ft->n_missing ; ++i )
590 ft->total_cases += f->count;
595 /* Frees the frequency table for variable V. */
597 cleanup_freq_tab (struct var_freqs *vf)
599 free (vf->tab.valid);
600 freq_hmap_destroy (&vf->tab.data, vf->width);
603 /* Parses the VARIABLES subcommand. */
605 frq_custom_variables (struct lexer *lexer, struct dataset *ds,
606 struct cmd_frequencies *cmd UNUSED, void *frq_ UNUSED)
608 struct frq_proc *frq = frq_;
609 struct variable **vars;
613 lex_match (lexer, '=');
614 if (lex_token (lexer) != T_ALL
615 && (lex_token (lexer) != T_ID
616 || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) == NULL))
619 /* Get list of current variables, to avoid duplicates. */
620 vars = xmalloc (frq->n_vars * sizeof *vars);
621 n_vars = frq->n_vars;
622 for (i = 0; i < frq->n_vars; i++)
623 vars[i] = frq->vars[i].var;
625 if (!parse_variables (lexer, dataset_dict (ds), &vars, &n_vars,
626 PV_APPEND | PV_NO_SCRATCH))
629 frq->vars = xrealloc (frq->vars, n_vars * sizeof *frq->vars);
630 for (i = frq->n_vars; i < n_vars; i++)
632 struct variable *var = vars[i];
633 struct var_freqs *vf = &frq->vars[i];
636 vf->tab.valid = vf->tab.missing = NULL;
637 vf->tab.dict = dataset_dict (ds);
640 vf->width = var_get_width (var);
641 vf->print = *var_get_print_format (var);
643 frq->n_vars = n_vars;
650 /* Parses the GROUPED subcommand, setting the n_grouped, grouped
651 fields of specified variables. */
653 frq_custom_grouped (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *frq_ UNUSED)
655 struct frq_proc *frq = frq_;
657 lex_match (lexer, '=');
658 if ((lex_token (lexer) == T_ID && dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
659 || lex_token (lexer) == T_ID)
664 /* Max, current size of list; list itself. */
670 const struct variable **v;
672 if (!parse_variables_const (lexer, dataset_dict (ds), &v, &n,
673 PV_NO_DUPLICATE | PV_NUMERIC))
675 if (lex_match (lexer, '('))
679 while (lex_integer (lexer))
684 dl = pool_nrealloc (frq->pool, dl, ml, sizeof *dl);
686 dl[nl++] = lex_tokval (lexer);
688 lex_match (lexer, ',');
690 /* Note that nl might still be 0 and dl might still be
691 NULL. That's okay. */
692 if (!lex_match (lexer, ')'))
695 msg (SE, _("`)' expected after GROUPED interval list."));
705 for (i = 0; i < n; i++)
709 for (j = 0; j < frq->n_vars; j++)
711 struct var_freqs *vf = &frq->vars[j];
714 if (vf->groups != NULL)
715 msg (SE, _("Variables %s specified multiple times on "
716 "GROUPED subcommand."), var_get_name (v[i]));
725 msg (SE, _("Variables %s specified on GROUPED but not on "
726 "VARIABLES."), var_get_name (v[i]));
732 if (!lex_match (lexer, '/'))
734 if ((lex_token (lexer) != T_ID || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
735 && lex_token (lexer) != T_ALL)
737 lex_put_back (lexer, '/');
745 /* Adds X to the list of percentiles, keeping the list in proper
746 order. If SHOW is true, the percentile will be shown in the statistics
747 box, otherwise it will be hidden. */
749 add_percentile (struct frq_proc *frq, double x, bool show,
750 size_t *allocated_percentiles)
754 /* Do nothing if it's already in the list */
755 for (i = 0; i < frq->n_percentiles; i++)
757 struct percentile *pc = &frq->percentiles[i];
759 if ( fabs(x - pc->p) < DBL_EPSILON )
761 if (show && !pc->show)
763 frq->n_show_percentiles++;
773 if (frq->n_percentiles >= *allocated_percentiles)
774 frq->percentiles = x2nrealloc (frq->percentiles, allocated_percentiles,
775 sizeof *frq->percentiles);
776 insert_element (frq->percentiles, frq->n_percentiles,
777 sizeof *frq->percentiles, i);
778 frq->percentiles[i].p = x;
779 frq->percentiles[i].show = show;
780 frq->n_percentiles++;
782 frq->n_show_percentiles++;
785 /* Comparison functions. */
787 /* Ascending numeric compare of values. */
789 compare_value_numeric_a (const void *a_, const void *b_,
790 const void *vf_ UNUSED)
792 const struct freq *a = a_;
793 const struct freq *b = b_;
795 if (a->value.f > b->value.f)
797 else if (a->value.f < b->value.f)
803 /* Ascending string compare of values. */
805 compare_value_alpha_a (const void *a_, const void *b_, const void *vf_)
807 const struct freq *a = a_;
808 const struct freq *b = b_;
809 const struct var_freqs *vf = vf_;
811 return value_compare_3way (&a->value, &b->value, vf->width);
814 /* Descending numeric compare of values. */
816 compare_value_numeric_d (const void *a, const void *b, const void *vf_ UNUSED)
818 return -compare_value_numeric_a (a, b, vf_);
821 /* Descending string compare of values. */
823 compare_value_alpha_d (const void *a, const void *b, const void *vf_)
825 return -compare_value_alpha_a (a, b, vf_);
828 /* Ascending numeric compare of frequency;
829 secondary key on ascending numeric value. */
831 compare_freq_numeric_a (const void *a_, const void *b_, const void *vf_ UNUSED)
833 const struct freq *a = a_;
834 const struct freq *b = b_;
836 if (a->count > b->count)
838 else if (a->count < b->count)
841 if (a->value.f > b->value.f)
843 else if (a->value.f < b->value.f)
849 /* Ascending numeric compare of frequency;
850 secondary key on ascending string value. */
852 compare_freq_alpha_a (const void *a_, const void *b_, const void *vf_)
854 const struct freq *a = a_;
855 const struct freq *b = b_;
856 const struct var_freqs *vf = vf_;
858 if (a->count > b->count)
860 else if (a->count < b->count)
863 return value_compare_3way (&a->value, &b->value, vf->width);
866 /* Descending numeric compare of frequency;
867 secondary key on ascending numeric value. */
869 compare_freq_numeric_d (const void *a_, const void *b_, const void *vf_ UNUSED)
871 const struct freq *a = a_;
872 const struct freq *b = b_;
874 if (a->count > b->count)
876 else if (a->count < b->count)
879 if (a->value.f > b->value.f)
881 else if (a->value.f < b->value.f)
887 /* Descending numeric compare of frequency;
888 secondary key on ascending string value. */
890 compare_freq_alpha_d (const void *a_, const void *b_, const void *vf_)
892 const struct freq *a = a_;
893 const struct freq *b = b_;
894 const struct var_freqs *vf = vf_;
896 if (a->count > b->count)
898 else if (a->count < b->count)
901 return value_compare_3way (&a->value, &b->value, vf->width);
904 /* Frequency table display. */
906 /* Displays a full frequency table for variable V. */
908 dump_freq_table (const struct var_freqs *vf, const struct variable *wv)
910 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
911 const struct freq_tab *ft = &vf->tab;
916 double cum_total = 0.0;
917 double cum_freq = 0.0;
919 static const char *headings[] = {
928 n_categories = ft->n_valid + ft->n_missing;
929 t = tab_create (6, n_categories + 2);
930 tab_headers (t, 0, 0, 1, 0);
932 for (x = 0; x < 6; x++)
933 tab_text (t, x, 0, TAB_CENTER | TAT_TITLE, gettext (headings[x]));
936 for (f = ft->valid; f < ft->missing; f++)
939 double percent, valid_percent;
941 cum_freq += f->count;
943 percent = f->count / ft->total_cases * 100.0;
944 valid_percent = f->count / ft->valid_cases * 100.0;
945 cum_total += valid_percent;
947 label = var_lookup_value_label (vf->var, &f->value);
949 tab_text (t, 0, r, TAB_LEFT, label);
951 tab_value (t, 1, r, TAB_NONE, &f->value, ft->dict, &vf->print);
952 tab_double (t, 2, r, TAB_NONE, f->count, wfmt);
953 tab_double (t, 3, r, TAB_NONE, percent, NULL);
954 tab_double (t, 4, r, TAB_NONE, valid_percent, NULL);
955 tab_double (t, 5, r, TAB_NONE, cum_total, NULL);
958 for (; f < &ft->valid[n_categories]; f++)
962 cum_freq += f->count;
964 label = var_lookup_value_label (vf->var, &f->value);
966 tab_text (t, 0, r, TAB_LEFT, label);
968 tab_value (t, 1, r, TAB_NONE, &f->value, ft->dict, &vf->print);
969 tab_double (t, 2, r, TAB_NONE, f->count, wfmt);
970 tab_double (t, 3, r, TAB_NONE,
971 f->count / ft->total_cases * 100.0, NULL);
972 tab_text (t, 4, r, TAB_NONE, _("Missing"));
976 tab_box (t, TAL_1, TAL_1, -1, TAL_1, 0, 0, 5, r);
977 tab_hline (t, TAL_2, 0, 5, 1);
978 tab_hline (t, TAL_2, 0, 5, r);
979 tab_joint_text (t, 0, r, 1, r, TAB_RIGHT | TAT_TITLE, _("Total"));
980 tab_vline (t, TAL_0, 1, r, r);
981 tab_double (t, 2, r, TAB_NONE, cum_freq, wfmt);
982 tab_fixed (t, 3, r, TAB_NONE, 100.0, 5, 1);
983 tab_fixed (t, 4, r, TAB_NONE, 100.0, 5, 1);
985 tab_title (t, "%s", var_to_string (vf->var));
989 /* Statistical display. */
991 /* Calculates all the pertinent statistics for variable V, putting them in
994 calc_stats (const struct frq_proc *frq,
995 const struct var_freqs *vf, double d[FRQ_N_STATS])
997 const struct freq_tab *ft = &vf->tab;
998 double W = ft->valid_cases;
1009 /* Calculate percentiles. */
1011 assert (ft->n_valid > 0);
1013 for (i = 0; i < frq->n_percentiles; i++)
1015 struct percentile *pc = &frq->percentiles[i];
1022 prev_value = SYSMIS;
1023 for (idx = 0; idx < ft->n_valid; ++idx)
1025 f = &ft->valid[idx];
1027 for (i = 0; i < frq->n_percentiles; i++)
1029 struct percentile *pc = &frq->percentiles[i];
1032 if ( pc->flag2 ) continue ;
1034 if ( settings_get_algorithm () != COMPATIBLE )
1036 (ft->valid_cases - 1) * pc->p;
1039 (ft->valid_cases + 1) * pc->p - 1;
1043 pc->x2 = f->value.f;
1044 pc->x1 = prev_value;
1051 if ( f->count > 1 && rank - (f->count - 1) > tp )
1053 pc->x2 = pc->x1 = f->value.f;
1064 prev_value = f->value.f;
1067 for (i = 0; i < frq->n_percentiles; i++)
1069 struct percentile *pc = &frq->percentiles[i];
1071 /* Catches the case when p == 100% */
1073 pc->x1 = pc->x2 = f->value.f;
1076 printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
1077 i,pc->p,pc->x1,pc->x2);
1081 for (i = 0; i < frq->n_percentiles; i++)
1083 struct percentile *pc = &frq->percentiles[i];
1087 if ( settings_get_algorithm () != COMPATIBLE )
1089 s = modf((ft->valid_cases - 1) * pc->p , &dummy);
1093 s = modf((ft->valid_cases + 1) * pc->p -1, &dummy);
1096 pc->value = pc->x1 + (pc->x2 - pc->x1) * s ;
1100 /* Calculate the mode. */
1103 for (f = ft->valid; f < ft->missing; f++)
1105 if (most_often < f->count)
1107 most_often = f->count;
1108 X_mode = f->value.f;
1110 else if (most_often == f->count)
1112 /* A duplicate mode is undefined.
1113 FIXME: keep track of *all* the modes. */
1118 /* Calculate moments. */
1119 m = moments_create (MOMENT_KURTOSIS);
1120 for (f = ft->valid; f < ft->missing; f++)
1121 moments_pass_one (m, f->value.f, f->count);
1122 for (f = ft->valid; f < ft->missing; f++)
1123 moments_pass_two (m, f->value.f, f->count);
1124 moments_calculate (m, NULL, &d[FRQ_MEAN], &d[FRQ_VARIANCE],
1125 &d[FRQ_SKEW], &d[FRQ_KURT]);
1126 moments_destroy (m);
1128 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1129 d[FRQ_MIN] = ft->valid[0].value.f;
1130 d[FRQ_MAX] = ft->valid[ft->n_valid - 1].value.f;
1131 d[FRQ_MODE] = X_mode;
1132 d[FRQ_RANGE] = d[FRQ_MAX] - d[FRQ_MIN];
1133 d[FRQ_SUM] = d[FRQ_MEAN] * W;
1134 d[FRQ_STDDEV] = sqrt (d[FRQ_VARIANCE]);
1135 d[FRQ_SEMEAN] = d[FRQ_STDDEV] / sqrt (W);
1136 d[FRQ_SESKEW] = calc_seskew (W);
1137 d[FRQ_SEKURT] = calc_sekurt (W);
1140 /* Displays a table of all the statistics requested for variable V. */
1142 dump_statistics (const struct frq_proc *frq, const struct var_freqs *vf,
1143 const struct variable *wv)
1145 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1146 const struct freq_tab *ft = &vf->tab;
1147 double stat_value[FRQ_N_STATS];
1148 struct tab_table *t;
1151 if (var_is_alpha (vf->var))
1154 if (ft->n_valid == 0)
1156 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1157 var_get_name (vf->var));
1160 calc_stats (frq, vf, stat_value);
1162 t = tab_create (3, frq->n_stats + frq->n_show_percentiles + 2);
1164 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1167 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1168 tab_vline (t, TAL_GAP , 1, 0, tab_nr(t) - 1 ) ;
1170 r=2; /* N missing and N valid are always dumped */
1172 for (i = 0; i < FRQ_N_STATS; i++)
1173 if (frq->stats & BIT_INDEX (i))
1175 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1176 gettext (st_name[i].s10));
1177 tab_double (t, 2, r, TAB_NONE, stat_value[i], NULL);
1181 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1182 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1183 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1185 tab_double (t, 2, 0, TAB_NONE, ft->valid_cases, wfmt);
1186 tab_double (t, 2, 1, TAB_NONE, ft->total_cases - ft->valid_cases, wfmt);
1188 for (i = 0; i < frq->n_percentiles; i++, r++)
1190 struct percentile *pc = &frq->percentiles[i];
1197 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1201 tab_text (t, 1, r, TAB_LEFT, _("50 (Median)"));
1203 tab_fixed (t, 1, r, TAB_LEFT, pc->p * 100, 3, 0);
1204 tab_double (t, 2, r, TAB_NONE, pc->value,
1205 var_get_print_format (vf->var));
1208 tab_title (t, "%s", var_to_string (vf->var));
1214 calculate_iqr (const struct frq_proc *frq)
1220 for (i = 0; i < frq->n_percentiles; i++)
1222 struct percentile *pc = &frq->percentiles[i];
1224 if (fabs (0.25 - pc->p) < DBL_EPSILON)
1226 else if (fabs (0.75 - pc->p) < DBL_EPSILON)
1230 return q1 == SYSMIS || q3 == SYSMIS ? SYSMIS : q3 - q1;
1234 chart_includes_value (const struct frq_chart *chart,
1235 const struct variable *var,
1236 const union value *value)
1238 if (!chart->include_missing && var_is_value_missing (var, value, MV_ANY))
1241 if (var_is_numeric (var)
1242 && ((chart->x_min != SYSMIS && value->f < chart->x_min)
1243 || (chart->x_max != SYSMIS && value->f > chart->x_max)))
1249 /* Create a gsl_histogram from a freq_tab */
1251 freq_tab_to_hist (const struct frq_proc *frq, const struct freq_tab *ft,
1252 const struct variable *var)
1254 double x_min, x_max, valid_freq;
1257 struct histogram *histogram;
1261 /* Find out the extremes of the x value, within the range to be included in
1262 the histogram, and sum the total frequency of those values. */
1266 for (i = 0; i < ft->n_valid; i++)
1268 const struct freq *f = &ft->valid[i];
1269 if (chart_includes_value (frq->hist, var, &f->value))
1271 x_min = MIN (x_min, f->value.f);
1272 x_max = MAX (x_max, f->value.f);
1273 valid_freq += f->count;
1277 /* Freedman-Diaconis' choice of bin width. */
1278 iqr = calculate_iqr (frq);
1281 double bin_width = 2 * iqr / pow (valid_freq, 1.0 / 3.0);
1282 bins = (x_max - x_min) / bin_width;
1285 else if (bins > 400)
1291 histogram = histogram_create (bins, x_min, x_max);
1292 for (i = 0; i < ft->n_valid; i++)
1294 const struct freq *f = &ft->valid[i];
1295 if (chart_includes_value (frq->hist, var, &f->value))
1296 histogram_add (histogram, f->value.f, f->count);
1303 add_slice (const struct frq_chart *pie, const struct freq *freq,
1304 const struct variable *var, struct slice *slice)
1306 if (chart_includes_value (pie, var, &freq->value))
1308 ds_init_empty (&slice->label);
1309 var_append_value_name (var, &freq->value, &slice->label);
1310 slice->magnitude = freq->count;
1317 /* Allocate an array of slices and fill them from the data in frq_tab
1318 n_slices will contain the number of slices allocated.
1319 The caller is responsible for freeing slices
1321 static struct slice *
1322 freq_tab_to_slice_array(const struct frq_chart *pie,
1323 const struct freq_tab *frq_tab,
1324 const struct variable *var,
1327 struct slice *slices;
1331 slices = xnmalloc (frq_tab->n_valid + frq_tab->n_missing, sizeof *slices);
1334 for (i = 0; i < frq_tab->n_valid; i++)
1335 n_slices += add_slice (pie, &frq_tab->valid[i], var, &slices[n_slices]);
1336 for (i = 0; i < frq_tab->n_missing; i++)
1337 n_slices += add_slice (pie, &frq_tab->missing[i], var, &slices[n_slices]);
1339 *n_slicesp = n_slices;
1347 do_piechart(const struct frq_chart *pie, const struct variable *var,
1348 const struct freq_tab *frq_tab)
1350 struct slice *slices;
1353 slices = freq_tab_to_slice_array (pie, frq_tab, var, &n_slices);
1356 msg (SW, _("Omitting pie chart for %s, which has only %d unique values."),
1357 var_get_name (var), n_slices);
1358 else if (n_slices > 50)
1359 msg (SW, _("Omitting pie chart for %s, which has over 50 unique values."),
1360 var_get_name (var));
1362 chart_item_submit (piechart_create (var_to_string(var), slices, n_slices));
1364 for (i = 0; i < n_slices; i++)
1365 ds_destroy (&slices[i].label);