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/>. */
20 * Remember that histograms, bar charts need mean, stddev.
27 #include <gsl/gsl_histogram.h>
29 #include <data/case.h>
30 #include <data/casegrouper.h>
31 #include <data/casereader.h>
32 #include <data/dictionary.h>
33 #include <data/format.h>
34 #include <data/procedure.h>
35 #include <data/settings.h>
36 #include <data/value-labels.h>
37 #include <data/variable.h>
38 #include <language/command.h>
39 #include <language/dictionary/split-file.h>
40 #include <language/lexer/lexer.h>
41 #include <libpspp/array.h>
42 #include <libpspp/bit-vector.h>
43 #include <libpspp/compiler.h>
44 #include <libpspp/hash.h>
45 #include <libpspp/message.h>
46 #include <libpspp/misc.h>
47 #include <libpspp/pool.h>
48 #include <libpspp/str.h>
49 #include <math/histogram.h>
50 #include <math/moments.h>
51 #include <output/chart-item.h>
52 #include <output/charts/piechart.h>
53 #include <output/charts/plot-hist.h>
54 #include <output/tab.h>
62 #define _(msgid) gettext (msgid)
63 #define N_(msgid) msgid
70 +format=cond:condense/onepage(*n:onepage_limit,"%s>=0")/!standard,
71 table:limit(n:limit,"%s>0")/notable/!table,
72 labels:!labels/nolabels,
73 sort:!avalue/dvalue/afreq/dfreq,
74 spaces:!single/double,
75 paging:newpage/!oldpage;
76 missing=miss:include/!exclude;
77 barchart(ba_)=:minimum(d:min),
79 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0");
80 piechart(pie_)=:minimum(d:min),
82 missing:missing/!nomissing;
83 histogram(hi_)=:minimum(d:min),
85 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
86 norm:!nonormal/normal,
87 incr:increment(d:inc,"%s>0");
88 hbar(hb_)=:minimum(d:min),
90 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
91 norm:!nonormal/normal,
92 incr:increment(d:inc,"%s>0");
95 +percentiles = double list;
96 +statistics[st_]=1|mean,2|semean,3|median,4|mode,5|stddev,6|variance,
97 7|kurtosis,8|skewness,9|range,10|minimum,11|maximum,12|sum,
98 13|default,14|seskewness,15|sekurtosis,all,none.
106 frq_mean = 0, frq_semean, frq_median, frq_mode, frq_stddev, frq_variance,
107 frq_kurt, frq_sekurt, frq_skew, frq_seskew, frq_range, frq_min, frq_max,
111 /* Description of a statistic. */
114 int st_indx; /* Index into a_statistics[]. */
115 const char *s10; /* Identifying string. */
118 /* Table of statistics, indexed by dsc_*. */
119 static const struct frq_info st_name[frq_n_stats + 1] =
121 {FRQ_ST_MEAN, N_("Mean")},
122 {FRQ_ST_SEMEAN, N_("S.E. Mean")},
123 {FRQ_ST_MEDIAN, N_("Median")},
124 {FRQ_ST_MODE, N_("Mode")},
125 {FRQ_ST_STDDEV, N_("Std Dev")},
126 {FRQ_ST_VARIANCE, N_("Variance")},
127 {FRQ_ST_KURTOSIS, N_("Kurtosis")},
128 {FRQ_ST_SEKURTOSIS, N_("S.E. Kurt")},
129 {FRQ_ST_SKEWNESS, N_("Skewness")},
130 {FRQ_ST_SESKEWNESS, N_("S.E. Skew")},
131 {FRQ_ST_RANGE, N_("Range")},
132 {FRQ_ST_MINIMUM, N_("Minimum")},
133 {FRQ_ST_MAXIMUM, N_("Maximum")},
134 {FRQ_ST_SUM, N_("Sum")},
138 /* Percentiles to calculate. */
142 double p; /* the %ile to be calculated */
143 double value; /* the %ile's value */
144 double x1; /* The datum value <= the percentile */
145 double x2; /* The datum value >= the percentile */
147 int flag2; /* Set to 1 if this percentile value has been found */
148 bool show; /* True to show this percentile in the statistics box. */
152 static void add_percentile (double x, bool show);
154 static struct percentile *percentiles;
155 static int n_percentiles, n_show_percentiles;
157 /* Groups of statistics. */
159 #define frq_default \
160 (BI (frq_mean) | BI (frq_stddev) | BI (frq_min) | BI (frq_max))
162 (BI (frq_sum) | BI(frq_min) | BI(frq_max) \
163 | BI(frq_mean) | BI(frq_semean) | BI(frq_stddev) \
164 | BI(frq_variance) | BI(frq_kurt) | BI(frq_sekurt) \
165 | BI(frq_skew) | BI(frq_seskew) | BI(frq_range) \
166 | BI(frq_range) | BI(frq_mode) | BI(frq_median))
168 /* Statistics; number of statistics. */
169 static unsigned long stats;
172 /* Types of graphs. */
175 GFT_NONE, /* Don't draw graphs. */
176 GFT_BAR, /* Draw bar charts. */
177 GFT_HIST, /* Draw histograms. */
178 GFT_PIE, /* Draw piechart */
179 GFT_HBAR /* Draw bar charts or histograms at our discretion. */
182 /* Parsed command. */
183 static struct cmd_frequencies cmd;
185 /* Summary of the barchart, histogram, and hbar subcommands. */
186 /* FIXME: These should not be mututally exclusive */
187 static int chart; /* NONE/BAR/HIST/HBAR/PIE. */
188 static double min, max; /* Minimum, maximum on y axis. */
189 static int format; /* FREQ/PERCENT: Scaling of y axis. */
190 static double scale, incr; /* FIXME */
191 static int normal; /* FIXME */
193 /* Variables for which to calculate statistics. */
194 static size_t n_variables;
195 static const struct variable **v_variables;
198 static struct pool *data_pool; /* For per-SPLIT FILE group data. */
199 static struct pool *syntax_pool; /* For syntax-related data. */
201 /* Frequency tables. */
203 /* Entire frequency table. */
206 struct hsh_table *data; /* Undifferentiated data. */
207 struct freq_mutable *valid; /* Valid freqs. */
208 int n_valid; /* Number of total freqs. */
209 const struct dictionary *dict; /* The dict from whence entries in the table
212 struct freq_mutable *missing; /* Missing freqs. */
213 int n_missing; /* Number of missing freqs. */
216 double total_cases; /* Sum of weights of all cases. */
217 double valid_cases; /* Sum of weights of valid cases. */
221 /* Per-variable frequency data. */
224 /* Freqency table. */
225 struct freq_tab tab; /* Frequencies table to use. */
228 int n_groups; /* Number of groups. */
229 double *groups; /* Groups. */
232 double stat[frq_n_stats];
234 /* Variable attributes. */
236 struct fmt_spec print;
239 static inline struct var_freqs *
240 get_var_freqs (const struct variable *v)
242 return var_get_aux (v);
245 static void determine_charts (void);
247 static void calc_stats (const struct variable *v, double d[frq_n_stats]);
249 static void precalc (struct casereader *, struct dataset *);
250 static void calc (const struct ccase *, const struct dataset *);
251 static void postcalc (const struct dataset *);
253 static void postprocess_freq_tab (const struct variable *);
254 static void dump_full ( const struct variable *, const struct variable *);
255 static void dump_condensed (const struct variable *, const struct variable *);
256 static void dump_statistics (const struct variable *, bool show_varname, const struct variable *);
257 static void cleanup_freq_tab (const struct variable *);
259 static hsh_compare_func compare_value_numeric_a, compare_value_alpha_a;
260 static hsh_compare_func compare_value_numeric_d, compare_value_alpha_d;
261 static hsh_compare_func compare_freq_numeric_a, compare_freq_alpha_a;
262 static hsh_compare_func compare_freq_numeric_d, compare_freq_alpha_d;
265 static void do_piechart(const struct variable *var,
266 const struct freq_tab *frq_tab);
269 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var);
273 /* Parser and outline. */
275 static int internal_cmd_frequencies (struct lexer *lexer, struct dataset *ds);
278 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
282 syntax_pool = pool_create ();
283 result = internal_cmd_frequencies (lexer, ds);
284 pool_destroy (syntax_pool);
286 pool_destroy (data_pool);
294 internal_cmd_frequencies (struct lexer *lexer, struct dataset *ds)
296 struct casegrouper *grouper;
297 struct casereader *input, *group;
302 n_show_percentiles = 0;
308 if (!parse_frequencies (lexer, ds, &cmd, NULL))
311 if (cmd.onepage_limit == LONG_MIN)
312 cmd.onepage_limit = 50;
314 /* Figure out statistics to calculate. */
316 if (cmd.a_statistics[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
317 stats |= frq_default;
318 if (cmd.a_statistics[FRQ_ST_ALL])
320 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
321 stats &= ~BIT_INDEX (frq_median);
322 for (i = 0; i < frq_n_stats; i++)
323 if (cmd.a_statistics[st_name[i].st_indx])
324 stats |= BIT_INDEX (i);
325 if (stats & frq_kurt)
326 stats |= BIT_INDEX (frq_sekurt);
327 if (stats & frq_skew)
328 stats |= BIT_INDEX (frq_seskew);
330 /* Calculate n_stats. */
332 for (i = 0; i < frq_n_stats; i++)
333 if ((stats & BIT_INDEX (i)))
338 if (chart != GFT_NONE || cmd.sbc_ntiles)
339 cmd.sort = FRQ_AVALUE;
341 /* Work out what percentiles need to be calculated */
342 if ( cmd.sbc_percentiles )
344 for ( i = 0 ; i < MAXLISTS ; ++i )
347 subc_list_double *ptl_list = &cmd.dl_percentiles[i];
348 for ( pl = 0 ; pl < subc_list_double_count(ptl_list); ++pl)
349 add_percentile (subc_list_double_at(ptl_list, pl) / 100.0, true);
352 if ( cmd.sbc_ntiles )
354 for ( i = 0 ; i < cmd.sbc_ntiles ; ++i )
357 for (j = 0; j <= cmd.n_ntiles[i]; ++j )
358 add_percentile (j / (double) cmd.n_ntiles[i], true);
361 if (stats & BIT_INDEX (frq_median))
363 /* Treat the median as the 50% percentile.
364 We output it in the percentiles table as "50 (Median)." */
365 add_percentile (0.5, true);
366 stats &= ~BIT_INDEX (frq_median);
369 if (chart == GFT_HIST)
371 add_percentile (0.25, false);
372 add_percentile (0.75, false);
376 input = casereader_create_filter_weight (proc_open (ds), dataset_dict (ds),
378 grouper = casegrouper_create_splits (input, dataset_dict (ds));
379 for (; casegrouper_get_next_group (grouper, &group);
380 casereader_destroy (group))
385 for (; (c = casereader_read (group)) != NULL; case_unref (c))
389 ok = casegrouper_destroy (grouper);
390 ok = proc_commit (ds) && ok;
392 free_frequencies(&cmd);
394 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
397 /* Figure out which charts the user requested. */
399 determine_charts (void)
401 int count = (!!cmd.sbc_histogram) + (!!cmd.sbc_barchart) +
402 (!!cmd.sbc_hbar) + (!!cmd.sbc_piechart);
412 msg (SW, _("At most one of BARCHART, HISTOGRAM, or HBAR should be "
413 "given. HBAR will be assumed. Argument values will be "
414 "given precedence increasing along the order given."));
416 else if (cmd.sbc_histogram)
418 else if (cmd.sbc_barchart)
420 else if (cmd.sbc_piechart)
431 if (cmd.sbc_barchart)
433 if (cmd.ba_min != SYSMIS)
435 if (cmd.ba_max != SYSMIS)
437 if (cmd.ba_scale == FRQ_FREQ)
442 else if (cmd.ba_scale == FRQ_PERCENT)
444 format = FRQ_PERCENT;
449 if (cmd.sbc_histogram)
451 if (cmd.hi_min != SYSMIS)
453 if (cmd.hi_max != SYSMIS)
455 if (cmd.hi_scale == FRQ_FREQ)
460 else if (cmd.hi_scale == FRQ_PERCENT)
462 format = FRQ_PERCENT;
465 if (cmd.hi_norm != FRQ_NONORMAL )
467 if (cmd.hi_incr == FRQ_INCREMENT)
473 if (cmd.hb_min != SYSMIS)
475 if (cmd.hb_max != SYSMIS)
477 if (cmd.hb_scale == FRQ_FREQ)
482 else if (cmd.hb_scale == FRQ_PERCENT)
484 format = FRQ_PERCENT;
489 if (cmd.hb_incr == FRQ_INCREMENT)
493 if (min != SYSMIS && max != SYSMIS && min >= max)
495 msg (SE, _("MAX must be greater than or equal to MIN, if both are "
496 "specified. However, MIN was specified as %g and MAX as %g. "
497 "MIN and MAX will be ignored."), min, max);
502 /* Add data from case C to the frequency table. */
504 calc (const struct ccase *c, const struct dataset *ds)
506 double weight = dict_get_case_weight (dataset_dict (ds), c, NULL);
509 for (i = 0; i < n_variables; i++)
511 const struct variable *v = v_variables[i];
512 const union value *val = case_data (c, v);
513 struct var_freqs *vf = get_var_freqs (v);
514 struct freq_tab *ft = &vf->tab;
516 struct freq_mutable target;
517 struct freq_mutable **fpp;
520 fpp = (struct freq_mutable **) hsh_probe (ft->data, &target);
523 (*fpp)->count += weight;
526 struct freq_mutable *fp = pool_alloc (data_pool, sizeof *fp);
528 value_init_pool (data_pool, &fp->value, vf->width);
529 value_copy (&fp->value, val, vf->width);
535 /* Prepares each variable that is the target of FREQUENCIES by setting
536 up its hash table. */
538 precalc (struct casereader *input, struct dataset *ds)
543 c = casereader_peek (input, 0);
546 output_split_file_values (ds, c);
550 pool_destroy (data_pool);
551 data_pool = pool_create ();
553 for (i = 0; i < n_variables; i++)
555 const struct variable *v = v_variables[i];
556 struct freq_tab *ft = &get_var_freqs (v)->tab;
558 ft->data = hsh_create (16, compare_freq, hash_freq, NULL, v);
562 /* Finishes up with the variables after frequencies have been
563 calculated. Displays statistics, percentiles, ... */
565 postcalc (const struct dataset *ds)
567 const struct dictionary *dict = dataset_dict (ds);
568 const struct variable *wv = dict_get_weight (dict);
571 for (i = 0; i < n_variables; i++)
573 const struct variable *v = v_variables[i];
574 struct var_freqs *vf = get_var_freqs (v);
575 struct freq_tab *ft = &vf->tab;
577 int dumped_freq_tab = 1;
579 postprocess_freq_tab (v);
581 /* Frequencies tables. */
582 n_categories = ft->n_valid + ft->n_missing;
583 if (cmd.table == FRQ_TABLE
584 || (cmd.table == FRQ_LIMIT && n_categories <= cmd.limit))
588 dump_condensed (v, wv);
594 if (n_categories > cmd.onepage_limit)
595 dump_condensed (v, wv);
607 dump_statistics (v, !dumped_freq_tab, wv);
611 if ( chart == GFT_HIST && var_is_numeric (v) && ft->n_valid > 0)
613 double d[frq_n_stats];
614 struct histogram *hist ;
618 hist = freq_tab_to_hist (ft,v);
620 chart_item_submit (histogram_chart_create (
621 hist->gsl_hist, var_to_string(v),
627 statistic_destroy (&hist->parent);
630 if ( chart == GFT_PIE)
632 do_piechart(v_variables[i], ft);
635 cleanup_freq_tab (v);
640 /* Returns the comparison function that should be used for
641 sorting a frequency table by FRQ_SORT using VAL_TYPE
643 static hsh_compare_func *
644 get_freq_comparator (int frq_sort, enum val_type val_type)
646 bool is_numeric = val_type == VAL_NUMERIC;
650 return is_numeric ? compare_value_numeric_a : compare_value_alpha_a;
652 return is_numeric ? compare_value_numeric_d : compare_value_alpha_d;
654 return is_numeric ? compare_freq_numeric_a : compare_freq_alpha_a;
656 return is_numeric ? compare_freq_numeric_d : compare_freq_alpha_d;
662 /* Returns true iff the value in struct freq_mutable F is non-missing
665 not_missing (const void *f_, const void *v_)
667 const struct freq_mutable *f = f_;
668 const struct variable *v = v_;
670 return !var_is_value_missing (v, &f->value, MV_ANY);
673 /* Summarizes the frequency table data for variable V. */
675 postprocess_freq_tab (const struct variable *v)
677 hsh_compare_func *compare;
681 struct freq_mutable *freqs, *f;
684 ft = &get_var_freqs (v)->tab;
685 compare = get_freq_comparator (cmd.sort, var_get_type (v));
687 /* Extract data from hash table. */
688 count = hsh_count (ft->data);
689 data = hsh_data (ft->data);
691 /* Copy dereferenced data into freqs. */
692 freqs = xnmalloc (count, sizeof *freqs);
693 for (i = 0; i < count; i++)
695 struct freq_mutable *f = data[i];
699 /* Put data into ft. */
701 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, v);
702 ft->missing = freqs + ft->n_valid;
703 ft->n_missing = count - ft->n_valid;
706 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare, v);
707 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, v);
709 /* Summary statistics. */
710 ft->valid_cases = 0.0;
711 for(i = 0 ; i < ft->n_valid ; ++i )
714 ft->valid_cases += f->count;
718 ft->total_cases = ft->valid_cases ;
719 for(i = 0 ; i < ft->n_missing ; ++i )
722 ft->total_cases += f->count;
727 /* Frees the frequency table for variable V. */
729 cleanup_freq_tab (const struct variable *v)
731 struct freq_tab *ft = &get_var_freqs (v)->tab;
733 hsh_destroy (ft->data);
736 /* Parses the VARIABLES subcommand, adding to
737 {n_variables,v_variables}. */
739 frq_custom_variables (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *aux UNUSED)
741 size_t old_n_variables = n_variables;
744 lex_match (lexer, '=');
745 if (lex_token (lexer) != T_ALL && (lex_token (lexer) != T_ID
746 || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) == NULL))
749 if (!parse_variables_const (lexer, dataset_dict (ds), &v_variables, &n_variables,
750 PV_APPEND | PV_NO_SCRATCH))
753 for (i = old_n_variables; i < n_variables; i++)
755 const struct variable *v = v_variables[i];
756 struct var_freqs *vf;
758 if (var_get_aux (v) != NULL)
760 msg (SE, _("Variable %s specified multiple times on VARIABLES "
761 "subcommand."), var_get_name (v));
764 vf = var_attach_aux (v, xmalloc (sizeof *vf), var_dtor_free);
765 vf->tab.valid = vf->tab.missing = NULL;
766 vf->tab.dict = dataset_dict (ds);
769 vf->width = var_get_width (v);
770 vf->print = *var_get_print_format (v);
775 /* Parses the GROUPED subcommand, setting the n_grouped, grouped
776 fields of specified variables. */
778 frq_custom_grouped (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *aux UNUSED)
780 lex_match (lexer, '=');
781 if ((lex_token (lexer) == T_ID && dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
782 || lex_token (lexer) == T_ID)
787 /* Max, current size of list; list itself. */
793 const struct variable **v;
795 if (!parse_variables_const (lexer, dataset_dict (ds), &v, &n,
796 PV_NO_DUPLICATE | PV_NUMERIC))
798 if (lex_match (lexer, '('))
802 while (lex_integer (lexer))
807 dl = pool_nrealloc (syntax_pool, dl, ml, sizeof *dl);
809 dl[nl++] = lex_tokval (lexer);
811 lex_match (lexer, ',');
813 /* Note that nl might still be 0 and dl might still be
814 NULL. That's okay. */
815 if (!lex_match (lexer, ')'))
818 msg (SE, _("`)' expected after GROUPED interval list."));
828 for (i = 0; i < n; i++)
829 if (var_get_aux (v[i]) == NULL)
830 msg (SE, _("Variables %s specified on GROUPED but not on "
831 "VARIABLES."), var_get_name (v[i]));
834 struct var_freqs *vf = get_var_freqs (v[i]);
836 if (vf->groups != NULL)
837 msg (SE, _("Variables %s specified multiple times on GROUPED "
838 "subcommand."), var_get_name (v[i]));
846 if (!lex_match (lexer, '/'))
848 if ((lex_token (lexer) != T_ID || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
849 && lex_token (lexer) != T_ALL)
851 lex_put_back (lexer, '/');
859 /* Adds X to the list of percentiles, keeping the list in proper
860 order. If SHOW is true, the percentile will be shown in the statistics
861 box, otherwise it will be hidden. */
863 add_percentile (double x, bool show)
867 for (i = 0; i < n_percentiles; i++)
869 /* Do nothing if it's already in the list */
870 if ( fabs(x - percentiles[i].p) < DBL_EPSILON )
872 if (show && !percentiles[i].show)
874 n_show_percentiles++;
875 percentiles[i].show = true;
880 if (x < percentiles[i].p)
884 if (i >= n_percentiles || x != percentiles[i].p)
886 percentiles = pool_nrealloc (syntax_pool, percentiles,
887 n_percentiles + 1, sizeof *percentiles);
888 insert_element (percentiles, n_percentiles, sizeof *percentiles, i);
889 percentiles[i].p = x;
890 percentiles[i].show = show;
893 n_show_percentiles++;
897 /* Comparison functions. */
899 /* Ascending numeric compare of values. */
901 compare_value_numeric_a (const void *a_, const void *b_, const void *aux UNUSED)
903 const struct freq_mutable *a = a_;
904 const struct freq_mutable *b = b_;
906 if (a->value.f > b->value.f)
908 else if (a->value.f < b->value.f)
914 /* Ascending string compare of values. */
916 compare_value_alpha_a (const void *a_, const void *b_, const void *v_)
918 const struct freq_mutable *a = a_;
919 const struct freq_mutable *b = b_;
920 const struct variable *v = v_;
921 struct var_freqs *vf = get_var_freqs (v);
923 return value_compare_3way (&a->value, &b->value, vf->width);
926 /* Descending numeric compare of values. */
928 compare_value_numeric_d (const void *a, const void *b, const void *aux UNUSED)
930 return -compare_value_numeric_a (a, b, aux);
933 /* Descending string compare of values. */
935 compare_value_alpha_d (const void *a, const void *b, const void *v)
937 return -compare_value_alpha_a (a, b, v);
940 /* Ascending numeric compare of frequency;
941 secondary key on ascending numeric value. */
943 compare_freq_numeric_a (const void *a_, const void *b_, const void *aux UNUSED)
945 const struct freq_mutable *a = a_;
946 const struct freq_mutable *b = b_;
948 if (a->count > b->count)
950 else if (a->count < b->count)
953 if (a->value.f > b->value.f)
955 else if (a->value.f < b->value.f)
961 /* Ascending numeric compare of frequency;
962 secondary key on ascending string value. */
964 compare_freq_alpha_a (const void *a_, const void *b_, const void *v_)
966 const struct freq_mutable *a = a_;
967 const struct freq_mutable *b = b_;
968 const struct variable *v = v_;
969 struct var_freqs *vf = get_var_freqs (v);
971 if (a->count > b->count)
973 else if (a->count < b->count)
976 return value_compare_3way (&a->value, &b->value, vf->width);
979 /* Descending numeric compare of frequency;
980 secondary key on ascending numeric value. */
982 compare_freq_numeric_d (const void *a_, const void *b_, const void *aux UNUSED)
984 const struct freq_mutable *a = a_;
985 const struct freq_mutable *b = b_;
987 if (a->count > b->count)
989 else if (a->count < b->count)
992 if (a->value.f > b->value.f)
994 else if (a->value.f < b->value.f)
1000 /* Descending numeric compare of frequency;
1001 secondary key on ascending string value. */
1003 compare_freq_alpha_d (const void *a_, const void *b_, const void *v_)
1005 const struct freq_mutable *a = a_;
1006 const struct freq_mutable *b = b_;
1007 const struct variable *v = v_;
1008 struct var_freqs *vf = get_var_freqs (v);
1010 if (a->count > b->count)
1012 else if (a->count < b->count)
1015 return value_compare_3way (&a->value, &b->value, vf->width);
1018 /* Frequency table display. */
1020 /* Displays a full frequency table for variable V. */
1022 dump_full (const struct variable *v, const struct variable *wv)
1024 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1026 struct var_freqs *vf;
1027 struct freq_tab *ft;
1028 struct freq_mutable *f;
1029 struct tab_table *t;
1031 double cum_total = 0.0;
1032 double cum_freq = 0.0;
1034 static const char *headings[] = {
1038 N_("Valid Percent"),
1042 const bool lab = (cmd.labels == FRQ_LABELS);
1044 vf = get_var_freqs (v);
1046 n_categories = ft->n_valid + ft->n_missing;
1047 t = tab_create (5 + lab, n_categories + 2);
1048 tab_headers (t, 0, 0, 1, 0);
1051 tab_text (t, 0, 0, TAB_CENTER | TAT_TITLE, _("Value Label"));
1053 for (x = 0; x < 5; x++)
1054 tab_text (t, lab + x, 0, TAB_CENTER | TAT_TITLE, gettext (headings[x]));
1057 for (f = ft->valid; f < ft->missing; f++)
1059 double percent, valid_percent;
1061 cum_freq += f->count;
1063 percent = f->count / ft->total_cases * 100.0;
1064 valid_percent = f->count / ft->valid_cases * 100.0;
1065 cum_total += valid_percent;
1069 const char *label = var_lookup_value_label (v, &f->value);
1071 tab_text (t, 0, r, TAB_LEFT, label);
1074 tab_value (t, 0 + lab, r, TAB_NONE, &f->value, ft->dict, &vf->print);
1075 tab_double (t, 1 + lab, r, TAB_NONE, f->count, wfmt);
1076 tab_double (t, 2 + lab, r, TAB_NONE, percent, NULL);
1077 tab_double (t, 3 + lab, r, TAB_NONE, valid_percent, NULL);
1078 tab_double (t, 4 + lab, r, TAB_NONE, cum_total, NULL);
1081 for (; f < &ft->valid[n_categories]; f++)
1083 cum_freq += f->count;
1087 const char *label = var_lookup_value_label (v, &f->value);
1089 tab_text (t, 0, r, TAB_LEFT, label);
1092 tab_value (t, 0 + lab, r, TAB_NONE, &f->value, ft->dict, &vf->print);
1093 tab_double (t, 1 + lab, r, TAB_NONE, f->count, wfmt);
1094 tab_double (t, 2 + lab, r, TAB_NONE,
1095 f->count / ft->total_cases * 100.0, NULL);
1096 tab_text (t, 3 + lab, r, TAB_NONE, _("Missing"));
1100 tab_box (t, TAL_1, TAL_1,
1101 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1103 tab_hline (t, TAL_2, 0, 4 + lab, 1);
1104 tab_hline (t, TAL_2, 0, 4 + lab, r);
1105 tab_joint_text (t, 0, r, 0 + lab, r, TAB_RIGHT | TAT_TITLE, _("Total"));
1106 tab_vline (t, TAL_0, 1, r, r);
1107 tab_double (t, 1 + lab, r, TAB_NONE, cum_freq, wfmt);
1108 tab_fixed (t, 2 + lab, r, TAB_NONE, 100.0, 5, 1);
1109 tab_fixed (t, 3 + lab, r, TAB_NONE, 100.0, 5, 1);
1111 tab_title (t, "%s", var_to_string (v));
1115 /* Display condensed frequency table for variable V. */
1117 dump_condensed (const struct variable *v, const struct variable *wv)
1119 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1121 struct var_freqs *vf;
1122 struct freq_tab *ft;
1123 struct freq_mutable *f;
1124 struct tab_table *t;
1126 double cum_total = 0.0;
1128 vf = get_var_freqs (v);
1130 n_categories = ft->n_valid + ft->n_missing;
1131 t = tab_create (4, n_categories + 2);
1133 tab_headers (t, 0, 0, 2, 0);
1134 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value"));
1135 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Freq"));
1136 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1137 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Cum"));
1138 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1141 for (f = ft->valid; f < ft->missing; f++)
1145 percent = f->count / ft->total_cases * 100.0;
1146 cum_total += f->count / ft->valid_cases * 100.0;
1148 tab_value (t, 0, r, TAB_NONE, &f->value, ft->dict, &vf->print);
1149 tab_double (t, 1, r, TAB_NONE, f->count, wfmt);
1150 tab_double (t, 2, r, TAB_NONE, percent, NULL);
1151 tab_double (t, 3, r, TAB_NONE, cum_total, NULL);
1154 for (; f < &ft->valid[n_categories]; f++)
1156 tab_value (t, 0, r, TAB_NONE, &f->value, ft->dict, &vf->print);
1157 tab_double (t, 1, r, TAB_NONE, f->count, wfmt);
1158 tab_double (t, 2, r, TAB_NONE,
1159 f->count / ft->total_cases * 100.0, NULL);
1163 tab_box (t, TAL_1, TAL_1,
1164 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1166 tab_hline (t, TAL_2, 0, 3, 2);
1167 tab_title (t, "%s", var_to_string (v));
1171 /* Statistical display. */
1173 /* Calculates all the pertinent statistics for variable V, putting them in
1176 calc_stats (const struct variable *v, double d[frq_n_stats])
1178 struct freq_tab *ft = &get_var_freqs (v)->tab;
1179 double W = ft->valid_cases;
1181 struct freq_mutable *f=0;
1189 /* Calculate percentiles. */
1191 assert (ft->n_valid > 0);
1193 for (i = 0; i < n_percentiles; i++)
1195 percentiles[i].flag = 0;
1196 percentiles[i].flag2 = 0;
1200 for (idx = 0; idx < ft->n_valid; ++idx)
1202 static double prev_value = SYSMIS;
1203 f = &ft->valid[idx];
1205 for (i = 0; i < n_percentiles; i++)
1208 if ( percentiles[i].flag2 ) continue ;
1210 if ( settings_get_algorithm () != COMPATIBLE )
1212 (ft->valid_cases - 1) * percentiles[i].p;
1215 (ft->valid_cases + 1) * percentiles[i].p - 1;
1217 if ( percentiles[i].flag )
1219 percentiles[i].x2 = f->value.f;
1220 percentiles[i].x1 = prev_value;
1221 percentiles[i].flag2 = 1;
1227 if ( f->count > 1 && rank - (f->count - 1) > tp )
1229 percentiles[i].x2 = percentiles[i].x1 = f->value.f;
1230 percentiles[i].flag2 = 1;
1234 percentiles[i].flag=1;
1240 prev_value = f->value.f;
1243 for (i = 0; i < n_percentiles; i++)
1245 /* Catches the case when p == 100% */
1246 if ( ! percentiles[i].flag2 )
1247 percentiles[i].x1 = percentiles[i].x2 = f->value.f;
1250 printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
1251 i,percentiles[i].p,percentiles[i].x1,percentiles[i].x2);
1255 for (i = 0; i < n_percentiles; i++)
1257 struct freq_tab *ft = &get_var_freqs (v)->tab;
1261 if ( settings_get_algorithm () != COMPATIBLE )
1263 s = modf((ft->valid_cases - 1) * percentiles[i].p , &dummy);
1267 s = modf((ft->valid_cases + 1) * percentiles[i].p -1, &dummy);
1270 percentiles[i].value = percentiles[i].x1 +
1271 ( percentiles[i].x2 - percentiles[i].x1) * s ;
1275 /* Calculate the mode. */
1278 for (f = ft->valid; f < ft->missing; f++)
1280 if (most_often < f->count)
1282 most_often = f->count;
1283 X_mode = f->value.f;
1285 else if (most_often == f->count)
1287 /* A duplicate mode is undefined.
1288 FIXME: keep track of *all* the modes. */
1293 /* Calculate moments. */
1294 m = moments_create (MOMENT_KURTOSIS);
1295 for (f = ft->valid; f < ft->missing; f++)
1296 moments_pass_one (m, f->value.f, f->count);
1297 for (f = ft->valid; f < ft->missing; f++)
1298 moments_pass_two (m, f->value.f, f->count);
1299 moments_calculate (m, NULL, &d[frq_mean], &d[frq_variance],
1300 &d[frq_skew], &d[frq_kurt]);
1301 moments_destroy (m);
1303 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1304 d[frq_min] = ft->valid[0].value.f;
1305 d[frq_max] = ft->valid[ft->n_valid - 1].value.f;
1306 d[frq_mode] = X_mode;
1307 d[frq_range] = d[frq_max] - d[frq_min];
1308 d[frq_sum] = d[frq_mean] * W;
1309 d[frq_stddev] = sqrt (d[frq_variance]);
1310 d[frq_semean] = d[frq_stddev] / sqrt (W);
1311 d[frq_seskew] = calc_seskew (W);
1312 d[frq_sekurt] = calc_sekurt (W);
1315 /* Displays a table of all the statistics requested for variable V. */
1317 dump_statistics (const struct variable *v, bool show_varname,
1318 const struct variable *wv)
1320 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1321 struct freq_tab *ft;
1322 double stat_value[frq_n_stats];
1323 struct tab_table *t;
1326 if (var_is_alpha (v))
1328 ft = &get_var_freqs (v)->tab;
1329 if (ft->n_valid == 0)
1331 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1335 calc_stats (v, stat_value);
1337 t = tab_create (3, n_stats + n_show_percentiles + 2);
1339 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1342 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1343 tab_vline (t, TAL_GAP , 1, 0, tab_nr(t) - 1 ) ;
1345 r=2; /* N missing and N valid are always dumped */
1347 for (i = 0; i < frq_n_stats; i++)
1348 if (stats & BIT_INDEX (i))
1350 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1351 gettext (st_name[i].s10));
1352 tab_double (t, 2, r, TAB_NONE, stat_value[i], NULL);
1356 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1357 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1358 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1360 tab_double (t, 2, 0, TAB_NONE, ft->valid_cases, wfmt);
1361 tab_double (t, 2, 1, TAB_NONE, ft->total_cases - ft->valid_cases, wfmt);
1363 for (i = 0; i < n_percentiles; i++, r++)
1365 if (!percentiles[i].show)
1370 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1373 if (percentiles[i].p == 0.5)
1374 tab_text (t, 1, r, TAB_LEFT, _("50 (Median)"));
1376 tab_fixed (t, 1, r, TAB_LEFT, percentiles[i].p * 100, 3, 0);
1377 tab_double (t, 2, r, TAB_NONE, percentiles[i].value,
1378 var_get_print_format (v));
1382 tab_title (t, "%s", var_to_string (v));
1389 calculate_iqr (void)
1395 for (i = 0; i < n_percentiles; i++)
1397 if (fabs (0.25 - percentiles[i].p) < DBL_EPSILON)
1398 q1 = percentiles[i].value;
1399 else if (fabs (0.75 - percentiles[i].p) < DBL_EPSILON)
1400 q3 = percentiles[i].value;
1403 return q1 == SYSMIS || q3 == SYSMIS ? SYSMIS : q3 - q1;
1406 /* Create a gsl_histogram from a freq_tab */
1408 freq_tab_to_hist (const struct freq_tab *ft, const struct variable *var)
1411 double x_min = DBL_MAX;
1412 double x_max = -DBL_MAX;
1414 struct histogram *hist;
1418 struct hsh_iterator hi;
1419 struct hsh_table *fh = ft->data;
1420 struct freq_mutable *frq;
1422 /* Find out the extremes of the x value */
1423 for ( frq = hsh_first(fh, &hi); frq != 0; frq = hsh_next(fh, &hi) )
1425 if (var_is_value_missing(var, &frq->value, MV_ANY))
1428 if ( frq->value.f < x_min ) x_min = frq->value.f ;
1429 if ( frq->value.f > x_max ) x_max = frq->value.f ;
1432 /* Freedman-Diaconis' choice of bin width. */
1433 iqr = calculate_iqr ();
1436 double bin_width = 2 * iqr / pow (ft->valid_cases, 1.0 / 3.0);
1437 bins = (x_max - x_min) / bin_width;
1440 else if (bins > 400)
1446 hist = histogram_create (bins, x_min, x_max);
1448 for( i = 0 ; i < ft->n_valid ; ++i )
1450 frq = &ft->valid[i];
1451 histogram_add (hist, frq->value.f, frq->count);
1458 static struct slice *
1459 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1460 const struct variable *var,
1464 /* Allocate an array of slices and fill them from the data in frq_tab
1465 n_slices will contain the number of slices allocated.
1466 The caller is responsible for freeing slices
1468 static struct slice *
1469 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1470 const struct variable *var,
1474 struct slice *slices;
1476 *n_slices = frq_tab->n_valid;
1478 slices = xnmalloc (*n_slices, sizeof *slices);
1480 for (i = 0 ; i < *n_slices ; ++i )
1482 const struct freq_mutable *frq = &frq_tab->valid[i];
1484 ds_init_empty (&slices[i].label);
1485 var_append_value_name (var, &frq->value, &slices[i].label);
1486 slices[i].magnitude = frq->count;
1496 do_piechart(const struct variable *var, const struct freq_tab *frq_tab)
1498 struct slice *slices;
1501 slices = freq_tab_to_slice_array(frq_tab, var, &n_slices);
1503 chart_item_submit (piechart_create (var_to_string(var), slices, n_slices));
1505 for (i = 0 ; i < n_slices ; ++i )
1506 ds_destroy (&slices[i].label);