1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2007, 2009 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 */
151 static void add_percentile (double x) ;
153 static struct percentile *percentiles;
154 static int n_percentiles;
156 /* Groups of statistics. */
158 #define frq_default \
159 (BI (frq_mean) | BI (frq_stddev) | BI (frq_min) | BI (frq_max))
161 (BI (frq_sum) | BI(frq_min) | BI(frq_max) \
162 | BI(frq_mean) | BI(frq_semean) | BI(frq_stddev) \
163 | BI(frq_variance) | BI(frq_kurt) | BI(frq_sekurt) \
164 | BI(frq_skew) | BI(frq_seskew) | BI(frq_range) \
165 | BI(frq_range) | BI(frq_mode) | BI(frq_median))
167 /* Statistics; number of statistics. */
168 static unsigned long stats;
171 /* Types of graphs. */
174 GFT_NONE, /* Don't draw graphs. */
175 GFT_BAR, /* Draw bar charts. */
176 GFT_HIST, /* Draw histograms. */
177 GFT_PIE, /* Draw piechart */
178 GFT_HBAR /* Draw bar charts or histograms at our discretion. */
181 /* Parsed command. */
182 static struct cmd_frequencies cmd;
184 /* Summary of the barchart, histogram, and hbar subcommands. */
185 /* FIXME: These should not be mututally exclusive */
186 static int chart; /* NONE/BAR/HIST/HBAR/PIE. */
187 static double min, max; /* Minimum, maximum on y axis. */
188 static int format; /* FREQ/PERCENT: Scaling of y axis. */
189 static double scale, incr; /* FIXME */
190 static int normal; /* FIXME */
192 /* Variables for which to calculate statistics. */
193 static size_t n_variables;
194 static const struct variable **v_variables;
197 static struct pool *data_pool; /* For per-SPLIT FILE group data. */
198 static struct pool *syntax_pool; /* For syntax-related data. */
200 /* Frequency tables. */
202 /* Entire frequency table. */
205 struct hsh_table *data; /* Undifferentiated data. */
206 struct freq_mutable *valid; /* Valid freqs. */
207 int n_valid; /* Number of total freqs. */
208 const struct dictionary *dict; /* The dict from whence entries in the table
211 struct freq_mutable *missing; /* Missing freqs. */
212 int n_missing; /* Number of missing freqs. */
215 double total_cases; /* Sum of weights of all cases. */
216 double valid_cases; /* Sum of weights of valid cases. */
220 /* Per-variable frequency data. */
223 /* Freqency table. */
224 struct freq_tab tab; /* Frequencies table to use. */
227 int n_groups; /* Number of groups. */
228 double *groups; /* Groups. */
231 double stat[frq_n_stats];
233 /* Variable attributes. */
235 struct fmt_spec print;
238 static inline struct var_freqs *
239 get_var_freqs (const struct variable *v)
241 return var_get_aux (v);
244 static void determine_charts (void);
246 static void calc_stats (const struct variable *v, double d[frq_n_stats]);
248 static void precalc (struct casereader *, struct dataset *);
249 static void calc (const struct ccase *, const struct dataset *);
250 static void postcalc (const struct dataset *);
252 static void postprocess_freq_tab (const struct variable *);
253 static void dump_full ( const struct variable *, const struct variable *);
254 static void dump_condensed (const struct variable *, const struct variable *);
255 static void dump_statistics (const struct variable *, bool show_varname, const struct variable *);
256 static void cleanup_freq_tab (const struct variable *);
258 static hsh_compare_func compare_value_numeric_a, compare_value_alpha_a;
259 static hsh_compare_func compare_value_numeric_d, compare_value_alpha_d;
260 static hsh_compare_func compare_freq_numeric_a, compare_freq_alpha_a;
261 static hsh_compare_func compare_freq_numeric_d, compare_freq_alpha_d;
264 static void do_piechart(const struct variable *var,
265 const struct freq_tab *frq_tab);
268 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var);
272 /* Parser and outline. */
274 static int internal_cmd_frequencies (struct lexer *lexer, struct dataset *ds);
277 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
281 syntax_pool = pool_create ();
282 result = internal_cmd_frequencies (lexer, ds);
283 pool_destroy (syntax_pool);
285 pool_destroy (data_pool);
293 internal_cmd_frequencies (struct lexer *lexer, struct dataset *ds)
295 struct casegrouper *grouper;
296 struct casereader *input, *group;
306 if (!parse_frequencies (lexer, ds, &cmd, NULL))
309 if (cmd.onepage_limit == LONG_MIN)
310 cmd.onepage_limit = 50;
312 /* Figure out statistics to calculate. */
314 if (cmd.a_statistics[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
315 stats |= frq_default;
316 if (cmd.a_statistics[FRQ_ST_ALL])
318 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
319 stats &= ~BIT_INDEX (frq_median);
320 for (i = 0; i < frq_n_stats; i++)
321 if (cmd.a_statistics[st_name[i].st_indx])
322 stats |= BIT_INDEX (i);
323 if (stats & frq_kurt)
324 stats |= BIT_INDEX (frq_sekurt);
325 if (stats & frq_skew)
326 stats |= BIT_INDEX (frq_seskew);
328 /* Calculate n_stats. */
330 for (i = 0; i < frq_n_stats; i++)
331 if ((stats & BIT_INDEX (i)))
336 if (chart != GFT_NONE || cmd.sbc_ntiles)
337 cmd.sort = FRQ_AVALUE;
339 /* Work out what percentiles need to be calculated */
340 if ( cmd.sbc_percentiles )
342 for ( i = 0 ; i < MAXLISTS ; ++i )
345 subc_list_double *ptl_list = &cmd.dl_percentiles[i];
346 for ( pl = 0 ; pl < subc_list_double_count(ptl_list); ++pl)
347 add_percentile (subc_list_double_at(ptl_list, pl) / 100.0 );
350 if ( cmd.sbc_ntiles )
352 for ( i = 0 ; i < cmd.sbc_ntiles ; ++i )
355 for (j = 0; j <= cmd.n_ntiles[i]; ++j )
356 add_percentile (j / (double) cmd.n_ntiles[i]);
359 if (stats & BIT_INDEX (frq_median))
361 /* Treat the median as the 50% percentile.
362 We output it in the percentiles table as "50 (Median)." */
363 add_percentile (0.5);
364 stats &= ~BIT_INDEX (frq_median);
369 input = casereader_create_filter_weight (proc_open (ds), dataset_dict (ds),
371 grouper = casegrouper_create_splits (input, dataset_dict (ds));
372 for (; casegrouper_get_next_group (grouper, &group);
373 casereader_destroy (group))
378 for (; (c = casereader_read (group)) != NULL; case_unref (c))
382 ok = casegrouper_destroy (grouper);
383 ok = proc_commit (ds) && ok;
385 free_frequencies(&cmd);
387 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
390 /* Figure out which charts the user requested. */
392 determine_charts (void)
394 int count = (!!cmd.sbc_histogram) + (!!cmd.sbc_barchart) +
395 (!!cmd.sbc_hbar) + (!!cmd.sbc_piechart);
405 msg (SW, _("At most one of BARCHART, HISTOGRAM, or HBAR should be "
406 "given. HBAR will be assumed. Argument values will be "
407 "given precedence increasing along the order given."));
409 else if (cmd.sbc_histogram)
411 else if (cmd.sbc_barchart)
413 else if (cmd.sbc_piechart)
424 if (cmd.sbc_barchart)
426 if (cmd.ba_min != SYSMIS)
428 if (cmd.ba_max != SYSMIS)
430 if (cmd.ba_scale == FRQ_FREQ)
435 else if (cmd.ba_scale == FRQ_PERCENT)
437 format = FRQ_PERCENT;
442 if (cmd.sbc_histogram)
444 if (cmd.hi_min != SYSMIS)
446 if (cmd.hi_max != SYSMIS)
448 if (cmd.hi_scale == FRQ_FREQ)
453 else if (cmd.hi_scale == FRQ_PERCENT)
455 format = FRQ_PERCENT;
458 if (cmd.hi_norm != FRQ_NONORMAL )
460 if (cmd.hi_incr == FRQ_INCREMENT)
466 if (cmd.hb_min != SYSMIS)
468 if (cmd.hb_max != SYSMIS)
470 if (cmd.hb_scale == FRQ_FREQ)
475 else if (cmd.hb_scale == FRQ_PERCENT)
477 format = FRQ_PERCENT;
482 if (cmd.hb_incr == FRQ_INCREMENT)
486 if (min != SYSMIS && max != SYSMIS && min >= max)
488 msg (SE, _("MAX must be greater than or equal to MIN, if both are "
489 "specified. However, MIN was specified as %g and MAX as %g. "
490 "MIN and MAX will be ignored."), min, max);
495 /* Add data from case C to the frequency table. */
497 calc (const struct ccase *c, const struct dataset *ds)
499 double weight = dict_get_case_weight (dataset_dict (ds), c, NULL);
502 for (i = 0; i < n_variables; i++)
504 const struct variable *v = v_variables[i];
505 const union value *val = case_data (c, v);
506 struct var_freqs *vf = get_var_freqs (v);
507 struct freq_tab *ft = &vf->tab;
509 struct freq_mutable target;
510 struct freq_mutable **fpp;
513 fpp = (struct freq_mutable **) hsh_probe (ft->data, &target);
516 (*fpp)->count += weight;
519 struct freq_mutable *fp = pool_alloc (data_pool, sizeof *fp);
521 value_init_pool (data_pool, &fp->value, vf->width);
522 value_copy (&fp->value, val, vf->width);
528 /* Prepares each variable that is the target of FREQUENCIES by setting
529 up its hash table. */
531 precalc (struct casereader *input, struct dataset *ds)
536 c = casereader_peek (input, 0);
539 output_split_file_values (ds, c);
543 pool_destroy (data_pool);
544 data_pool = pool_create ();
546 for (i = 0; i < n_variables; i++)
548 const struct variable *v = v_variables[i];
549 struct freq_tab *ft = &get_var_freqs (v)->tab;
551 ft->data = hsh_create (16, compare_freq, hash_freq, NULL, v);
555 /* Finishes up with the variables after frequencies have been
556 calculated. Displays statistics, percentiles, ... */
558 postcalc (const struct dataset *ds)
560 const struct dictionary *dict = dataset_dict (ds);
561 const struct variable *wv = dict_get_weight (dict);
564 for (i = 0; i < n_variables; i++)
566 const struct variable *v = v_variables[i];
567 struct var_freqs *vf = get_var_freqs (v);
568 struct freq_tab *ft = &vf->tab;
570 int dumped_freq_tab = 1;
572 postprocess_freq_tab (v);
574 /* Frequencies tables. */
575 n_categories = ft->n_valid + ft->n_missing;
576 if (cmd.table == FRQ_TABLE
577 || (cmd.table == FRQ_LIMIT && n_categories <= cmd.limit))
581 dump_condensed (v, wv);
587 if (n_categories > cmd.onepage_limit)
588 dump_condensed (v, wv);
600 dump_statistics (v, !dumped_freq_tab, wv);
604 if ( chart == GFT_HIST && var_is_numeric (v) )
606 double d[frq_n_stats];
607 struct histogram *hist ;
611 hist = freq_tab_to_hist (ft,v);
613 chart_item_submit (histogram_chart_create (
614 hist->gsl_hist, var_to_string(v),
620 statistic_destroy (&hist->parent);
623 if ( chart == GFT_PIE)
625 do_piechart(v_variables[i], ft);
628 cleanup_freq_tab (v);
633 /* Returns the comparison function that should be used for
634 sorting a frequency table by FRQ_SORT using VAL_TYPE
636 static hsh_compare_func *
637 get_freq_comparator (int frq_sort, enum val_type val_type)
639 bool is_numeric = val_type == VAL_NUMERIC;
643 return is_numeric ? compare_value_numeric_a : compare_value_alpha_a;
645 return is_numeric ? compare_value_numeric_d : compare_value_alpha_d;
647 return is_numeric ? compare_freq_numeric_a : compare_freq_alpha_a;
649 return is_numeric ? compare_freq_numeric_d : compare_freq_alpha_d;
655 /* Returns true iff the value in struct freq_mutable F is non-missing
658 not_missing (const void *f_, const void *v_)
660 const struct freq_mutable *f = f_;
661 const struct variable *v = v_;
663 return !var_is_value_missing (v, &f->value, MV_ANY);
666 /* Summarizes the frequency table data for variable V. */
668 postprocess_freq_tab (const struct variable *v)
670 hsh_compare_func *compare;
674 struct freq_mutable *freqs, *f;
677 ft = &get_var_freqs (v)->tab;
678 compare = get_freq_comparator (cmd.sort, var_get_type (v));
680 /* Extract data from hash table. */
681 count = hsh_count (ft->data);
682 data = hsh_data (ft->data);
684 /* Copy dereferenced data into freqs. */
685 freqs = xnmalloc (count, sizeof *freqs);
686 for (i = 0; i < count; i++)
688 struct freq_mutable *f = data[i];
692 /* Put data into ft. */
694 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, v);
695 ft->missing = freqs + ft->n_valid;
696 ft->n_missing = count - ft->n_valid;
699 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare, v);
700 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, v);
702 /* Summary statistics. */
703 ft->valid_cases = 0.0;
704 for(i = 0 ; i < ft->n_valid ; ++i )
707 ft->valid_cases += f->count;
711 ft->total_cases = ft->valid_cases ;
712 for(i = 0 ; i < ft->n_missing ; ++i )
715 ft->total_cases += f->count;
720 /* Frees the frequency table for variable V. */
722 cleanup_freq_tab (const struct variable *v)
724 struct freq_tab *ft = &get_var_freqs (v)->tab;
726 hsh_destroy (ft->data);
729 /* Parses the VARIABLES subcommand, adding to
730 {n_variables,v_variables}. */
732 frq_custom_variables (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *aux UNUSED)
734 size_t old_n_variables = n_variables;
737 lex_match (lexer, '=');
738 if (lex_token (lexer) != T_ALL && (lex_token (lexer) != T_ID
739 || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) == NULL))
742 if (!parse_variables_const (lexer, dataset_dict (ds), &v_variables, &n_variables,
743 PV_APPEND | PV_NO_SCRATCH))
746 for (i = old_n_variables; i < n_variables; i++)
748 const struct variable *v = v_variables[i];
749 struct var_freqs *vf;
751 if (var_get_aux (v) != NULL)
753 msg (SE, _("Variable %s specified multiple times on VARIABLES "
754 "subcommand."), var_get_name (v));
757 vf = var_attach_aux (v, xmalloc (sizeof *vf), var_dtor_free);
758 vf->tab.valid = vf->tab.missing = NULL;
759 vf->tab.dict = dataset_dict (ds);
762 vf->width = var_get_width (v);
763 vf->print = *var_get_print_format (v);
768 /* Parses the GROUPED subcommand, setting the n_grouped, grouped
769 fields of specified variables. */
771 frq_custom_grouped (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *aux UNUSED)
773 lex_match (lexer, '=');
774 if ((lex_token (lexer) == T_ID && dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
775 || lex_token (lexer) == T_ID)
780 /* Max, current size of list; list itself. */
786 const struct variable **v;
788 if (!parse_variables_const (lexer, dataset_dict (ds), &v, &n,
789 PV_NO_DUPLICATE | PV_NUMERIC))
791 if (lex_match (lexer, '('))
795 while (lex_integer (lexer))
800 dl = pool_nrealloc (syntax_pool, dl, ml, sizeof *dl);
802 dl[nl++] = lex_tokval (lexer);
804 lex_match (lexer, ',');
806 /* Note that nl might still be 0 and dl might still be
807 NULL. That's okay. */
808 if (!lex_match (lexer, ')'))
811 msg (SE, _("`)' expected after GROUPED interval list."));
821 for (i = 0; i < n; i++)
822 if (var_get_aux (v[i]) == NULL)
823 msg (SE, _("Variables %s specified on GROUPED but not on "
824 "VARIABLES."), var_get_name (v[i]));
827 struct var_freqs *vf = get_var_freqs (v[i]);
829 if (vf->groups != NULL)
830 msg (SE, _("Variables %s specified multiple times on GROUPED "
831 "subcommand."), var_get_name (v[i]));
839 if (!lex_match (lexer, '/'))
841 if ((lex_token (lexer) != T_ID || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
842 && lex_token (lexer) != T_ALL)
844 lex_put_back (lexer, '/');
852 /* Adds X to the list of percentiles, keeping the list in proper
855 add_percentile (double x)
859 for (i = 0; i < n_percentiles; i++)
861 /* Do nothing if it's already in the list */
862 if ( fabs(x - percentiles[i].p) < DBL_EPSILON )
865 if (x < percentiles[i].p)
869 if (i >= n_percentiles || x != percentiles[i].p)
871 percentiles = pool_nrealloc (syntax_pool, percentiles,
872 n_percentiles + 1, sizeof *percentiles);
873 insert_element (percentiles, n_percentiles, sizeof *percentiles, i);
874 percentiles[i].p = x;
879 /* Comparison functions. */
881 /* Ascending numeric compare of values. */
883 compare_value_numeric_a (const void *a_, const void *b_, const void *aux UNUSED)
885 const struct freq_mutable *a = a_;
886 const struct freq_mutable *b = b_;
888 if (a->value.f > b->value.f)
890 else if (a->value.f < b->value.f)
896 /* Ascending string compare of values. */
898 compare_value_alpha_a (const void *a_, const void *b_, const void *v_)
900 const struct freq_mutable *a = a_;
901 const struct freq_mutable *b = b_;
902 const struct variable *v = v_;
903 struct var_freqs *vf = get_var_freqs (v);
905 return value_compare_3way (&a->value, &b->value, vf->width);
908 /* Descending numeric compare of values. */
910 compare_value_numeric_d (const void *a, const void *b, const void *aux UNUSED)
912 return -compare_value_numeric_a (a, b, aux);
915 /* Descending string compare of values. */
917 compare_value_alpha_d (const void *a, const void *b, const void *v)
919 return -compare_value_alpha_a (a, b, v);
922 /* Ascending numeric compare of frequency;
923 secondary key on ascending numeric value. */
925 compare_freq_numeric_a (const void *a_, const void *b_, const void *aux UNUSED)
927 const struct freq_mutable *a = a_;
928 const struct freq_mutable *b = b_;
930 if (a->count > b->count)
932 else if (a->count < b->count)
935 if (a->value.f > b->value.f)
937 else if (a->value.f < b->value.f)
943 /* Ascending numeric compare of frequency;
944 secondary key on ascending string value. */
946 compare_freq_alpha_a (const void *a_, const void *b_, const void *v_)
948 const struct freq_mutable *a = a_;
949 const struct freq_mutable *b = b_;
950 const struct variable *v = v_;
951 struct var_freqs *vf = get_var_freqs (v);
953 if (a->count > b->count)
955 else if (a->count < b->count)
958 return value_compare_3way (&a->value, &b->value, vf->width);
961 /* Descending numeric compare of frequency;
962 secondary key on ascending numeric value. */
964 compare_freq_numeric_d (const void *a_, const void *b_, const void *aux UNUSED)
966 const struct freq_mutable *a = a_;
967 const struct freq_mutable *b = b_;
969 if (a->count > b->count)
971 else if (a->count < b->count)
974 if (a->value.f > b->value.f)
976 else if (a->value.f < b->value.f)
982 /* Descending numeric compare of frequency;
983 secondary key on ascending string value. */
985 compare_freq_alpha_d (const void *a_, const void *b_, const void *v_)
987 const struct freq_mutable *a = a_;
988 const struct freq_mutable *b = b_;
989 const struct variable *v = v_;
990 struct var_freqs *vf = get_var_freqs (v);
992 if (a->count > b->count)
994 else if (a->count < b->count)
997 return value_compare_3way (&a->value, &b->value, vf->width);
1000 /* Frequency table display. */
1002 /* Displays a full frequency table for variable V. */
1004 dump_full (const struct variable *v, const struct variable *wv)
1006 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1008 struct var_freqs *vf;
1009 struct freq_tab *ft;
1010 struct freq_mutable *f;
1011 struct tab_table *t;
1013 double cum_total = 0.0;
1014 double cum_freq = 0.0;
1016 static const char *headings[] = {
1020 N_("Valid Percent"),
1024 const bool lab = (cmd.labels == FRQ_LABELS);
1026 vf = get_var_freqs (v);
1028 n_categories = ft->n_valid + ft->n_missing;
1029 t = tab_create (5 + lab, n_categories + 2);
1030 tab_headers (t, 0, 0, 1, 0);
1033 tab_text (t, 0, 0, TAB_CENTER | TAT_TITLE, _("Value Label"));
1035 for (x = 0; x < 5; x++)
1036 tab_text (t, lab + x, 0, TAB_CENTER | TAT_TITLE, gettext (headings[x]));
1039 for (f = ft->valid; f < ft->missing; f++)
1041 double percent, valid_percent;
1043 cum_freq += f->count;
1045 percent = f->count / ft->total_cases * 100.0;
1046 valid_percent = f->count / ft->valid_cases * 100.0;
1047 cum_total += valid_percent;
1051 const char *label = var_lookup_value_label (v, &f->value);
1053 tab_text (t, 0, r, TAB_LEFT, label);
1056 tab_value (t, 0 + lab, r, TAB_NONE, &f->value, ft->dict, &vf->print);
1057 tab_double (t, 1 + lab, r, TAB_NONE, f->count, wfmt);
1058 tab_double (t, 2 + lab, r, TAB_NONE, percent, NULL);
1059 tab_double (t, 3 + lab, r, TAB_NONE, valid_percent, NULL);
1060 tab_double (t, 4 + lab, r, TAB_NONE, cum_total, NULL);
1063 for (; f < &ft->valid[n_categories]; f++)
1065 cum_freq += f->count;
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,
1077 f->count / ft->total_cases * 100.0, NULL);
1078 tab_text (t, 3 + lab, r, TAB_NONE, _("Missing"));
1082 tab_box (t, TAL_1, TAL_1,
1083 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1085 tab_hline (t, TAL_2, 0, 4 + lab, 1);
1086 tab_hline (t, TAL_2, 0, 4 + lab, r);
1087 tab_joint_text (t, 0, r, 0 + lab, r, TAB_RIGHT | TAT_TITLE, _("Total"));
1088 tab_vline (t, TAL_0, 1, r, r);
1089 tab_double (t, 1 + lab, r, TAB_NONE, cum_freq, wfmt);
1090 tab_fixed (t, 2 + lab, r, TAB_NONE, 100.0, 5, 1);
1091 tab_fixed (t, 3 + lab, r, TAB_NONE, 100.0, 5, 1);
1093 tab_title (t, "%s", var_to_string (v));
1097 /* Display condensed frequency table for variable V. */
1099 dump_condensed (const struct variable *v, const struct variable *wv)
1101 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1103 struct var_freqs *vf;
1104 struct freq_tab *ft;
1105 struct freq_mutable *f;
1106 struct tab_table *t;
1108 double cum_total = 0.0;
1110 vf = get_var_freqs (v);
1112 n_categories = ft->n_valid + ft->n_missing;
1113 t = tab_create (4, n_categories + 2);
1115 tab_headers (t, 0, 0, 2, 0);
1116 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value"));
1117 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Freq"));
1118 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1119 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Cum"));
1120 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1123 for (f = ft->valid; f < ft->missing; f++)
1127 percent = f->count / ft->total_cases * 100.0;
1128 cum_total += f->count / ft->valid_cases * 100.0;
1130 tab_value (t, 0, r, TAB_NONE, &f->value, ft->dict, &vf->print);
1131 tab_double (t, 1, r, TAB_NONE, f->count, wfmt);
1132 tab_double (t, 2, r, TAB_NONE, percent, NULL);
1133 tab_double (t, 3, r, TAB_NONE, cum_total, NULL);
1136 for (; f < &ft->valid[n_categories]; f++)
1138 tab_value (t, 0, r, TAB_NONE, &f->value, ft->dict, &vf->print);
1139 tab_double (t, 1, r, TAB_NONE, f->count, wfmt);
1140 tab_double (t, 2, r, TAB_NONE,
1141 f->count / ft->total_cases * 100.0, NULL);
1145 tab_box (t, TAL_1, TAL_1,
1146 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1148 tab_hline (t, TAL_2, 0, 3, 2);
1149 tab_title (t, "%s", var_to_string (v));
1153 /* Statistical display. */
1155 /* Calculates all the pertinent statistics for variable V, putting
1156 them in array D[]. FIXME: This could be made much more optimal. */
1158 calc_stats (const struct variable *v, double d[frq_n_stats])
1160 struct freq_tab *ft = &get_var_freqs (v)->tab;
1161 double W = ft->valid_cases;
1163 struct freq_mutable *f=0;
1171 /* Calculate percentiles. */
1173 for (i = 0; i < n_percentiles; i++)
1175 percentiles[i].flag = 0;
1176 percentiles[i].flag2 = 0;
1180 for (idx = 0; idx < ft->n_valid; ++idx)
1182 static double prev_value = SYSMIS;
1183 f = &ft->valid[idx];
1185 for (i = 0; i < n_percentiles; i++)
1188 if ( percentiles[i].flag2 ) continue ;
1190 if ( settings_get_algorithm () != COMPATIBLE )
1192 (ft->valid_cases - 1) * percentiles[i].p;
1195 (ft->valid_cases + 1) * percentiles[i].p - 1;
1197 if ( percentiles[i].flag )
1199 percentiles[i].x2 = f->value.f;
1200 percentiles[i].x1 = prev_value;
1201 percentiles[i].flag2 = 1;
1207 if ( f->count > 1 && rank - (f->count - 1) > tp )
1209 percentiles[i].x2 = percentiles[i].x1 = f->value.f;
1210 percentiles[i].flag2 = 1;
1214 percentiles[i].flag=1;
1220 prev_value = f->value.f;
1223 for (i = 0; i < n_percentiles; i++)
1225 /* Catches the case when p == 100% */
1226 if ( ! percentiles[i].flag2 )
1227 percentiles[i].x1 = percentiles[i].x2 = f->value.f;
1230 printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
1231 i,percentiles[i].p,percentiles[i].x1,percentiles[i].x2);
1235 for (i = 0; i < n_percentiles; i++)
1237 struct freq_tab *ft = &get_var_freqs (v)->tab;
1241 if ( settings_get_algorithm () != COMPATIBLE )
1243 s = modf((ft->valid_cases - 1) * percentiles[i].p , &dummy);
1247 s = modf((ft->valid_cases + 1) * percentiles[i].p -1, &dummy);
1250 percentiles[i].value = percentiles[i].x1 +
1251 ( percentiles[i].x2 - percentiles[i].x1) * s ;
1255 /* Calculate the mode. */
1258 for (f = ft->valid; f < ft->missing; f++)
1260 if (most_often < f->count)
1262 most_often = f->count;
1263 X_mode = f->value.f;
1265 else if (most_often == f->count)
1267 /* A duplicate mode is undefined.
1268 FIXME: keep track of *all* the modes. */
1273 /* Calculate moments. */
1274 m = moments_create (MOMENT_KURTOSIS);
1275 for (f = ft->valid; f < ft->missing; f++)
1276 moments_pass_one (m, f->value.f, f->count);
1277 for (f = ft->valid; f < ft->missing; f++)
1278 moments_pass_two (m, f->value.f, f->count);
1279 moments_calculate (m, NULL, &d[frq_mean], &d[frq_variance],
1280 &d[frq_skew], &d[frq_kurt]);
1281 moments_destroy (m);
1283 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1284 d[frq_min] = ft->valid[0].value.f;
1285 d[frq_max] = ft->valid[ft->n_valid - 1].value.f;
1286 d[frq_mode] = X_mode;
1287 d[frq_range] = d[frq_max] - d[frq_min];
1288 d[frq_sum] = d[frq_mean] * W;
1289 d[frq_stddev] = sqrt (d[frq_variance]);
1290 d[frq_semean] = d[frq_stddev] / sqrt (W);
1291 d[frq_seskew] = calc_seskew (W);
1292 d[frq_sekurt] = calc_sekurt (W);
1295 /* Displays a table of all the statistics requested for variable V. */
1297 dump_statistics (const struct variable *v, bool show_varname,
1298 const struct variable *wv)
1300 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1301 struct freq_tab *ft;
1302 double stat_value[frq_n_stats];
1303 struct tab_table *t;
1306 if (var_is_alpha (v))
1308 ft = &get_var_freqs (v)->tab;
1309 if (ft->n_valid == 0)
1311 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1315 calc_stats (v, stat_value);
1317 t = tab_create (3, n_stats + n_percentiles + 2);
1319 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1322 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1323 tab_vline (t, TAL_GAP , 1, 0, tab_nr(t) - 1 ) ;
1325 r=2; /* N missing and N valid are always dumped */
1327 for (i = 0; i < frq_n_stats; i++)
1328 if (stats & BIT_INDEX (i))
1330 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1331 gettext (st_name[i].s10));
1332 tab_double (t, 2, r, TAB_NONE, stat_value[i], NULL);
1336 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1337 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1338 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1340 tab_double (t, 2, 0, TAB_NONE, ft->valid_cases, wfmt);
1341 tab_double (t, 2, 1, TAB_NONE, ft->total_cases - ft->valid_cases, wfmt);
1343 for (i = 0; i < n_percentiles; i++, r++)
1347 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1350 if (percentiles[i].p == 0.5)
1351 tab_text (t, 1, r, TAB_LEFT, _("50 (Median)"));
1353 tab_fixed (t, 1, r, TAB_LEFT, percentiles[i].p * 100, 3, 0);
1354 tab_double (t, 2, r, TAB_NONE, percentiles[i].value,
1355 var_get_print_format (v));
1359 tab_title (t, "%s", var_to_string (v));
1366 /* Create a gsl_histogram from a freq_tab */
1368 freq_tab_to_hist (const struct freq_tab *ft, const struct variable *var)
1371 double x_min = DBL_MAX;
1372 double x_max = -DBL_MAX;
1374 struct histogram *hist;
1375 const double bins = 11;
1377 struct hsh_iterator hi;
1378 struct hsh_table *fh = ft->data;
1379 struct freq_mutable *frq;
1381 /* Find out the extremes of the x value */
1382 for ( frq = hsh_first(fh, &hi); frq != 0; frq = hsh_next(fh, &hi) )
1384 if (var_is_value_missing(var, &frq->value, MV_ANY))
1387 if ( frq->value.f < x_min ) x_min = frq->value.f ;
1388 if ( frq->value.f > x_max ) x_max = frq->value.f ;
1391 hist = histogram_create (bins, x_min, x_max);
1393 for( i = 0 ; i < ft->n_valid ; ++i )
1395 frq = &ft->valid[i];
1396 histogram_add (hist, frq->value.f, frq->count);
1403 static struct slice *
1404 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1405 const struct variable *var,
1409 /* Allocate an array of slices and fill them from the data in frq_tab
1410 n_slices will contain the number of slices allocated.
1411 The caller is responsible for freeing slices
1413 static struct slice *
1414 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1415 const struct variable *var,
1419 struct slice *slices;
1421 *n_slices = frq_tab->n_valid;
1423 slices = xnmalloc (*n_slices, sizeof *slices);
1425 for (i = 0 ; i < *n_slices ; ++i )
1427 const struct freq_mutable *frq = &frq_tab->valid[i];
1429 ds_init_empty (&slices[i].label);
1430 var_append_value_name (var, &frq->value, &slices[i].label);
1431 slices[i].magnitude = frq->count;
1441 do_piechart(const struct variable *var, const struct freq_tab *frq_tab)
1443 struct slice *slices;
1446 slices = freq_tab_to_slice_array(frq_tab, var, &n_slices);
1448 chart_item_submit (piechart_create (var_to_string(var), slices, n_slices));
1450 for (i = 0 ; i < n_slices ; ++i )
1451 ds_destroy (&slices[i].label);