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.h>
52 #include <output/charts/piechart.h>
53 #include <output/charts/plot-hist.h>
54 #include <output/manager.h>
55 #include <output/output.h>
56 #include <output/table.h>
64 #define _(msgid) gettext (msgid)
65 #define N_(msgid) msgid
72 +format=cond:condense/onepage(*n:onepage_limit,"%s>=0")/!standard,
73 table:limit(n:limit,"%s>0")/notable/!table,
74 labels:!labels/nolabels,
75 sort:!avalue/dvalue/afreq/dfreq,
76 spaces:!single/double,
77 paging:newpage/!oldpage;
78 missing=miss:include/!exclude;
79 barchart(ba_)=:minimum(d:min),
81 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0");
82 piechart(pie_)=:minimum(d:min),
84 missing:missing/!nomissing;
85 histogram(hi_)=:minimum(d:min),
87 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
88 norm:!nonormal/normal,
89 incr:increment(d:inc,"%s>0");
90 hbar(hb_)=:minimum(d:min),
92 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
93 norm:!nonormal/normal,
94 incr:increment(d:inc,"%s>0");
97 +percentiles = double list;
98 +statistics[st_]=1|mean,2|semean,3|median,4|mode,5|stddev,6|variance,
99 7|kurtosis,8|skewness,9|range,10|minimum,11|maximum,12|sum,
100 13|default,14|seskewness,15|sekurtosis,all,none.
108 frq_mean = 0, frq_semean, frq_median, frq_mode, frq_stddev, frq_variance,
109 frq_kurt, frq_sekurt, frq_skew, frq_seskew, frq_range, frq_min, frq_max,
113 /* Description of a statistic. */
116 int st_indx; /* Index into a_statistics[]. */
117 const char *s10; /* Identifying string. */
120 /* Table of statistics, indexed by dsc_*. */
121 static const struct frq_info st_name[frq_n_stats + 1] =
123 {FRQ_ST_MEAN, N_("Mean")},
124 {FRQ_ST_SEMEAN, N_("S.E. Mean")},
125 {FRQ_ST_MEDIAN, N_("Median")},
126 {FRQ_ST_MODE, N_("Mode")},
127 {FRQ_ST_STDDEV, N_("Std Dev")},
128 {FRQ_ST_VARIANCE, N_("Variance")},
129 {FRQ_ST_KURTOSIS, N_("Kurtosis")},
130 {FRQ_ST_SEKURTOSIS, N_("S.E. Kurt")},
131 {FRQ_ST_SKEWNESS, N_("Skewness")},
132 {FRQ_ST_SESKEWNESS, N_("S.E. Skew")},
133 {FRQ_ST_RANGE, N_("Range")},
134 {FRQ_ST_MINIMUM, N_("Minimum")},
135 {FRQ_ST_MAXIMUM, N_("Maximum")},
136 {FRQ_ST_SUM, N_("Sum")},
140 /* Percentiles to calculate. */
144 double p; /* the %ile to be calculated */
145 double value; /* the %ile's value */
146 double x1; /* The datum value <= the percentile */
147 double x2; /* The datum value >= the percentile */
149 int flag2; /* Set to 1 if this percentile value has been found */
153 static void add_percentile (double x) ;
155 static struct percentile *percentiles;
156 static int n_percentiles;
158 /* Groups of statistics. */
160 #define frq_default \
161 (BI (frq_mean) | BI (frq_stddev) | BI (frq_min) | BI (frq_max))
163 (BI (frq_sum) | BI(frq_min) | BI(frq_max) \
164 | BI(frq_mean) | BI(frq_semean) | BI(frq_stddev) \
165 | BI(frq_variance) | BI(frq_kurt) | BI(frq_sekurt) \
166 | BI(frq_skew) | BI(frq_seskew) | BI(frq_range) \
167 | BI(frq_range) | BI(frq_mode) | BI(frq_median))
169 /* Statistics; number of statistics. */
170 static unsigned long stats;
173 /* Types of graphs. */
176 GFT_NONE, /* Don't draw graphs. */
177 GFT_BAR, /* Draw bar charts. */
178 GFT_HIST, /* Draw histograms. */
179 GFT_PIE, /* Draw piechart */
180 GFT_HBAR /* Draw bar charts or histograms at our discretion. */
183 /* Parsed command. */
184 static struct cmd_frequencies cmd;
186 /* Summary of the barchart, histogram, and hbar subcommands. */
187 /* FIXME: These should not be mututally exclusive */
188 static int chart; /* NONE/BAR/HIST/HBAR/PIE. */
189 static double min, max; /* Minimum, maximum on y axis. */
190 static int format; /* FREQ/PERCENT: Scaling of y axis. */
191 static double scale, incr; /* FIXME */
192 static int normal; /* FIXME */
194 /* Variables for which to calculate statistics. */
195 static size_t n_variables;
196 static const struct variable **v_variables;
199 static struct pool *data_pool; /* For per-SPLIT FILE group data. */
200 static struct pool *syntax_pool; /* For syntax-related data. */
202 /* Frequency tables. */
204 /* Entire frequency table. */
207 struct hsh_table *data; /* Undifferentiated data. */
208 struct freq_mutable *valid; /* Valid freqs. */
209 int n_valid; /* Number of total freqs. */
210 const struct dictionary *dict; /* The dict from whence entries in the table
213 struct freq_mutable *missing; /* Missing freqs. */
214 int n_missing; /* Number of missing freqs. */
217 double total_cases; /* Sum of weights of all cases. */
218 double valid_cases; /* Sum of weights of valid cases. */
222 /* Per-variable frequency data. */
225 /* Freqency table. */
226 struct freq_tab tab; /* Frequencies table to use. */
229 int n_groups; /* Number of groups. */
230 double *groups; /* Groups. */
233 double stat[frq_n_stats];
235 /* Variable attributes. */
237 struct fmt_spec print;
240 static inline struct var_freqs *
241 get_var_freqs (const struct variable *v)
243 return var_get_aux (v);
246 static void determine_charts (void);
248 static void calc_stats (const struct variable *v, double d[frq_n_stats]);
250 static void precalc (struct casereader *, struct dataset *);
251 static void calc (const struct ccase *, const struct dataset *);
252 static void postcalc (const struct dataset *);
254 static void postprocess_freq_tab (const struct variable *);
255 static void dump_full ( const struct variable *, const struct variable *);
256 static void dump_condensed (const struct variable *, const struct variable *);
257 static void dump_statistics (const struct variable *, bool show_varname, const struct variable *);
258 static void cleanup_freq_tab (const struct variable *);
260 static hsh_compare_func compare_value_numeric_a, compare_value_alpha_a;
261 static hsh_compare_func compare_value_numeric_d, compare_value_alpha_d;
262 static hsh_compare_func compare_freq_numeric_a, compare_freq_alpha_a;
263 static hsh_compare_func compare_freq_numeric_d, compare_freq_alpha_d;
266 static void do_piechart(const struct variable *var,
267 const struct freq_tab *frq_tab);
270 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var);
274 /* Parser and outline. */
276 static int internal_cmd_frequencies (struct lexer *lexer, struct dataset *ds);
279 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
283 syntax_pool = pool_create ();
284 result = internal_cmd_frequencies (lexer, ds);
285 pool_destroy (syntax_pool);
287 pool_destroy (data_pool);
295 internal_cmd_frequencies (struct lexer *lexer, struct dataset *ds)
297 struct casegrouper *grouper;
298 struct casereader *input, *group;
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 );
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]);
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);
366 stats &= ~BIT_INDEX (frq_median);
371 input = casereader_create_filter_weight (proc_open (ds), dataset_dict (ds),
373 grouper = casegrouper_create_splits (input, dataset_dict (ds));
374 for (; casegrouper_get_next_group (grouper, &group);
375 casereader_destroy (group))
380 for (; (c = casereader_read (group)) != NULL; case_unref (c))
384 ok = casegrouper_destroy (grouper);
385 ok = proc_commit (ds) && ok;
387 free_frequencies(&cmd);
389 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
392 /* Figure out which charts the user requested. */
394 determine_charts (void)
396 int count = (!!cmd.sbc_histogram) + (!!cmd.sbc_barchart) +
397 (!!cmd.sbc_hbar) + (!!cmd.sbc_piechart);
407 msg (SW, _("At most one of BARCHART, HISTOGRAM, or HBAR should be "
408 "given. HBAR will be assumed. Argument values will be "
409 "given precedence increasing along the order given."));
411 else if (cmd.sbc_histogram)
413 else if (cmd.sbc_barchart)
415 else if (cmd.sbc_piechart)
426 if (cmd.sbc_barchart)
428 if (cmd.ba_min != SYSMIS)
430 if (cmd.ba_max != SYSMIS)
432 if (cmd.ba_scale == FRQ_FREQ)
437 else if (cmd.ba_scale == FRQ_PERCENT)
439 format = FRQ_PERCENT;
444 if (cmd.sbc_histogram)
446 if (cmd.hi_min != SYSMIS)
448 if (cmd.hi_max != SYSMIS)
450 if (cmd.hi_scale == FRQ_FREQ)
455 else if (cmd.hi_scale == FRQ_PERCENT)
457 format = FRQ_PERCENT;
460 if (cmd.hi_norm != FRQ_NONORMAL )
462 if (cmd.hi_incr == FRQ_INCREMENT)
468 if (cmd.hb_min != SYSMIS)
470 if (cmd.hb_max != SYSMIS)
472 if (cmd.hb_scale == FRQ_FREQ)
477 else if (cmd.hb_scale == FRQ_PERCENT)
479 format = FRQ_PERCENT;
484 if (cmd.hb_incr == FRQ_INCREMENT)
488 if (min != SYSMIS && max != SYSMIS && min >= max)
490 msg (SE, _("MAX must be greater than or equal to MIN, if both are "
491 "specified. However, MIN was specified as %g and MAX as %g. "
492 "MIN and MAX will be ignored."), min, max);
497 /* Add data from case C to the frequency table. */
499 calc (const struct ccase *c, const struct dataset *ds)
501 double weight = dict_get_case_weight (dataset_dict (ds), c, NULL);
504 for (i = 0; i < n_variables; i++)
506 const struct variable *v = v_variables[i];
507 const union value *val = case_data (c, v);
508 struct var_freqs *vf = get_var_freqs (v);
509 struct freq_tab *ft = &vf->tab;
511 struct freq_mutable target;
512 struct freq_mutable **fpp;
515 fpp = (struct freq_mutable **) hsh_probe (ft->data, &target);
518 (*fpp)->count += weight;
521 struct freq_mutable *fp = pool_alloc (data_pool, sizeof *fp);
523 value_init_pool (data_pool, &fp->value, vf->width);
524 value_copy (&fp->value, val, vf->width);
530 /* Prepares each variable that is the target of FREQUENCIES by setting
531 up its hash table. */
533 precalc (struct casereader *input, struct dataset *ds)
538 c = casereader_peek (input, 0);
541 output_split_file_values (ds, c);
545 pool_destroy (data_pool);
546 data_pool = pool_create ();
548 for (i = 0; i < n_variables; i++)
550 const struct variable *v = v_variables[i];
551 struct freq_tab *ft = &get_var_freqs (v)->tab;
553 ft->data = hsh_create (16, compare_freq, hash_freq, NULL, v);
557 /* Finishes up with the variables after frequencies have been
558 calculated. Displays statistics, percentiles, ... */
560 postcalc (const struct dataset *ds)
562 const struct dictionary *dict = dataset_dict (ds);
563 const struct variable *wv = dict_get_weight (dict);
566 for (i = 0; i < n_variables; i++)
568 const struct variable *v = v_variables[i];
569 struct var_freqs *vf = get_var_freqs (v);
570 struct freq_tab *ft = &vf->tab;
572 int dumped_freq_tab = 1;
574 postprocess_freq_tab (v);
576 /* Frequencies tables. */
577 n_categories = ft->n_valid + ft->n_missing;
578 if (cmd.table == FRQ_TABLE
579 || (cmd.table == FRQ_LIMIT && n_categories <= cmd.limit))
583 dump_condensed (v, wv);
589 if (n_categories > cmd.onepage_limit)
590 dump_condensed (v, wv);
602 dump_statistics (v, !dumped_freq_tab, wv);
606 if ( chart == GFT_HIST && var_is_numeric (v) )
608 double d[frq_n_stats];
609 struct histogram *hist ;
613 hist = freq_tab_to_hist (ft,v);
615 histogram_plot_n (hist, var_to_string(v),
621 statistic_destroy ((struct statistic *)hist);
624 if ( chart == GFT_PIE)
626 do_piechart(v_variables[i], ft);
629 cleanup_freq_tab (v);
634 /* Returns the comparison function that should be used for
635 sorting a frequency table by FRQ_SORT using VAL_TYPE
637 static hsh_compare_func *
638 get_freq_comparator (int frq_sort, enum val_type val_type)
640 bool is_numeric = val_type == VAL_NUMERIC;
644 return is_numeric ? compare_value_numeric_a : compare_value_alpha_a;
646 return is_numeric ? compare_value_numeric_d : compare_value_alpha_d;
648 return is_numeric ? compare_freq_numeric_a : compare_freq_alpha_a;
650 return is_numeric ? compare_freq_numeric_d : compare_freq_alpha_d;
656 /* Returns true iff the value in struct freq_mutable F is non-missing
659 not_missing (const void *f_, const void *v_)
661 const struct freq_mutable *f = f_;
662 const struct variable *v = v_;
664 return !var_is_value_missing (v, &f->value, MV_ANY);
667 /* Summarizes the frequency table data for variable V. */
669 postprocess_freq_tab (const struct variable *v)
671 hsh_compare_func *compare;
675 struct freq_mutable *freqs, *f;
678 ft = &get_var_freqs (v)->tab;
679 compare = get_freq_comparator (cmd.sort, var_get_type (v));
681 /* Extract data from hash table. */
682 count = hsh_count (ft->data);
683 data = hsh_data (ft->data);
685 /* Copy dereferenced data into freqs. */
686 freqs = xnmalloc (count, sizeof *freqs);
687 for (i = 0; i < count; i++)
689 struct freq_mutable *f = data[i];
693 /* Put data into ft. */
695 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, v);
696 ft->missing = freqs + ft->n_valid;
697 ft->n_missing = count - ft->n_valid;
700 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare, v);
701 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, v);
703 /* Summary statistics. */
704 ft->valid_cases = 0.0;
705 for(i = 0 ; i < ft->n_valid ; ++i )
708 ft->valid_cases += f->count;
712 ft->total_cases = ft->valid_cases ;
713 for(i = 0 ; i < ft->n_missing ; ++i )
716 ft->total_cases += f->count;
721 /* Frees the frequency table for variable V. */
723 cleanup_freq_tab (const struct variable *v)
725 struct freq_tab *ft = &get_var_freqs (v)->tab;
727 hsh_destroy (ft->data);
730 /* Parses the VARIABLES subcommand, adding to
731 {n_variables,v_variables}. */
733 frq_custom_variables (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *aux UNUSED)
735 size_t old_n_variables = n_variables;
738 lex_match (lexer, '=');
739 if (lex_token (lexer) != T_ALL && (lex_token (lexer) != T_ID
740 || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) == NULL))
743 if (!parse_variables_const (lexer, dataset_dict (ds), &v_variables, &n_variables,
744 PV_APPEND | PV_NO_SCRATCH))
747 for (i = old_n_variables; i < n_variables; i++)
749 const struct variable *v = v_variables[i];
750 struct var_freqs *vf;
752 if (var_get_aux (v) != NULL)
754 msg (SE, _("Variable %s specified multiple times on VARIABLES "
755 "subcommand."), var_get_name (v));
758 vf = var_attach_aux (v, xmalloc (sizeof *vf), var_dtor_free);
759 vf->tab.valid = vf->tab.missing = NULL;
760 vf->tab.dict = dataset_dict (ds);
763 vf->width = var_get_width (v);
764 vf->print = *var_get_print_format (v);
769 /* Parses the GROUPED subcommand, setting the n_grouped, grouped
770 fields of specified variables. */
772 frq_custom_grouped (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *aux UNUSED)
774 lex_match (lexer, '=');
775 if ((lex_token (lexer) == T_ID && dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
776 || lex_token (lexer) == T_ID)
781 /* Max, current size of list; list itself. */
787 const struct variable **v;
789 if (!parse_variables_const (lexer, dataset_dict (ds), &v, &n,
790 PV_NO_DUPLICATE | PV_NUMERIC))
792 if (lex_match (lexer, '('))
796 while (lex_integer (lexer))
801 dl = pool_nrealloc (syntax_pool, dl, ml, sizeof *dl);
803 dl[nl++] = lex_tokval (lexer);
805 lex_match (lexer, ',');
807 /* Note that nl might still be 0 and dl might still be
808 NULL. That's okay. */
809 if (!lex_match (lexer, ')'))
812 msg (SE, _("`)' expected after GROUPED interval list."));
822 for (i = 0; i < n; i++)
823 if (var_get_aux (v[i]) == NULL)
824 msg (SE, _("Variables %s specified on GROUPED but not on "
825 "VARIABLES."), var_get_name (v[i]));
828 struct var_freqs *vf = get_var_freqs (v[i]);
830 if (vf->groups != NULL)
831 msg (SE, _("Variables %s specified multiple times on GROUPED "
832 "subcommand."), var_get_name (v[i]));
840 if (!lex_match (lexer, '/'))
842 if ((lex_token (lexer) != T_ID || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
843 && lex_token (lexer) != T_ALL)
845 lex_put_back (lexer, '/');
853 /* Adds X to the list of percentiles, keeping the list in proper
856 add_percentile (double x)
860 for (i = 0; i < n_percentiles; i++)
862 /* Do nothing if it's already in the list */
863 if ( fabs(x - percentiles[i].p) < DBL_EPSILON )
866 if (x < percentiles[i].p)
870 if (i >= n_percentiles || x != percentiles[i].p)
872 percentiles = pool_nrealloc (syntax_pool, percentiles,
873 n_percentiles + 1, sizeof *percentiles);
874 insert_element (percentiles, n_percentiles, sizeof *percentiles, i);
875 percentiles[i].p = x;
880 /* Comparison functions. */
882 /* Ascending numeric compare of values. */
884 compare_value_numeric_a (const void *a_, const void *b_, const void *aux UNUSED)
886 const struct freq_mutable *a = a_;
887 const struct freq_mutable *b = b_;
889 if (a->value.f > b->value.f)
891 else if (a->value.f < b->value.f)
897 /* Ascending string compare of values. */
899 compare_value_alpha_a (const void *a_, const void *b_, const void *v_)
901 const struct freq_mutable *a = a_;
902 const struct freq_mutable *b = b_;
903 const struct variable *v = v_;
904 struct var_freqs *vf = get_var_freqs (v);
906 return value_compare_3way (&a->value, &b->value, vf->width);
909 /* Descending numeric compare of values. */
911 compare_value_numeric_d (const void *a, const void *b, const void *aux UNUSED)
913 return -compare_value_numeric_a (a, b, aux);
916 /* Descending string compare of values. */
918 compare_value_alpha_d (const void *a, const void *b, const void *v)
920 return -compare_value_alpha_a (a, b, v);
923 /* Ascending numeric compare of frequency;
924 secondary key on ascending numeric value. */
926 compare_freq_numeric_a (const void *a_, const void *b_, const void *aux UNUSED)
928 const struct freq_mutable *a = a_;
929 const struct freq_mutable *b = b_;
931 if (a->count > b->count)
933 else if (a->count < b->count)
936 if (a->value.f > b->value.f)
938 else if (a->value.f < b->value.f)
944 /* Ascending numeric compare of frequency;
945 secondary key on ascending string value. */
947 compare_freq_alpha_a (const void *a_, const void *b_, const void *v_)
949 const struct freq_mutable *a = a_;
950 const struct freq_mutable *b = b_;
951 const struct variable *v = v_;
952 struct var_freqs *vf = get_var_freqs (v);
954 if (a->count > b->count)
956 else if (a->count < b->count)
959 return value_compare_3way (&a->value, &b->value, vf->width);
962 /* Descending numeric compare of frequency;
963 secondary key on ascending numeric value. */
965 compare_freq_numeric_d (const void *a_, const void *b_, const void *aux UNUSED)
967 const struct freq_mutable *a = a_;
968 const struct freq_mutable *b = b_;
970 if (a->count > b->count)
972 else if (a->count < b->count)
975 if (a->value.f > b->value.f)
977 else if (a->value.f < b->value.f)
983 /* Descending numeric compare of frequency;
984 secondary key on ascending string value. */
986 compare_freq_alpha_d (const void *a_, const void *b_, const void *v_)
988 const struct freq_mutable *a = a_;
989 const struct freq_mutable *b = b_;
990 const struct variable *v = v_;
991 struct var_freqs *vf = get_var_freqs (v);
993 if (a->count > b->count)
995 else if (a->count < b->count)
998 return value_compare_3way (&a->value, &b->value, vf->width);
1001 /* Frequency table display. */
1003 /* Sets the widths of all the columns and heights of all the rows in
1004 table T for driver D. */
1006 full_dim (struct tab_table *t, struct outp_driver *d, void *aux UNUSED)
1011 if (cmd.labels == FRQ_LABELS)
1013 t->w[0] = MIN (tab_natural_width (t, d, 0), d->prop_em_width * 15);
1018 for (;i < columns; i++)
1019 t->w[i] = MAX (tab_natural_width (t, d, i), d->prop_em_width * 8);
1021 for (i = 0; i < t->nr; i++)
1022 t->h[i] = d->font_height;
1025 /* Displays a full frequency table for variable V. */
1027 dump_full (const struct variable *v, const struct variable *wv)
1029 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1031 struct var_freqs *vf;
1032 struct freq_tab *ft;
1033 struct freq_mutable *f;
1034 struct tab_table *t;
1036 double cum_total = 0.0;
1037 double cum_freq = 0.0;
1039 static const char *headings[] = {
1043 N_("Valid Percent"),
1047 const bool lab = (cmd.labels == FRQ_LABELS);
1049 vf = get_var_freqs (v);
1051 n_categories = ft->n_valid + ft->n_missing;
1052 t = tab_create (5 + lab, n_categories + 2, 0);
1053 tab_headers (t, 0, 0, 1, 0);
1054 tab_dim (t, full_dim, NULL);
1057 tab_text (t, 0, 0, TAB_CENTER | TAT_TITLE, _("Value Label"));
1059 for (x = 0; x < 5; x++)
1060 tab_text (t, lab + x, 0, TAB_CENTER | TAT_TITLE, gettext (headings[x]));
1063 for (f = ft->valid; f < ft->missing; f++)
1065 double percent, valid_percent;
1067 cum_freq += f->count;
1069 percent = f->count / ft->total_cases * 100.0;
1070 valid_percent = f->count / ft->valid_cases * 100.0;
1071 cum_total += valid_percent;
1075 const char *label = var_lookup_value_label (v, &f->value);
1077 tab_text (t, 0, r, TAB_LEFT, label);
1080 tab_value (t, 0 + lab, r, TAB_NONE, &f->value, ft->dict, &vf->print);
1081 tab_double (t, 1 + lab, r, TAB_NONE, f->count, wfmt);
1082 tab_double (t, 2 + lab, r, TAB_NONE, percent, NULL);
1083 tab_double (t, 3 + lab, r, TAB_NONE, valid_percent, NULL);
1084 tab_double (t, 4 + lab, r, TAB_NONE, cum_total, NULL);
1087 for (; f < &ft->valid[n_categories]; f++)
1089 cum_freq += f->count;
1093 const char *label = var_lookup_value_label (v, &f->value);
1095 tab_text (t, 0, r, TAB_LEFT, label);
1098 tab_value (t, 0 + lab, r, TAB_NONE, &f->value, ft->dict, &vf->print);
1099 tab_double (t, 1 + lab, r, TAB_NONE, f->count, wfmt);
1100 tab_double (t, 2 + lab, r, TAB_NONE,
1101 f->count / ft->total_cases * 100.0, NULL);
1102 tab_text (t, 3 + lab, r, TAB_NONE, _("Missing"));
1106 tab_box (t, TAL_1, TAL_1,
1107 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1109 tab_hline (t, TAL_2, 0, 4 + lab, 1);
1110 tab_hline (t, TAL_2, 0, 4 + lab, r);
1111 tab_joint_text (t, 0, r, 0 + lab, r, TAB_RIGHT | TAT_TITLE, _("Total"));
1112 tab_vline (t, TAL_0, 1, r, r);
1113 tab_double (t, 1 + lab, r, TAB_NONE, cum_freq, wfmt);
1114 tab_fixed (t, 2 + lab, r, TAB_NONE, 100.0, 5, 1);
1115 tab_fixed (t, 3 + lab, r, TAB_NONE, 100.0, 5, 1);
1117 tab_title (t, "%s", var_to_string (v));
1121 /* Sets the widths of all the columns and heights of all the rows in
1122 table T for driver D. */
1124 condensed_dim (struct tab_table *t, struct outp_driver *d, void *aux UNUSED)
1126 int cum_w = MAX (outp_string_width (d, _("Cum"), OUTP_PROPORTIONAL),
1127 MAX (outp_string_width (d, _("Cum"), OUTP_PROPORTIONAL),
1128 outp_string_width (d, "000", OUTP_PROPORTIONAL)));
1132 for (i = 0; i < 2; i++)
1133 t->w[i] = MAX (tab_natural_width (t, d, i), d->prop_em_width * 8);
1134 for (i = 2; i < 4; i++)
1136 for (i = 0; i < t->nr; i++)
1137 t->h[i] = d->font_height;
1140 /* Display condensed frequency table for variable V. */
1142 dump_condensed (const struct variable *v, const struct variable *wv)
1144 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1146 struct var_freqs *vf;
1147 struct freq_tab *ft;
1148 struct freq_mutable *f;
1149 struct tab_table *t;
1151 double cum_total = 0.0;
1153 vf = get_var_freqs (v);
1155 n_categories = ft->n_valid + ft->n_missing;
1156 t = tab_create (4, n_categories + 2, 0);
1158 tab_headers (t, 0, 0, 2, 0);
1159 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value"));
1160 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Freq"));
1161 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1162 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Cum"));
1163 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1164 tab_dim (t, condensed_dim, NULL);
1167 for (f = ft->valid; f < ft->missing; f++)
1171 percent = f->count / ft->total_cases * 100.0;
1172 cum_total += f->count / ft->valid_cases * 100.0;
1174 tab_value (t, 0, r, TAB_NONE, &f->value, ft->dict, &vf->print);
1175 tab_double (t, 1, r, TAB_NONE, f->count, wfmt);
1176 tab_double (t, 2, r, TAB_NONE, percent, NULL);
1177 tab_double (t, 3, r, TAB_NONE, cum_total, NULL);
1180 for (; f < &ft->valid[n_categories]; f++)
1182 tab_value (t, 0, r, TAB_NONE, &f->value, ft->dict, &vf->print);
1183 tab_double (t, 1, r, TAB_NONE, f->count, wfmt);
1184 tab_double (t, 2, r, TAB_NONE,
1185 f->count / ft->total_cases * 100.0, NULL);
1189 tab_box (t, TAL_1, TAL_1,
1190 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1192 tab_hline (t, TAL_2, 0, 3, 2);
1193 tab_title (t, "%s", var_to_string (v));
1194 tab_columns (t, SOM_COL_DOWN, 1);
1198 /* Statistical display. */
1200 /* Calculates all the pertinent statistics for variable V, putting
1201 them in array D[]. FIXME: This could be made much more optimal. */
1203 calc_stats (const struct variable *v, double d[frq_n_stats])
1205 struct freq_tab *ft = &get_var_freqs (v)->tab;
1206 double W = ft->valid_cases;
1208 struct freq_mutable *f=0;
1216 /* Calculate percentiles. */
1218 for (i = 0; i < n_percentiles; i++)
1220 percentiles[i].flag = 0;
1221 percentiles[i].flag2 = 0;
1225 for (idx = 0; idx < ft->n_valid; ++idx)
1227 static double prev_value = SYSMIS;
1228 f = &ft->valid[idx];
1230 for (i = 0; i < n_percentiles; i++)
1233 if ( percentiles[i].flag2 ) continue ;
1235 if ( settings_get_algorithm () != COMPATIBLE )
1237 (ft->valid_cases - 1) * percentiles[i].p;
1240 (ft->valid_cases + 1) * percentiles[i].p - 1;
1242 if ( percentiles[i].flag )
1244 percentiles[i].x2 = f->value.f;
1245 percentiles[i].x1 = prev_value;
1246 percentiles[i].flag2 = 1;
1252 if ( f->count > 1 && rank - (f->count - 1) > tp )
1254 percentiles[i].x2 = percentiles[i].x1 = f->value.f;
1255 percentiles[i].flag2 = 1;
1259 percentiles[i].flag=1;
1265 prev_value = f->value.f;
1268 for (i = 0; i < n_percentiles; i++)
1270 /* Catches the case when p == 100% */
1271 if ( ! percentiles[i].flag2 )
1272 percentiles[i].x1 = percentiles[i].x2 = f->value.f;
1275 printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
1276 i,percentiles[i].p,percentiles[i].x1,percentiles[i].x2);
1280 for (i = 0; i < n_percentiles; i++)
1282 struct freq_tab *ft = &get_var_freqs (v)->tab;
1286 if ( settings_get_algorithm () != COMPATIBLE )
1288 s = modf((ft->valid_cases - 1) * percentiles[i].p , &dummy);
1292 s = modf((ft->valid_cases + 1) * percentiles[i].p -1, &dummy);
1295 percentiles[i].value = percentiles[i].x1 +
1296 ( percentiles[i].x2 - percentiles[i].x1) * s ;
1300 /* Calculate the mode. */
1303 for (f = ft->valid; f < ft->missing; f++)
1305 if (most_often < f->count)
1307 most_often = f->count;
1308 X_mode = f->value.f;
1310 else if (most_often == f->count)
1312 /* A duplicate mode is undefined.
1313 FIXME: keep track of *all* the modes. */
1318 /* Calculate moments. */
1319 m = moments_create (MOMENT_KURTOSIS);
1320 for (f = ft->valid; f < ft->missing; f++)
1321 moments_pass_one (m, f->value.f, f->count);
1322 for (f = ft->valid; f < ft->missing; f++)
1323 moments_pass_two (m, f->value.f, f->count);
1324 moments_calculate (m, NULL, &d[frq_mean], &d[frq_variance],
1325 &d[frq_skew], &d[frq_kurt]);
1326 moments_destroy (m);
1328 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1329 d[frq_min] = ft->valid[0].value.f;
1330 d[frq_max] = ft->valid[ft->n_valid - 1].value.f;
1331 d[frq_mode] = X_mode;
1332 d[frq_range] = d[frq_max] - d[frq_min];
1333 d[frq_sum] = d[frq_mean] * W;
1334 d[frq_stddev] = sqrt (d[frq_variance]);
1335 d[frq_semean] = d[frq_stddev] / sqrt (W);
1336 d[frq_seskew] = calc_seskew (W);
1337 d[frq_sekurt] = calc_sekurt (W);
1340 /* Displays a table of all the statistics requested for variable V. */
1342 dump_statistics (const struct variable *v, bool show_varname,
1343 const struct variable *wv)
1345 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1346 struct freq_tab *ft;
1347 double stat_value[frq_n_stats];
1348 struct tab_table *t;
1351 if (var_is_alpha (v))
1353 ft = &get_var_freqs (v)->tab;
1354 if (ft->n_valid == 0)
1356 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1360 calc_stats (v, stat_value);
1362 t = tab_create (3, n_stats + n_percentiles + 2, 0);
1363 tab_dim (t, tab_natural_dimensions, NULL);
1365 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1368 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1369 tab_vline (t, TAL_GAP , 1, 0, tab_nr(t) - 1 ) ;
1371 r=2; /* N missing and N valid are always dumped */
1373 for (i = 0; i < frq_n_stats; i++)
1374 if (stats & BIT_INDEX (i))
1376 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1377 gettext (st_name[i].s10));
1378 tab_double (t, 2, r, TAB_NONE, stat_value[i], NULL);
1382 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1383 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1384 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1386 tab_double (t, 2, 0, TAB_NONE, ft->valid_cases, wfmt);
1387 tab_double (t, 2, 1, TAB_NONE, ft->total_cases - ft->valid_cases, wfmt);
1389 for (i = 0; i < n_percentiles; i++, r++)
1393 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1396 if (percentiles[i].p == 0.5)
1397 tab_text (t, 1, r, TAB_LEFT, _("50 (Median)"));
1399 tab_fixed (t, 1, r, TAB_LEFT, percentiles[i].p * 100, 3, 0);
1400 tab_double (t, 2, r, TAB_NONE, percentiles[i].value,
1401 var_get_print_format (v));
1404 tab_columns (t, SOM_COL_DOWN, 1);
1406 tab_title (t, "%s", var_to_string (v));
1408 tab_flags (t, SOMF_NO_TITLE);
1415 /* Create a gsl_histogram from a freq_tab */
1417 freq_tab_to_hist (const struct freq_tab *ft, const struct variable *var)
1420 double x_min = DBL_MAX;
1421 double x_max = -DBL_MAX;
1423 struct statistic *hist;
1424 const double bins = 11;
1426 struct hsh_iterator hi;
1427 struct hsh_table *fh = ft->data;
1428 struct freq_mutable *frq;
1430 /* Find out the extremes of the x value */
1431 for ( frq = hsh_first(fh, &hi); frq != 0; frq = hsh_next(fh, &hi) )
1433 if (var_is_value_missing(var, &frq->value, MV_ANY))
1436 if ( frq->value.f < x_min ) x_min = frq->value.f ;
1437 if ( frq->value.f > x_max ) x_max = frq->value.f ;
1440 hist = histogram_create (bins, x_min, x_max);
1442 for( i = 0 ; i < ft->n_valid ; ++i )
1444 frq = &ft->valid[i];
1445 histogram_add ((struct histogram *)hist, frq->value.f, frq->count);
1448 return (struct histogram *)hist;
1452 static struct slice *
1453 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1454 const struct variable *var,
1458 /* Allocate an array of slices and fill them from the data in frq_tab
1459 n_slices will contain the number of slices allocated.
1460 The caller is responsible for freeing slices
1462 static struct slice *
1463 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1464 const struct variable *var,
1468 struct slice *slices;
1470 *n_slices = frq_tab->n_valid;
1472 slices = xnmalloc (*n_slices, sizeof *slices);
1474 for (i = 0 ; i < *n_slices ; ++i )
1476 const struct freq_mutable *frq = &frq_tab->valid[i];
1478 ds_init_empty (&slices[i].label);
1479 var_append_value_name (var, &frq->value, &slices[i].label);
1480 slices[i].magnetude = frq->count;
1490 do_piechart(const struct variable *var, const struct freq_tab *frq_tab)
1492 struct slice *slices;
1495 slices = freq_tab_to_slice_array(frq_tab, var, &n_slices);
1497 piechart_plot(var_to_string(var), slices, n_slices);
1499 for (i = 0 ; i < n_slices ; ++i )
1501 ds_destroy (&slices[i].label);