1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2007 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/alloc.h>
42 #include <libpspp/array.h>
43 #include <libpspp/bit-vector.h>
44 #include <libpspp/compiler.h>
45 #include <libpspp/hash.h>
46 #include <libpspp/magic.h>
47 #include <libpspp/message.h>
48 #include <libpspp/misc.h>
49 #include <libpspp/pool.h>
50 #include <libpspp/str.h>
51 #include <math/histogram.h>
52 #include <math/moments.h>
53 #include <output/chart.h>
54 #include <output/charts/piechart.h>
55 #include <output/charts/plot-hist.h>
56 #include <output/manager.h>
57 #include <output/output.h>
58 #include <output/table.h>
65 #define _(msgid) gettext (msgid)
66 #define N_(msgid) msgid
73 +format=cond:condense/onepage(*n:onepage_limit,"%s>=0")/!standard,
74 table:limit(n:limit,"%s>0")/notable/!table,
75 labels:!labels/nolabels,
76 sort:!avalue/dvalue/afreq/dfreq,
77 spaces:!single/double,
78 paging:newpage/!oldpage;
79 missing=miss:include/!exclude;
80 barchart(ba_)=:minimum(d:min),
82 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0");
83 piechart(pie_)=:minimum(d:min),
85 missing:missing/!nomissing;
86 histogram(hi_)=:minimum(d:min),
88 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
89 norm:!nonormal/normal,
90 incr:increment(d:inc,"%s>0");
91 hbar(hb_)=:minimum(d:min),
93 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
94 norm:!nonormal/normal,
95 incr:increment(d:inc,"%s>0");
98 +percentiles = double list;
99 +statistics[st_]=1|mean,2|semean,3|median,4|mode,5|stddev,6|variance,
100 7|kurtosis,8|skewness,9|range,10|minimum,11|maximum,12|sum,
101 13|default,14|seskewness,15|sekurtosis,all,none.
109 frq_mean = 0, frq_semean, frq_median, frq_mode, frq_stddev, frq_variance,
110 frq_kurt, frq_sekurt, frq_skew, frq_seskew, frq_range, frq_min, frq_max,
114 /* Description of a statistic. */
117 int st_indx; /* Index into a_statistics[]. */
118 const char *s10; /* Identifying string. */
121 /* Table of statistics, indexed by dsc_*. */
122 static const struct frq_info st_name[frq_n_stats + 1] =
124 {FRQ_ST_MEAN, N_("Mean")},
125 {FRQ_ST_SEMEAN, N_("S.E. Mean")},
126 {FRQ_ST_MEDIAN, N_("Median")},
127 {FRQ_ST_MODE, N_("Mode")},
128 {FRQ_ST_STDDEV, N_("Std Dev")},
129 {FRQ_ST_VARIANCE, N_("Variance")},
130 {FRQ_ST_KURTOSIS, N_("Kurtosis")},
131 {FRQ_ST_SEKURTOSIS, N_("S.E. Kurt")},
132 {FRQ_ST_SKEWNESS, N_("Skewness")},
133 {FRQ_ST_SESKEWNESS, N_("S.E. Skew")},
134 {FRQ_ST_RANGE, N_("Range")},
135 {FRQ_ST_MINIMUM, N_("Minimum")},
136 {FRQ_ST_MAXIMUM, N_("Maximum")},
137 {FRQ_ST_SUM, N_("Sum")},
141 /* Percentiles to calculate. */
145 double p; /* the %ile to be calculated */
146 double value; /* the %ile's value */
147 double x1; /* The datum value <= the percentile */
148 double x2; /* The datum value >= the percentile */
150 int flag2; /* Set to 1 if this percentile value has been found */
154 static void add_percentile (double x) ;
156 static struct percentile *percentiles;
157 static int n_percentiles;
159 static int implicit_50th ;
161 /* Groups of statistics. */
163 #define frq_default \
164 (BI (frq_mean) | BI (frq_stddev) | BI (frq_min) | BI (frq_max))
166 (BI (frq_sum) | BI(frq_min) | BI(frq_max) \
167 | BI(frq_mean) | BI(frq_semean) | BI(frq_stddev) \
168 | BI(frq_variance) | BI(frq_kurt) | BI(frq_sekurt) \
169 | BI(frq_skew) | BI(frq_seskew) | BI(frq_range) \
170 | BI(frq_range) | BI(frq_mode) | BI(frq_median))
172 /* Statistics; number of statistics. */
173 static unsigned long stats;
176 /* Types of graphs. */
179 GFT_NONE, /* Don't draw graphs. */
180 GFT_BAR, /* Draw bar charts. */
181 GFT_HIST, /* Draw histograms. */
182 GFT_PIE, /* Draw piechart */
183 GFT_HBAR /* Draw bar charts or histograms at our discretion. */
186 /* Parsed command. */
187 static struct cmd_frequencies cmd;
189 /* Summary of the barchart, histogram, and hbar subcommands. */
190 /* FIXME: These should not be mututally exclusive */
191 static int chart; /* NONE/BAR/HIST/HBAR/PIE. */
192 static double min, max; /* Minimum, maximum on y axis. */
193 static int format; /* FREQ/PERCENT: Scaling of y axis. */
194 static double scale, incr; /* FIXME */
195 static int normal; /* FIXME */
197 /* Variables for which to calculate statistics. */
198 static size_t n_variables;
199 static const struct variable **v_variables;
202 static struct pool *data_pool; /* For per-SPLIT FILE group data. */
203 static struct pool *syntax_pool; /* For syntax-related data. */
205 /* Frequency tables. */
207 /* Entire frequency table. */
210 struct hsh_table *data; /* Undifferentiated data. */
211 struct freq *valid; /* Valid freqs. */
212 int n_valid; /* Number of total freqs. */
214 struct freq *missing; /* Missing freqs. */
215 int n_missing; /* Number of missing freqs. */
218 double total_cases; /* Sum of weights of all cases. */
219 double valid_cases; /* Sum of weights of valid cases. */
223 /* Per-variable frequency data. */
226 /* Freqency table. */
227 struct freq_tab tab; /* Frequencies table to use. */
230 int n_groups; /* Number of groups. */
231 double *groups; /* Groups. */
234 double stat[frq_n_stats];
236 /* Width and format for analysis and display.
237 This is normally the same as "width" and "print" in struct
238 variable, but in SPSS-compatible mode only the first
239 MAX_SHORT_STRING bytes of long string variables are
242 struct fmt_spec print;
245 static inline struct var_freqs *
246 get_var_freqs (const struct variable *v)
248 return var_get_aux (v);
251 static void determine_charts (void);
253 static void calc_stats (const struct variable *v, double d[frq_n_stats]);
255 static void precalc (struct casereader *, struct dataset *);
256 static void calc (const struct ccase *, const struct dataset *);
257 static void postcalc (void);
259 static void postprocess_freq_tab (const struct variable *);
260 static void dump_full (const struct variable *);
261 static void dump_condensed (const struct variable *);
262 static void dump_statistics (const struct variable *, int show_varname);
263 static void cleanup_freq_tab (const struct variable *);
265 static hsh_compare_func compare_value_numeric_a, compare_value_alpha_a;
266 static hsh_compare_func compare_value_numeric_d, compare_value_alpha_d;
267 static hsh_compare_func compare_freq_numeric_a, compare_freq_alpha_a;
268 static hsh_compare_func compare_freq_numeric_d, compare_freq_alpha_d;
271 static void do_piechart(const struct variable *var,
272 const struct freq_tab *frq_tab);
275 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var);
279 /* Parser and outline. */
281 static int internal_cmd_frequencies (struct lexer *lexer, struct dataset *ds);
284 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
288 syntax_pool = pool_create ();
289 result = internal_cmd_frequencies (lexer, ds);
290 pool_destroy (syntax_pool);
292 pool_destroy (data_pool);
300 internal_cmd_frequencies (struct lexer *lexer, struct dataset *ds)
302 struct casegrouper *grouper;
303 struct casereader *input, *group;
313 if (!parse_frequencies (lexer, ds, &cmd, NULL))
316 if (cmd.onepage_limit == NOT_LONG)
317 cmd.onepage_limit = 50;
319 /* Figure out statistics to calculate. */
321 if (cmd.a_statistics[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
322 stats |= frq_default;
323 if (cmd.a_statistics[FRQ_ST_ALL])
325 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
326 stats &= ~frq_median;
327 for (i = 0; i < frq_n_stats; i++)
328 if (cmd.a_statistics[st_name[i].st_indx])
329 stats |= BIT_INDEX (i);
330 if (stats & frq_kurt)
332 if (stats & frq_skew)
335 /* Calculate n_stats. */
337 for (i = 0; i < frq_n_stats; i++)
338 if ((stats & BIT_INDEX (i)))
343 if (chart != GFT_NONE || cmd.sbc_ntiles)
344 cmd.sort = FRQ_AVALUE;
346 /* Work out what percentiles need to be calculated */
347 if ( cmd.sbc_percentiles )
349 for ( i = 0 ; i < MAXLISTS ; ++i )
352 subc_list_double *ptl_list = &cmd.dl_percentiles[i];
353 for ( pl = 0 ; pl < subc_list_double_count(ptl_list); ++pl)
354 add_percentile (subc_list_double_at(ptl_list, pl) / 100.0 );
357 if ( cmd.sbc_ntiles )
359 for ( i = 0 ; i < cmd.sbc_ntiles ; ++i )
362 for (j = 0; j <= cmd.n_ntiles[i]; ++j )
363 add_percentile (j / (double) cmd.n_ntiles[i]);
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 (; casereader_read (group, &c); case_destroy (&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;
512 target.value = (union value *) val;
513 fpp = (struct freq **) hsh_probe (ft->data, &target);
516 (*fpp)->count += weight;
519 struct freq *fp = pool_alloc (data_pool, sizeof *fp);
521 fp->value = pool_clone (data_pool,
523 MAX (MAX_SHORT_STRING, vf->width));
529 /* Prepares each variable that is the target of FREQUENCIES by setting
530 up its hash table. */
532 precalc (struct casereader *input, struct dataset *ds)
537 if (!casereader_peek (input, 0, &c))
539 output_split_file_values (ds, &c);
542 pool_destroy (data_pool);
543 data_pool = pool_create ();
545 for (i = 0; i < n_variables; i++)
547 const struct variable *v = v_variables[i];
548 struct freq_tab *ft = &get_var_freqs (v)->tab;
550 ft->data = hsh_create (16, compare_freq, hash_freq, NULL, v);
554 /* Finishes up with the variables after frequencies have been
555 calculated. Displays statistics, percentiles, ... */
561 for (i = 0; i < n_variables; i++)
563 const struct variable *v = v_variables[i];
564 struct var_freqs *vf = get_var_freqs (v);
565 struct freq_tab *ft = &vf->tab;
567 int dumped_freq_tab = 1;
569 postprocess_freq_tab (v);
571 /* Frequencies tables. */
572 n_categories = ft->n_valid + ft->n_missing;
573 if (cmd.table == FRQ_TABLE
574 || (cmd.table == FRQ_LIMIT && n_categories <= cmd.limit))
584 if (n_categories > cmd.onepage_limit)
597 dump_statistics (v, !dumped_freq_tab);
601 if ( chart == GFT_HIST)
603 double d[frq_n_stats];
604 struct normal_curve norm;
605 gsl_histogram *hist ;
608 norm.N = vf->tab.valid_cases;
611 norm.mean = d[frq_mean];
612 norm.stddev = d[frq_stddev];
614 hist = freq_tab_to_hist(ft,v);
616 histogram_plot(hist, var_to_string(v), &norm, normal);
618 gsl_histogram_free(hist);
622 if ( chart == GFT_PIE)
624 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 VAR_TYPE
637 static hsh_compare_func *
638 get_freq_comparator (int frq_sort, enum var_type var_type)
640 bool is_numeric = var_type == VAR_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 F is non-missing
659 not_missing (const void *f_, const void *v_)
661 const struct freq *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 *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 *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;
762 vf->width = var_get_width (v);
763 vf->print = *var_get_print_format (v);
764 if (vf->width > MAX_SHORT_STRING && get_algorithm () == COMPATIBLE)
766 enum fmt_type type = var_get_print_format (v)->type;
767 vf->width = MAX_SHORT_STRING;
768 vf->print.w = MAX_SHORT_STRING * (type == FMT_AHEX ? 2 : 1);
774 /* Parses the GROUPED subcommand, setting the n_grouped, grouped
775 fields of specified variables. */
777 frq_custom_grouped (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *aux UNUSED)
779 lex_match (lexer, '=');
780 if ((lex_token (lexer) == T_ID && dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
781 || lex_token (lexer) == T_ID)
786 /* Max, current size of list; list itself. */
792 const struct variable **v;
794 if (!parse_variables_const (lexer, dataset_dict (ds), &v, &n,
795 PV_NO_DUPLICATE | PV_NUMERIC))
797 if (lex_match (lexer, '('))
801 while (lex_integer (lexer))
806 dl = pool_nrealloc (syntax_pool, dl, ml, sizeof *dl);
808 dl[nl++] = lex_tokval (lexer);
810 lex_match (lexer, ',');
812 /* Note that nl might still be 0 and dl might still be
813 NULL. That's okay. */
814 if (!lex_match (lexer, ')'))
817 msg (SE, _("`)' expected after GROUPED interval list."));
827 for (i = 0; i < n; i++)
828 if (var_get_aux (v[i]) == NULL)
829 msg (SE, _("Variables %s specified on GROUPED but not on "
830 "VARIABLES."), var_get_name (v[i]));
833 struct var_freqs *vf = get_var_freqs (v[i]);
835 if (vf->groups != NULL)
836 msg (SE, _("Variables %s specified multiple times on GROUPED "
837 "subcommand."), var_get_name (v[i]));
845 if (!lex_match (lexer, '/'))
847 if ((lex_token (lexer) != T_ID || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
848 && lex_token (lexer) != T_ALL)
850 lex_put_back (lexer, '/');
858 /* Adds X to the list of percentiles, keeping the list in proper
861 add_percentile (double x)
865 for (i = 0; i < n_percentiles; i++)
867 /* Do nothing if it's already in the list */
868 if ( fabs(x - percentiles[i].p) < DBL_EPSILON )
871 if (x < percentiles[i].p)
875 if (i >= n_percentiles || x != percentiles[i].p)
877 percentiles = pool_nrealloc (syntax_pool, percentiles,
878 n_percentiles + 1, sizeof *percentiles);
880 if (i < n_percentiles)
881 memmove (&percentiles[i + 1], &percentiles[i],
882 (n_percentiles - i) * sizeof (struct percentile) );
884 percentiles[i].p = x;
889 /* Comparison functions. */
891 /* Ascending numeric compare of values. */
893 compare_value_numeric_a (const void *a_, const void *b_, const void *aux UNUSED)
895 const struct freq *a = a_;
896 const struct freq *b = b_;
898 if (a->value[0].f > b->value[0].f)
900 else if (a->value[0].f < b->value[0].f)
906 /* Ascending string compare of values. */
908 compare_value_alpha_a (const void *a_, const void *b_, const void *v_)
910 const struct freq *a = a_;
911 const struct freq *b = b_;
912 const struct variable *v = v_;
913 struct var_freqs *vf = get_var_freqs (v);
915 return memcmp (a->value[0].s, b->value[0].s, vf->width);
918 /* Descending numeric compare of values. */
920 compare_value_numeric_d (const void *a, const void *b, const void *aux UNUSED)
922 return -compare_value_numeric_a (a, b, aux);
925 /* Descending string compare of values. */
927 compare_value_alpha_d (const void *a, const void *b, const void *v)
929 return -compare_value_alpha_a (a, b, v);
932 /* Ascending numeric compare of frequency;
933 secondary key on ascending numeric value. */
935 compare_freq_numeric_a (const void *a_, const void *b_, const void *aux UNUSED)
937 const struct freq *a = a_;
938 const struct freq *b = b_;
940 if (a->count > b->count)
942 else if (a->count < b->count)
945 if (a->value[0].f > b->value[0].f)
947 else if (a->value[0].f < b->value[0].f)
953 /* Ascending numeric compare of frequency;
954 secondary key on ascending string value. */
956 compare_freq_alpha_a (const void *a_, const void *b_, const void *v_)
958 const struct freq *a = a_;
959 const struct freq *b = b_;
960 const struct variable *v = v_;
961 struct var_freqs *vf = get_var_freqs (v);
963 if (a->count > b->count)
965 else if (a->count < b->count)
968 return memcmp (a->value[0].s, b->value[0].s, vf->width);
971 /* Descending numeric compare of frequency;
972 secondary key on ascending numeric value. */
974 compare_freq_numeric_d (const void *a_, const void *b_, const void *aux UNUSED)
976 const struct freq *a = a_;
977 const struct freq *b = b_;
979 if (a->count > b->count)
981 else if (a->count < b->count)
984 if (a->value[0].f > b->value[0].f)
986 else if (a->value[0].f < b->value[0].f)
992 /* Descending numeric compare of frequency;
993 secondary key on ascending string value. */
995 compare_freq_alpha_d (const void *a_, const void *b_, const void *v_)
997 const struct freq *a = a_;
998 const struct freq *b = b_;
999 const struct variable *v = v_;
1000 struct var_freqs *vf = get_var_freqs (v);
1002 if (a->count > b->count)
1004 else if (a->count < b->count)
1007 return memcmp (a->value[0].s, b->value[0].s, vf->width);
1010 /* Frequency table display. */
1012 /* Sets the widths of all the columns and heights of all the rows in
1013 table T for driver D. */
1015 full_dim (struct tab_table *t, struct outp_driver *d)
1020 if (cmd.labels == FRQ_LABELS)
1022 t->w[0] = MIN (tab_natural_width (t, d, 0), d->prop_em_width * 15);
1027 for (;i < columns; i++)
1028 t->w[i] = MAX (tab_natural_width (t, d, i), d->prop_em_width * 8);
1030 for (i = 0; i < t->nr; i++)
1031 t->h[i] = d->font_height;
1034 /* Displays a full frequency table for variable V. */
1036 dump_full (const struct variable *v)
1039 struct var_freqs *vf;
1040 struct freq_tab *ft;
1042 struct tab_table *t;
1044 double cum_total = 0.0;
1045 double cum_freq = 0.0;
1053 const struct init *p;
1055 static const struct init vec[] =
1057 {4, 0, N_("Valid")},
1059 {1, 1, N_("Value")},
1060 {2, 1, N_("Frequency")},
1061 {3, 1, N_("Percent")},
1062 {4, 1, N_("Percent")},
1063 {5, 1, N_("Percent")},
1071 const bool lab = (cmd.labels == FRQ_LABELS);
1073 vf = get_var_freqs (v);
1075 n_categories = ft->n_valid + ft->n_missing;
1076 t = tab_create (5 + lab, n_categories + 3, 0);
1077 tab_headers (t, 0, 0, 2, 0);
1078 tab_dim (t, full_dim);
1081 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value Label"));
1083 for (p = vec; p->s; p++)
1084 tab_text (t, lab ? p->c : p->c - 1, p->r,
1085 TAB_CENTER | TAT_TITLE, gettext (p->s));
1088 for (f = ft->valid; f < ft->missing; f++)
1090 double percent, valid_percent;
1092 cum_freq += f->count;
1094 percent = f->count / ft->total_cases * 100.0;
1095 valid_percent = f->count / ft->valid_cases * 100.0;
1096 cum_total += valid_percent;
1100 const char *label = var_lookup_value_label (v, &f->value[0]);
1102 tab_text (t, 0, r, TAB_LEFT, label);
1105 tab_value (t, 0 + lab, r, TAB_NONE, f->value, &vf->print);
1106 tab_float (t, 1 + lab, r, TAB_NONE, f->count, 8, 0);
1107 tab_float (t, 2 + lab, r, TAB_NONE, percent, 5, 1);
1108 tab_float (t, 3 + lab, r, TAB_NONE, valid_percent, 5, 1);
1109 tab_float (t, 4 + lab, r, TAB_NONE, cum_total, 5, 1);
1112 for (; f < &ft->valid[n_categories]; f++)
1114 cum_freq += f->count;
1118 const char *label = var_lookup_value_label (v, &f->value[0]);
1120 tab_text (t, 0, r, TAB_LEFT, label);
1123 tab_value (t, 0 + lab, r, TAB_NONE, f->value, &vf->print);
1124 tab_float (t, 1 + lab, r, TAB_NONE, f->count, 8, 0);
1125 tab_float (t, 2 + lab, r, TAB_NONE,
1126 f->count / ft->total_cases * 100.0, 5, 1);
1127 tab_text (t, 3 + lab, r, TAB_NONE, _("Missing"));
1131 tab_box (t, TAL_1, TAL_1,
1132 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1134 tab_hline (t, TAL_2, 0, 4 + lab, 2);
1135 tab_hline (t, TAL_2, 0, 4 + lab, r);
1136 tab_joint_text (t, 0, r, 0 + lab, r, TAB_RIGHT | TAT_TITLE, _("Total"));
1137 tab_vline (t, TAL_0, 1, r, r);
1138 tab_float (t, 1 + lab, r, TAB_NONE, cum_freq, 8, 0);
1139 tab_float (t, 2 + lab, r, TAB_NONE, 100.0, 5, 1);
1140 tab_float (t, 3 + lab, r, TAB_NONE, 100.0, 5, 1);
1142 tab_title (t, "%s", var_to_string (v));
1146 /* Sets the widths of all the columns and heights of all the rows in
1147 table T for driver D. */
1149 condensed_dim (struct tab_table *t, struct outp_driver *d)
1151 int cum_w = MAX (outp_string_width (d, _("Cum"), OUTP_PROPORTIONAL),
1152 MAX (outp_string_width (d, _("Cum"), OUTP_PROPORTIONAL),
1153 outp_string_width (d, "000", OUTP_PROPORTIONAL)));
1157 for (i = 0; i < 2; i++)
1158 t->w[i] = MAX (tab_natural_width (t, d, i), d->prop_em_width * 8);
1159 for (i = 2; i < 4; i++)
1161 for (i = 0; i < t->nr; i++)
1162 t->h[i] = d->font_height;
1165 /* Display condensed frequency table for variable V. */
1167 dump_condensed (const struct variable *v)
1170 struct var_freqs *vf;
1171 struct freq_tab *ft;
1173 struct tab_table *t;
1175 double cum_total = 0.0;
1177 vf = get_var_freqs (v);
1179 n_categories = ft->n_valid + ft->n_missing;
1180 t = tab_create (4, n_categories + 2, 0);
1182 tab_headers (t, 0, 0, 2, 0);
1183 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value"));
1184 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Freq"));
1185 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1186 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Cum"));
1187 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1188 tab_dim (t, condensed_dim);
1191 for (f = ft->valid; f < ft->missing; f++)
1195 percent = f->count / ft->total_cases * 100.0;
1196 cum_total += f->count / ft->valid_cases * 100.0;
1198 tab_value (t, 0, r, TAB_NONE, f->value, &vf->print);
1199 tab_float (t, 1, r, TAB_NONE, f->count, 8, 0);
1200 tab_float (t, 2, r, TAB_NONE, percent, 3, 0);
1201 tab_float (t, 3, r, TAB_NONE, cum_total, 3, 0);
1204 for (; f < &ft->valid[n_categories]; f++)
1206 tab_value (t, 0, r, TAB_NONE, f->value, &vf->print);
1207 tab_float (t, 1, r, TAB_NONE, f->count, 8, 0);
1208 tab_float (t, 2, r, TAB_NONE,
1209 f->count / ft->total_cases * 100.0, 3, 0);
1213 tab_box (t, TAL_1, TAL_1,
1214 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1216 tab_hline (t, TAL_2, 0, 3, 2);
1217 tab_title (t, "%s", var_to_string (v));
1218 tab_columns (t, SOM_COL_DOWN, 1);
1222 /* Statistical display. */
1224 /* Calculates all the pertinent statistics for variable V, putting
1225 them in array D[]. FIXME: This could be made much more optimal. */
1227 calc_stats (const struct variable *v, double d[frq_n_stats])
1229 struct freq_tab *ft = &get_var_freqs (v)->tab;
1230 double W = ft->valid_cases;
1239 double *median_value;
1241 /* Calculate percentiles. */
1243 /* If the 50th percentile was not explicitly requested then we must
1244 calculate it anyway --- it's the median */
1246 for (i = 0; i < n_percentiles; i++)
1248 if (percentiles[i].p == 0.5)
1250 median_value = &percentiles[i].value;
1255 if ( 0 == median_value )
1257 add_percentile (0.5);
1261 for (i = 0; i < n_percentiles; i++)
1263 percentiles[i].flag = 0;
1264 percentiles[i].flag2 = 0;
1268 for (idx = 0; idx < ft->n_valid; ++idx)
1270 static double prev_value = SYSMIS;
1271 f = &ft->valid[idx];
1273 for (i = 0; i < n_percentiles; i++)
1276 if ( percentiles[i].flag2 ) continue ;
1278 if ( get_algorithm() != COMPATIBLE )
1280 (ft->valid_cases - 1) * percentiles[i].p;
1283 (ft->valid_cases + 1) * percentiles[i].p - 1;
1285 if ( percentiles[i].flag )
1287 percentiles[i].x2 = f->value[0].f;
1288 percentiles[i].x1 = prev_value;
1289 percentiles[i].flag2 = 1;
1295 if ( f->count > 1 && rank - (f->count - 1) > tp )
1297 percentiles[i].x2 = percentiles[i].x1 = f->value[0].f;
1298 percentiles[i].flag2 = 1;
1302 percentiles[i].flag=1;
1308 prev_value = f->value[0].f;
1311 for (i = 0; i < n_percentiles; i++)
1313 /* Catches the case when p == 100% */
1314 if ( ! percentiles[i].flag2 )
1315 percentiles[i].x1 = percentiles[i].x2 = f->value[0].f;
1318 printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
1319 i,percentiles[i].p,percentiles[i].x1,percentiles[i].x2);
1323 for (i = 0; i < n_percentiles; i++)
1325 struct freq_tab *ft = &get_var_freqs (v)->tab;
1329 if ( get_algorithm() != COMPATIBLE )
1331 s = modf((ft->valid_cases - 1) * percentiles[i].p , &dummy);
1335 s = modf((ft->valid_cases + 1) * percentiles[i].p -1, &dummy);
1338 percentiles[i].value = percentiles[i].x1 +
1339 ( percentiles[i].x2 - percentiles[i].x1) * s ;
1341 if ( percentiles[i].p == 0.50)
1342 median_value = &percentiles[i].value;
1346 /* Calculate the mode. */
1349 for (f = ft->valid; f < ft->missing; f++)
1351 if (most_often < f->count)
1353 most_often = f->count;
1354 X_mode = f->value[0].f;
1356 else if (most_often == f->count)
1358 /* A duplicate mode is undefined.
1359 FIXME: keep track of *all* the modes. */
1364 /* Calculate moments. */
1365 m = moments_create (MOMENT_KURTOSIS);
1366 for (f = ft->valid; f < ft->missing; f++)
1367 moments_pass_one (m, f->value[0].f, f->count);
1368 for (f = ft->valid; f < ft->missing; f++)
1369 moments_pass_two (m, f->value[0].f, f->count);
1370 moments_calculate (m, NULL, &d[frq_mean], &d[frq_variance],
1371 &d[frq_skew], &d[frq_kurt]);
1372 moments_destroy (m);
1374 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1375 d[frq_min] = ft->valid[0].value[0].f;
1376 d[frq_max] = ft->valid[ft->n_valid - 1].value[0].f;
1377 d[frq_mode] = X_mode;
1378 d[frq_range] = d[frq_max] - d[frq_min];
1379 d[frq_median] = *median_value;
1380 d[frq_sum] = d[frq_mean] * W;
1381 d[frq_stddev] = sqrt (d[frq_variance]);
1382 d[frq_semean] = d[frq_stddev] / sqrt (W);
1383 d[frq_seskew] = calc_seskew (W);
1384 d[frq_sekurt] = calc_sekurt (W);
1387 /* Displays a table of all the statistics requested for variable V. */
1389 dump_statistics (const struct variable *v, int show_varname)
1391 struct freq_tab *ft;
1392 double stat_value[frq_n_stats];
1393 struct tab_table *t;
1396 int n_explicit_percentiles = n_percentiles;
1398 if ( implicit_50th && n_percentiles > 0 )
1401 if (var_is_alpha (v))
1403 ft = &get_var_freqs (v)->tab;
1404 if (ft->n_valid == 0)
1406 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1410 calc_stats (v, stat_value);
1412 t = tab_create (3, n_stats + n_explicit_percentiles + 2, 0);
1413 tab_dim (t, tab_natural_dimensions);
1415 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1418 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1419 tab_vline (t, TAL_GAP , 1, 0, tab_nr(t) - 1 ) ;
1421 r=2; /* N missing and N valid are always dumped */
1423 for (i = 0; i < frq_n_stats; i++)
1424 if (stats & BIT_INDEX (i))
1426 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1427 gettext (st_name[i].s10));
1428 tab_float (t, 2, r, TAB_NONE, stat_value[i], 11, 3);
1432 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1433 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1434 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1436 tab_float(t, 2, 0, TAB_NONE, ft->valid_cases, 11, 0);
1437 tab_float(t, 2, 1, TAB_NONE, ft->total_cases - ft->valid_cases, 11, 0);
1440 for (i = 0; i < n_explicit_percentiles; i++, r++)
1444 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1447 tab_float (t, 1, r, TAB_LEFT, percentiles[i].p * 100, 3, 0 );
1448 tab_float (t, 2, r, TAB_NONE, percentiles[i].value, 11, 3);
1452 tab_columns (t, SOM_COL_DOWN, 1);
1454 tab_title (t, "%s", var_to_string (v));
1456 tab_flags (t, SOMF_NO_TITLE);
1463 /* Create a gsl_histogram from a freq_tab */
1465 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var)
1468 double x_min = DBL_MAX;
1469 double x_max = -DBL_MAX;
1471 gsl_histogram *hist;
1472 const double bins = 11;
1474 struct hsh_iterator hi;
1475 struct hsh_table *fh = ft->data;
1478 /* Find out the extremes of the x value */
1479 for ( frq = hsh_first(fh, &hi); frq != 0; frq = hsh_next(fh, &hi) )
1481 if (var_is_value_missing(var, frq->value, MV_ANY))
1484 if ( frq->value[0].f < x_min ) x_min = frq->value[0].f ;
1485 if ( frq->value[0].f > x_max ) x_max = frq->value[0].f ;
1488 hist = histogram_create(bins, x_min, x_max);
1490 for( i = 0 ; i < ft->n_valid ; ++i )
1492 frq = &ft->valid[i];
1493 gsl_histogram_accumulate(hist, frq->value[0].f, frq->count);
1500 static struct slice *
1501 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1502 const struct variable *var,
1506 /* Allocate an array of slices and fill them from the data in frq_tab
1507 n_slices will contain the number of slices allocated.
1508 The caller is responsible for freeing slices
1510 static struct slice *
1511 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1512 const struct variable *var,
1516 struct slice *slices;
1518 *n_slices = frq_tab->n_valid;
1520 slices = xnmalloc (*n_slices, sizeof *slices);
1522 for (i = 0 ; i < *n_slices ; ++i )
1524 const struct freq *frq = &frq_tab->valid[i];
1526 slices[i].label = var_get_value_name (var, frq->value);
1527 slices[i].magnetude = frq->count;
1537 do_piechart(const struct variable *var, const struct freq_tab *frq_tab)
1539 struct slice *slices;
1542 slices = freq_tab_to_slice_array(frq_tab, var, &n_slices);
1544 piechart_plot(var_to_string(var), slices, n_slices);