1 /* PSPP - computes sample statistics.
2 Copyright (C) 1997-9, 2000 Free Software Foundation, Inc.
3 Written by Ben Pfaff <blp@gnu.org>.
5 This program is free software; you can redistribute it and/or
6 modify it under the terms of the GNU General Public License as
7 published by the Free Software Foundation; either version 2 of the
8 License, or (at your option) any later version.
10 This program is distributed in the hope that it will be useful, but
11 WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with this program; if not, write to the Free Software
17 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
23 * Remember that histograms, bar charts need mean, stddev.
30 #include <gsl/gsl_histogram.h>
32 #include <data/case.h>
33 #include <data/dictionary.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/message.h>
49 #include <libpspp/misc.h>
50 #include <libpspp/pool.h>
51 #include <libpspp/str.h>
52 #include <math/histogram.h>
53 #include <math/moments.h>
54 #include <output/chart.h>
55 #include <output/charts/piechart.h>
56 #include <output/charts/plot-hist.h>
57 #include <output/manager.h>
58 #include <output/output.h>
59 #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 static int implicit_50th ;
160 /* Groups of statistics. */
162 #define frq_default \
163 (BI (frq_mean) | BI (frq_stddev) | BI (frq_min) | BI (frq_max))
165 (BI (frq_sum) | BI(frq_min) | BI(frq_max) \
166 | BI(frq_mean) | BI(frq_semean) | BI(frq_stddev) \
167 | BI(frq_variance) | BI(frq_kurt) | BI(frq_sekurt) \
168 | BI(frq_skew) | BI(frq_seskew) | BI(frq_range) \
169 | BI(frq_range) | BI(frq_mode) | BI(frq_median))
171 /* Statistics; number of statistics. */
172 static unsigned long stats;
175 /* Types of graphs. */
178 GFT_NONE, /* Don't draw graphs. */
179 GFT_BAR, /* Draw bar charts. */
180 GFT_HIST, /* Draw histograms. */
181 GFT_PIE, /* Draw piechart */
182 GFT_HBAR /* Draw bar charts or histograms at our discretion. */
185 /* Parsed command. */
186 static struct cmd_frequencies cmd;
188 /* Summary of the barchart, histogram, and hbar subcommands. */
189 /* FIXME: These should not be mututally exclusive */
190 static int chart; /* NONE/BAR/HIST/HBAR/PIE. */
191 static double min, max; /* Minimum, maximum on y axis. */
192 static int format; /* FREQ/PERCENT: Scaling of y axis. */
193 static double scale, incr; /* FIXME */
194 static int normal; /* FIXME */
196 /* Variables for which to calculate statistics. */
197 static size_t n_variables;
198 static struct variable **v_variables;
200 /* Arenas used to store semi-permanent storage. */
201 static struct pool *int_pool; /* Integer mode. */
202 static struct pool *gen_pool; /* General mode. */
204 /* Frequency tables. */
206 /* Frequency table entry. */
209 union value *v; /* The value. */
210 double c; /* The number of occurrences of the value. */
213 /* Types of frequency tables. */
220 /* Entire frequency table. */
223 int mode; /* FRQM_GENERAL or FRQM_INTEGER. */
226 struct hsh_table *data; /* Undifferentiated data. */
229 double *vector; /* Frequencies proper. */
230 int min, max; /* The boundaries of the table. */
231 double out_of_range; /* Sum of weights of out-of-range values. */
232 double sysmis; /* Sum of weights of SYSMIS values. */
235 struct freq *valid; /* Valid freqs. */
236 int n_valid; /* Number of total freqs. */
238 struct freq *missing; /* Missing freqs. */
239 int n_missing; /* Number of missing freqs. */
242 double total_cases; /* Sum of weights of all cases. */
243 double valid_cases; /* Sum of weights of valid cases. */
247 /* Per-variable frequency data. */
250 /* Freqency table. */
251 struct freq_tab tab; /* Frequencies table to use. */
254 int n_groups; /* Number of groups. */
255 double *groups; /* Groups. */
258 double stat[frq_n_stats];
260 /* Width and format for analysis and display.
261 This is normally the same as "width" and "print" in struct
262 variable, but in SPSS-compatible mode only the first
263 MAX_SHORT_STRING bytes of long string variables are
266 struct fmt_spec print;
269 static inline struct var_freqs *
270 get_var_freqs (const struct variable *v)
273 assert (v->aux != NULL);
277 static void determine_charts (void);
279 static void calc_stats (struct variable *v, double d[frq_n_stats]);
281 static void precalc (const struct ccase *, void *, const struct dataset *);
282 static bool calc (const struct ccase *, void *, const struct dataset *);
283 static bool postcalc (void *, const struct dataset *);
285 static void postprocess_freq_tab (struct variable *);
286 static void dump_full (struct variable *);
287 static void dump_condensed (struct variable *);
288 static void dump_statistics (struct variable *, int show_varname);
289 static void cleanup_freq_tab (struct variable *);
291 static hsh_hash_func hash_value_numeric, hash_value_alpha;
292 static hsh_compare_func compare_value_numeric_a, compare_value_alpha_a;
293 static hsh_compare_func compare_value_numeric_d, compare_value_alpha_d;
294 static hsh_compare_func compare_freq_numeric_a, compare_freq_alpha_a;
295 static hsh_compare_func compare_freq_numeric_d, compare_freq_alpha_d;
298 static void do_piechart(const struct variable *var,
299 const struct freq_tab *frq_tab);
302 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var);
306 /* Parser and outline. */
308 static int internal_cmd_frequencies (struct lexer *lexer, struct dataset *ds);
311 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
315 int_pool = pool_create ();
316 result = internal_cmd_frequencies (lexer, ds);
317 pool_destroy (int_pool);
319 pool_destroy (gen_pool);
327 internal_cmd_frequencies (struct lexer *lexer, struct dataset *ds)
338 if (!parse_frequencies (lexer, ds, &cmd, NULL))
341 if (cmd.onepage_limit == NOT_LONG)
342 cmd.onepage_limit = 50;
344 /* Figure out statistics to calculate. */
346 if (cmd.a_statistics[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
347 stats |= frq_default;
348 if (cmd.a_statistics[FRQ_ST_ALL])
350 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
351 stats &= ~frq_median;
352 for (i = 0; i < frq_n_stats; i++)
353 if (cmd.a_statistics[st_name[i].st_indx])
354 stats |= BIT_INDEX (i);
355 if (stats & frq_kurt)
357 if (stats & frq_skew)
360 /* Calculate n_stats. */
362 for (i = 0; i < frq_n_stats; i++)
363 if ((stats & BIT_INDEX (i)))
368 if (chart != GFT_NONE || cmd.sbc_ntiles)
369 cmd.sort = FRQ_AVALUE;
371 /* Work out what percentiles need to be calculated */
372 if ( cmd.sbc_percentiles )
374 for ( i = 0 ; i < MAXLISTS ; ++i )
377 subc_list_double *ptl_list = &cmd.dl_percentiles[i];
378 for ( pl = 0 ; pl < subc_list_double_count(ptl_list); ++pl)
379 add_percentile (subc_list_double_at(ptl_list, pl) / 100.0 );
382 if ( cmd.sbc_ntiles )
384 for ( i = 0 ; i < cmd.sbc_ntiles ; ++i )
387 for (j = 0; j <= cmd.n_ntiles[i]; ++j )
388 add_percentile (j / (double) cmd.n_ntiles[i]);
394 ok = procedure_with_splits (ds, precalc, calc, postcalc, NULL);
396 free_frequencies(&cmd);
398 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
401 /* Figure out which charts the user requested. */
403 determine_charts (void)
405 int count = (!!cmd.sbc_histogram) + (!!cmd.sbc_barchart) +
406 (!!cmd.sbc_hbar) + (!!cmd.sbc_piechart);
416 msg (SW, _("At most one of BARCHART, HISTOGRAM, or HBAR should be "
417 "given. HBAR will be assumed. Argument values will be "
418 "given precedence increasing along the order given."));
420 else if (cmd.sbc_histogram)
422 else if (cmd.sbc_barchart)
424 else if (cmd.sbc_piechart)
435 if (cmd.sbc_barchart)
437 if (cmd.ba_min != SYSMIS)
439 if (cmd.ba_max != SYSMIS)
441 if (cmd.ba_scale == FRQ_FREQ)
446 else if (cmd.ba_scale == FRQ_PERCENT)
448 format = FRQ_PERCENT;
453 if (cmd.sbc_histogram)
455 if (cmd.hi_min != SYSMIS)
457 if (cmd.hi_max != SYSMIS)
459 if (cmd.hi_scale == FRQ_FREQ)
464 else if (cmd.hi_scale == FRQ_PERCENT)
466 format = FRQ_PERCENT;
469 if (cmd.hi_norm != FRQ_NONORMAL )
471 if (cmd.hi_incr == FRQ_INCREMENT)
477 if (cmd.hb_min != SYSMIS)
479 if (cmd.hb_max != SYSMIS)
481 if (cmd.hb_scale == FRQ_FREQ)
486 else if (cmd.hb_scale == FRQ_PERCENT)
488 format = FRQ_PERCENT;
493 if (cmd.hb_incr == FRQ_INCREMENT)
497 if (min != SYSMIS && max != SYSMIS && min >= max)
499 msg (SE, _("MAX must be greater than or equal to MIN, if both are "
500 "specified. However, MIN was specified as %g and MAX as %g. "
501 "MIN and MAX will be ignored."), min, max);
506 /* Add data from case C to the frequency table. */
508 calc (const struct ccase *c, void *aux UNUSED, const struct dataset *ds)
512 bool bad_warn = true;
514 weight = dict_get_case_weight (dataset_dict (ds), c, &bad_warn);
516 for (i = 0; i < n_variables; i++)
518 const struct variable *v = v_variables[i];
519 const union value *val = case_data (c, v->fv);
520 struct var_freqs *vf = get_var_freqs (v);
521 struct freq_tab *ft = &vf->tab;
531 target.v = (union value *) val;
532 fpp = (struct freq **) hsh_probe (ft->data, &target);
538 struct freq *fp = pool_alloc (gen_pool, sizeof *fp);
540 fp->v = pool_clone (gen_pool,
541 val, MAX (MAX_SHORT_STRING, vf->width));
548 if (val->f == SYSMIS)
549 ft->sysmis += weight;
550 else if (val->f > INT_MIN+1 && val->f < INT_MAX-1)
553 if (i >= ft->min && i <= ft->max)
554 ft->vector[i - ft->min] += weight;
557 ft->out_of_range += weight;
566 /* Prepares each variable that is the target of FREQUENCIES by setting
567 up its hash table. */
569 precalc (const struct ccase *first, void *aux UNUSED, const struct dataset *ds)
573 output_split_file_values (ds, first);
575 pool_destroy (gen_pool);
576 gen_pool = pool_create ();
578 for (i = 0; i < n_variables; i++)
580 struct variable *v = v_variables[i];
581 struct freq_tab *ft = &get_var_freqs (v)->tab;
583 if (ft->mode == FRQM_GENERAL)
586 hsh_compare_func *compare;
588 if (v->type == NUMERIC)
590 hash = hash_value_numeric;
591 compare = compare_value_numeric_a;
595 hash = hash_value_alpha;
596 compare = compare_value_alpha_a;
598 ft->data = hsh_create (16, compare, hash, NULL, v);
604 for (j = (ft->max - ft->min); j >= 0; j--)
606 ft->out_of_range = 0.0;
612 /* Finishes up with the variables after frequencies have been
613 calculated. Displays statistics, percentiles, ... */
615 postcalc (void *aux UNUSED, const struct dataset *ds UNUSED)
619 for (i = 0; i < n_variables; i++)
621 struct variable *v = v_variables[i];
622 struct var_freqs *vf = get_var_freqs (v);
623 struct freq_tab *ft = &vf->tab;
625 int dumped_freq_tab = 1;
627 postprocess_freq_tab (v);
629 /* Frequencies tables. */
630 n_categories = ft->n_valid + ft->n_missing;
631 if (cmd.table == FRQ_TABLE
632 || (cmd.table == FRQ_LIMIT && n_categories <= cmd.limit))
642 if (n_categories > cmd.onepage_limit)
655 dump_statistics (v, !dumped_freq_tab);
659 if ( chart == GFT_HIST)
661 double d[frq_n_stats];
662 struct normal_curve norm;
663 gsl_histogram *hist ;
666 norm.N = vf->tab.valid_cases;
669 norm.mean = d[frq_mean];
670 norm.stddev = d[frq_stddev];
672 hist = freq_tab_to_hist(ft,v);
674 histogram_plot(hist, var_to_string(v), &norm, normal);
676 gsl_histogram_free(hist);
680 if ( chart == GFT_PIE)
682 do_piechart(v_variables[i], ft);
687 cleanup_freq_tab (v);
694 /* Returns the comparison function that should be used for
695 sorting a frequency table by FRQ_SORT using VAR_TYPE
697 static hsh_compare_func *
698 get_freq_comparator (int frq_sort, int var_type)
700 /* Note that q2c generates tags beginning with 1000. */
701 switch (frq_sort | (var_type << 16))
703 case FRQ_AVALUE | (NUMERIC << 16): return compare_value_numeric_a;
704 case FRQ_AVALUE | (ALPHA << 16): return compare_value_alpha_a;
705 case FRQ_DVALUE | (NUMERIC << 16): return compare_value_numeric_d;
706 case FRQ_DVALUE | (ALPHA << 16): return compare_value_alpha_d;
707 case FRQ_AFREQ | (NUMERIC << 16): return compare_freq_numeric_a;
708 case FRQ_AFREQ | (ALPHA << 16): return compare_freq_alpha_a;
709 case FRQ_DFREQ | (NUMERIC << 16): return compare_freq_numeric_d;
710 case FRQ_DFREQ | (ALPHA << 16): return compare_freq_alpha_d;
711 default: NOT_REACHED ();
717 /* Returns true iff the value in struct freq F is non-missing
720 not_missing (const void *f_, const void *v_)
722 const struct freq *f = f_;
723 const struct variable *v = v_;
725 return !mv_is_value_missing (&v->miss, f->v);
728 /* Summarizes the frequency table data for variable V. */
730 postprocess_freq_tab (struct variable *v)
732 hsh_compare_func *compare;
736 struct freq *freqs, *f;
739 ft = &get_var_freqs (v)->tab;
740 assert (ft->mode == FRQM_GENERAL);
741 compare = get_freq_comparator (cmd.sort, v->type);
743 /* Extract data from hash table. */
744 count = hsh_count (ft->data);
745 data = hsh_data (ft->data);
747 /* Copy dereferenced data into freqs. */
748 freqs = xnmalloc (count, sizeof *freqs);
749 for (i = 0; i < count; i++)
751 struct freq *f = data[i];
755 /* Put data into ft. */
757 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, v);
758 ft->missing = freqs + ft->n_valid;
759 ft->n_missing = count - ft->n_valid;
762 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare, v);
763 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, v);
765 /* Summary statistics. */
766 ft->valid_cases = 0.0;
767 for(i = 0 ; i < ft->n_valid ; ++i )
770 ft->valid_cases += f->c;
774 ft->total_cases = ft->valid_cases ;
775 for(i = 0 ; i < ft->n_missing ; ++i )
778 ft->total_cases += f->c;
783 /* Frees the frequency table for variable V. */
785 cleanup_freq_tab (struct variable *v)
787 struct freq_tab *ft = &get_var_freqs (v)->tab;
788 assert (ft->mode == FRQM_GENERAL);
790 hsh_destroy (ft->data);
793 /* Parses the VARIABLES subcommand, adding to
794 {n_variables,v_variables}. */
796 frq_custom_variables (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *aux UNUSED)
799 int min = 0, max = 0;
801 size_t old_n_variables = n_variables;
804 lex_match (lexer, '=');
805 if (lex_token (lexer) != T_ALL && (lex_token (lexer) != T_ID
806 || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) == NULL))
809 if (!parse_variables (lexer, dataset_dict (ds), &v_variables, &n_variables,
810 PV_APPEND | PV_NO_SCRATCH))
813 if (!lex_match (lexer, '('))
818 if (!lex_force_int (lexer))
820 min = lex_integer (lexer);
822 if (!lex_force_match (lexer, ','))
824 if (!lex_force_int (lexer))
826 max = lex_integer (lexer);
828 if (!lex_force_match (lexer, ')'))
832 msg (SE, _("Upper limit of integer mode value range must be "
833 "greater than lower limit."));
838 for (i = old_n_variables; i < n_variables; i++)
840 struct variable *v = v_variables[i];
841 struct var_freqs *vf;
845 msg (SE, _("Variable %s specified multiple times on VARIABLES "
846 "subcommand."), v->name);
849 if (mode == FRQM_INTEGER && v->type != NUMERIC)
851 msg (SE, _("Integer mode specified, but %s is not a numeric "
852 "variable."), v->name);
856 vf = var_attach_aux (v, xmalloc (sizeof *vf), var_dtor_free);
858 vf->tab.valid = vf->tab.missing = NULL;
859 if (mode == FRQM_INTEGER)
863 vf->tab.vector = pool_nalloc (int_pool,
864 max - min + 1, sizeof *vf->tab.vector);
867 vf->tab.vector = NULL;
870 vf->width = v->width;
871 vf->print = v->print;
872 if (vf->width > MAX_SHORT_STRING && get_algorithm () == COMPATIBLE)
874 vf->width = MAX_SHORT_STRING;
875 vf->print.w = MAX_SHORT_STRING * (v->print.type == FMT_AHEX ? 2 : 1);
881 /* Parses the GROUPED subcommand, setting the n_grouped, grouped
882 fields of specified variables. */
884 frq_custom_grouped (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *aux UNUSED)
886 lex_match (lexer, '=');
887 if ((lex_token (lexer) == T_ID && dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
888 || lex_token (lexer) == T_ID)
893 /* Max, current size of list; list itself. */
901 if (!parse_variables (lexer, dataset_dict (ds), &v, &n,
902 PV_NO_DUPLICATE | PV_NUMERIC))
904 if (lex_match (lexer, '('))
908 while (lex_integer (lexer))
913 dl = pool_nrealloc (int_pool, dl, ml, sizeof *dl);
915 dl[nl++] = lex_tokval (lexer);
917 lex_match (lexer, ',');
919 /* Note that nl might still be 0 and dl might still be
920 NULL. That's okay. */
921 if (!lex_match (lexer, ')'))
924 msg (SE, _("`)' expected after GROUPED interval list."));
934 for (i = 0; i < n; i++)
935 if (v[i]->aux == NULL)
936 msg (SE, _("Variables %s specified on GROUPED but not on "
937 "VARIABLES."), v[i]->name);
940 struct var_freqs *vf = get_var_freqs (v[i]);
942 if (vf->groups != NULL)
943 msg (SE, _("Variables %s specified multiple times on GROUPED "
944 "subcommand."), v[i]->name);
952 if (!lex_match (lexer, '/'))
954 if ((lex_token (lexer) != T_ID || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
955 && lex_token (lexer) != T_ALL)
957 lex_put_back (lexer, '/');
965 /* Adds X to the list of percentiles, keeping the list in proper
968 add_percentile (double x)
972 for (i = 0; i < n_percentiles; i++)
974 /* Do nothing if it's already in the list */
975 if ( fabs(x - percentiles[i].p) < DBL_EPSILON )
978 if (x < percentiles[i].p)
982 if (i >= n_percentiles || x != percentiles[i].p)
984 percentiles = pool_nrealloc (int_pool, percentiles,
985 n_percentiles + 1, sizeof *percentiles);
987 if (i < n_percentiles)
988 memmove (&percentiles[i + 1], &percentiles[i],
989 (n_percentiles - i) * sizeof (struct percentile) );
991 percentiles[i].p = x;
996 /* Comparison functions. */
998 /* Hash of numeric values. */
1000 hash_value_numeric (const void *value_, const void *aux UNUSED)
1002 const struct freq *value = value_;
1003 return hsh_hash_double (value->v[0].f);
1006 /* Hash of string values. */
1008 hash_value_alpha (const void *value_, const void *v_)
1010 const struct freq *value = value_;
1011 const struct variable *v = v_;
1012 struct var_freqs *vf = get_var_freqs (v);
1014 return hsh_hash_bytes (value->v[0].s, vf->width);
1017 /* Ascending numeric compare of values. */
1019 compare_value_numeric_a (const void *a_, const void *b_, const void *aux UNUSED)
1021 const struct freq *a = a_;
1022 const struct freq *b = b_;
1024 if (a->v[0].f > b->v[0].f)
1026 else if (a->v[0].f < b->v[0].f)
1032 /* Ascending string compare of values. */
1034 compare_value_alpha_a (const void *a_, const void *b_, const void *v_)
1036 const struct freq *a = a_;
1037 const struct freq *b = b_;
1038 const struct variable *v = v_;
1039 struct var_freqs *vf = get_var_freqs (v);
1041 return memcmp (a->v[0].s, b->v[0].s, vf->width);
1044 /* Descending numeric compare of values. */
1046 compare_value_numeric_d (const void *a, const void *b, const void *aux UNUSED)
1048 return -compare_value_numeric_a (a, b, aux);
1051 /* Descending string compare of values. */
1053 compare_value_alpha_d (const void *a, const void *b, const void *v)
1055 return -compare_value_alpha_a (a, b, v);
1058 /* Ascending numeric compare of frequency;
1059 secondary key on ascending numeric value. */
1061 compare_freq_numeric_a (const void *a_, const void *b_, const void *aux UNUSED)
1063 const struct freq *a = a_;
1064 const struct freq *b = b_;
1068 else if (a->c < b->c)
1071 if (a->v[0].f > b->v[0].f)
1073 else if (a->v[0].f < b->v[0].f)
1079 /* Ascending numeric compare of frequency;
1080 secondary key on ascending string value. */
1082 compare_freq_alpha_a (const void *a_, const void *b_, const void *v_)
1084 const struct freq *a = a_;
1085 const struct freq *b = b_;
1086 const struct variable *v = v_;
1087 struct var_freqs *vf = get_var_freqs (v);
1091 else if (a->c < b->c)
1094 return memcmp (a->v[0].s, b->v[0].s, vf->width);
1097 /* Descending numeric compare of frequency;
1098 secondary key on ascending numeric value. */
1100 compare_freq_numeric_d (const void *a_, const void *b_, const void *aux UNUSED)
1102 const struct freq *a = a_;
1103 const struct freq *b = b_;
1107 else if (a->c < b->c)
1110 if (a->v[0].f > b->v[0].f)
1112 else if (a->v[0].f < b->v[0].f)
1118 /* Descending numeric compare of frequency;
1119 secondary key on ascending string value. */
1121 compare_freq_alpha_d (const void *a_, const void *b_, const void *v_)
1123 const struct freq *a = a_;
1124 const struct freq *b = b_;
1125 const struct variable *v = v_;
1126 struct var_freqs *vf = get_var_freqs (v);
1130 else if (a->c < b->c)
1133 return memcmp (a->v[0].s, b->v[0].s, vf->width);
1136 /* Frequency table display. */
1138 /* Sets the widths of all the columns and heights of all the rows in
1139 table T for driver D. */
1141 full_dim (struct tab_table *t, struct outp_driver *d)
1143 int lab = cmd.labels == FRQ_LABELS;
1147 t->w[0] = MIN (tab_natural_width (t, d, 0), d->prop_em_width * 15);
1148 for (i = lab; i < lab + 5; i++)
1149 t->w[i] = MAX (tab_natural_width (t, d, i), d->prop_em_width * 8);
1150 for (i = 0; i < t->nr; i++)
1151 t->h[i] = d->font_height;
1154 /* Displays a full frequency table for variable V. */
1156 dump_full (struct variable *v)
1159 struct var_freqs *vf;
1160 struct freq_tab *ft;
1162 struct tab_table *t;
1164 double cum_total = 0.0;
1165 double cum_freq = 0.0;
1173 const struct init *p;
1175 static const struct init vec[] =
1177 {4, 0, N_("Valid")},
1179 {1, 1, N_("Value")},
1180 {2, 1, N_("Frequency")},
1181 {3, 1, N_("Percent")},
1182 {4, 1, N_("Percent")},
1183 {5, 1, N_("Percent")},
1191 int lab = cmd.labels == FRQ_LABELS;
1193 vf = get_var_freqs (v);
1195 n_categories = ft->n_valid + ft->n_missing;
1196 t = tab_create (5 + lab, n_categories + 3, 0);
1197 tab_headers (t, 0, 0, 2, 0);
1198 tab_dim (t, full_dim);
1201 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value Label"));
1202 for (p = vec; p->s; p++)
1203 tab_text (t, p->c - (p->r ? !lab : 0), p->r,
1204 TAB_CENTER | TAT_TITLE, gettext (p->s));
1207 for (f = ft->valid; f < ft->missing; f++)
1209 double percent, valid_percent;
1213 percent = f->c / ft->total_cases * 100.0;
1214 valid_percent = f->c / ft->valid_cases * 100.0;
1215 cum_total += valid_percent;
1219 const char *label = val_labs_find (v->val_labs, f->v[0]);
1221 tab_text (t, 0, r, TAB_LEFT, label);
1224 tab_value (t, 0 + lab, r, TAB_NONE, f->v, &vf->print);
1225 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1226 tab_float (t, 2 + lab, r, TAB_NONE, percent, 5, 1);
1227 tab_float (t, 3 + lab, r, TAB_NONE, valid_percent, 5, 1);
1228 tab_float (t, 4 + lab, r, TAB_NONE, cum_total, 5, 1);
1231 for (; f < &ft->valid[n_categories]; f++)
1237 const char *label = val_labs_find (v->val_labs, f->v[0]);
1239 tab_text (t, 0, r, TAB_LEFT, label);
1242 tab_value (t, 0 + lab, r, TAB_NONE, f->v, &vf->print);
1243 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1244 tab_float (t, 2 + lab, r, TAB_NONE,
1245 f->c / ft->total_cases * 100.0, 5, 1);
1246 tab_text (t, 3 + lab, r, TAB_NONE, _("Missing"));
1250 tab_box (t, TAL_1, TAL_1,
1251 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1253 tab_hline (t, TAL_2, 0, 4 + lab, 2);
1254 tab_hline (t, TAL_2, 0, 4 + lab, r);
1255 tab_joint_text (t, 0, r, 0 + lab, r, TAB_RIGHT | TAT_TITLE, _("Total"));
1256 tab_vline (t, TAL_0, 1, r, r);
1257 tab_float (t, 1 + lab, r, TAB_NONE, cum_freq, 8, 0);
1258 tab_float (t, 2 + lab, r, TAB_NONE, 100.0, 5, 1);
1259 tab_float (t, 3 + lab, r, TAB_NONE, 100.0, 5, 1);
1261 tab_title (t, "%s: %s", v->name, v->label ? v->label : "");
1266 /* Sets the widths of all the columns and heights of all the rows in
1267 table T for driver D. */
1269 condensed_dim (struct tab_table *t, struct outp_driver *d)
1271 int cum_w = MAX (outp_string_width (d, _("Cum"), OUTP_PROPORTIONAL),
1272 MAX (outp_string_width (d, _("Cum"), OUTP_PROPORTIONAL),
1273 outp_string_width (d, "000", OUTP_PROPORTIONAL)));
1277 for (i = 0; i < 2; i++)
1278 t->w[i] = MAX (tab_natural_width (t, d, i), d->prop_em_width * 8);
1279 for (i = 2; i < 4; i++)
1281 for (i = 0; i < t->nr; i++)
1282 t->h[i] = d->font_height;
1285 /* Display condensed frequency table for variable V. */
1287 dump_condensed (struct variable *v)
1290 struct var_freqs *vf;
1291 struct freq_tab *ft;
1293 struct tab_table *t;
1295 double cum_total = 0.0;
1297 vf = get_var_freqs (v);
1299 n_categories = ft->n_valid + ft->n_missing;
1300 t = tab_create (4, n_categories + 2, 0);
1302 tab_headers (t, 0, 0, 2, 0);
1303 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value"));
1304 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Freq"));
1305 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1306 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Cum"));
1307 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1308 tab_dim (t, condensed_dim);
1311 for (f = ft->valid; f < ft->missing; f++)
1315 percent = f->c / ft->total_cases * 100.0;
1316 cum_total += f->c / ft->valid_cases * 100.0;
1318 tab_value (t, 0, r, TAB_NONE, f->v, &vf->print);
1319 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1320 tab_float (t, 2, r, TAB_NONE, percent, 3, 0);
1321 tab_float (t, 3, r, TAB_NONE, cum_total, 3, 0);
1324 for (; f < &ft->valid[n_categories]; f++)
1326 tab_value (t, 0, r, TAB_NONE, f->v, &vf->print);
1327 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1328 tab_float (t, 2, r, TAB_NONE,
1329 f->c / ft->total_cases * 100.0, 3, 0);
1333 tab_box (t, TAL_1, TAL_1,
1334 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1336 tab_hline (t, TAL_2, 0, 3, 2);
1337 tab_title (t, "%s: %s", v->name, v->label ? v->label : "");
1338 tab_columns (t, SOM_COL_DOWN, 1);
1342 /* Statistical display. */
1344 /* Calculates all the pertinent statistics for variable V, putting
1345 them in array D[]. FIXME: This could be made much more optimal. */
1347 calc_stats (struct variable *v, double d[frq_n_stats])
1349 struct freq_tab *ft = &get_var_freqs (v)->tab;
1350 double W = ft->valid_cases;
1359 double *median_value;
1361 /* Calculate percentiles. */
1363 /* If the 50th percentile was not explicitly requested then we must
1364 calculate it anyway --- it's the median */
1366 for (i = 0; i < n_percentiles; i++)
1368 if (percentiles[i].p == 0.5)
1370 median_value = &percentiles[i].value;
1375 if ( 0 == median_value )
1377 add_percentile (0.5);
1381 for (i = 0; i < n_percentiles; i++)
1383 percentiles[i].flag = 0;
1384 percentiles[i].flag2 = 0;
1388 for (idx = 0; idx < ft->n_valid; ++idx)
1390 static double prev_value = SYSMIS;
1391 f = &ft->valid[idx];
1393 for (i = 0; i < n_percentiles; i++)
1396 if ( percentiles[i].flag2 ) continue ;
1398 if ( get_algorithm() != COMPATIBLE )
1400 (ft->valid_cases - 1) * percentiles[i].p;
1403 (ft->valid_cases + 1) * percentiles[i].p - 1;
1405 if ( percentiles[i].flag )
1407 percentiles[i].x2 = f->v[0].f;
1408 percentiles[i].x1 = prev_value;
1409 percentiles[i].flag2 = 1;
1415 if ( f->c > 1 && rank - (f->c - 1) > tp )
1417 percentiles[i].x2 = percentiles[i].x1 = f->v[0].f;
1418 percentiles[i].flag2 = 1;
1422 percentiles[i].flag=1;
1428 prev_value = f->v[0].f;
1431 for (i = 0; i < n_percentiles; i++)
1433 /* Catches the case when p == 100% */
1434 if ( ! percentiles[i].flag2 )
1435 percentiles[i].x1 = percentiles[i].x2 = f->v[0].f;
1438 printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
1439 i,percentiles[i].p,percentiles[i].x1,percentiles[i].x2);
1443 for (i = 0; i < n_percentiles; i++)
1445 struct freq_tab *ft = &get_var_freqs (v)->tab;
1449 if ( get_algorithm() != COMPATIBLE )
1451 s = modf((ft->valid_cases - 1) * percentiles[i].p , &dummy);
1455 s = modf((ft->valid_cases + 1) * percentiles[i].p -1, &dummy);
1458 percentiles[i].value = percentiles[i].x1 +
1459 ( percentiles[i].x2 - percentiles[i].x1) * s ;
1461 if ( percentiles[i].p == 0.50)
1462 median_value = &percentiles[i].value;
1466 /* Calculate the mode. */
1469 for (f = ft->valid; f < ft->missing; f++)
1471 if (most_often < f->c)
1476 else if (most_often == f->c)
1478 /* A duplicate mode is undefined.
1479 FIXME: keep track of *all* the modes. */
1484 /* Calculate moments. */
1485 m = moments_create (MOMENT_KURTOSIS);
1486 for (f = ft->valid; f < ft->missing; f++)
1487 moments_pass_one (m, f->v[0].f, f->c);
1488 for (f = ft->valid; f < ft->missing; f++)
1489 moments_pass_two (m, f->v[0].f, f->c);
1490 moments_calculate (m, NULL, &d[frq_mean], &d[frq_variance],
1491 &d[frq_skew], &d[frq_kurt]);
1492 moments_destroy (m);
1494 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1495 d[frq_min] = ft->valid[0].v[0].f;
1496 d[frq_max] = ft->valid[ft->n_valid - 1].v[0].f;
1497 d[frq_mode] = X_mode;
1498 d[frq_range] = d[frq_max] - d[frq_min];
1499 d[frq_median] = *median_value;
1500 d[frq_sum] = d[frq_mean] * W;
1501 d[frq_stddev] = sqrt (d[frq_variance]);
1502 d[frq_semean] = d[frq_stddev] / sqrt (W);
1503 d[frq_seskew] = calc_seskew (W);
1504 d[frq_sekurt] = calc_sekurt (W);
1507 /* Displays a table of all the statistics requested for variable V. */
1509 dump_statistics (struct variable *v, int show_varname)
1511 struct freq_tab *ft;
1512 double stat_value[frq_n_stats];
1513 struct tab_table *t;
1516 int n_explicit_percentiles = n_percentiles;
1518 if ( implicit_50th && n_percentiles > 0 )
1521 if (v->type == ALPHA)
1523 ft = &get_var_freqs (v)->tab;
1524 if (ft->n_valid == 0)
1526 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1530 calc_stats (v, stat_value);
1532 t = tab_create (3, n_stats + n_explicit_percentiles + 2, 0);
1533 tab_dim (t, tab_natural_dimensions);
1535 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1538 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1539 tab_vline (t, TAL_GAP , 1, 0, tab_nr(t) - 1 ) ;
1541 r=2; /* N missing and N valid are always dumped */
1543 for (i = 0; i < frq_n_stats; i++)
1544 if (stats & BIT_INDEX (i))
1546 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1547 gettext (st_name[i].s10));
1548 tab_float (t, 2, r, TAB_NONE, stat_value[i], 11, 3);
1552 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1553 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1554 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1556 tab_float(t, 2, 0, TAB_NONE, ft->valid_cases, 11, 0);
1557 tab_float(t, 2, 1, TAB_NONE, ft->total_cases - ft->valid_cases, 11, 0);
1560 for (i = 0; i < n_explicit_percentiles; i++, r++)
1564 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1567 tab_float (t, 1, r, TAB_LEFT, percentiles[i].p * 100, 3, 0 );
1568 tab_float (t, 2, r, TAB_NONE, percentiles[i].value, 11, 3);
1572 tab_columns (t, SOM_COL_DOWN, 1);
1576 tab_title (t, "%s: %s", v->name, v->label);
1578 tab_title (t, "%s", v->name);
1581 tab_flags (t, SOMF_NO_TITLE);
1588 /* Create a gsl_histogram from a freq_tab */
1590 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var)
1593 double x_min = DBL_MAX;
1594 double x_max = -DBL_MAX;
1596 gsl_histogram *hist;
1597 const double bins = 11;
1599 struct hsh_iterator hi;
1600 struct hsh_table *fh = ft->data;
1603 /* Find out the extremes of the x value */
1604 for ( frq = hsh_first(fh, &hi); frq != 0; frq = hsh_next(fh, &hi) )
1606 if ( mv_is_value_missing(&var->miss, frq->v))
1609 if ( frq->v[0].f < x_min ) x_min = frq->v[0].f ;
1610 if ( frq->v[0].f > x_max ) x_max = frq->v[0].f ;
1613 hist = histogram_create(bins, x_min, x_max);
1615 for( i = 0 ; i < ft->n_valid ; ++i )
1617 frq = &ft->valid[i];
1618 gsl_histogram_accumulate(hist, frq->v[0].f, frq->c);
1625 static struct slice *
1626 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1627 const struct variable *var,
1631 /* Allocate an array of slices and fill them from the data in frq_tab
1632 n_slices will contain the number of slices allocated.
1633 The caller is responsible for freeing slices
1635 static struct slice *
1636 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1637 const struct variable *var,
1641 struct slice *slices;
1643 *n_slices = frq_tab->n_valid;
1645 slices = xnmalloc (*n_slices, sizeof *slices);
1647 for (i = 0 ; i < *n_slices ; ++i )
1649 const struct freq *frq = &frq_tab->valid[i];
1651 slices[i].label = value_to_string(frq->v, var);
1653 slices[i].magnetude = frq->c;
1663 do_piechart(const struct variable *var, const struct freq_tab *frq_tab)
1665 struct slice *slices;
1668 slices = freq_tab_to_slice_array(frq_tab, var, &n_slices);
1670 piechart_plot(var_to_string(var), slices, n_slices);