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 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 (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 *);
282 static bool calc (const struct ccase *, void *);
283 static void postcalc (void *);
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 (void);
311 cmd_frequencies (void)
315 int_pool = pool_create ();
316 result = internal_cmd_frequencies ();
317 pool_destroy (int_pool);
319 pool_destroy (gen_pool);
327 internal_cmd_frequencies (void)
338 if (!parse_frequencies (&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 (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)
514 weight = dict_get_case_weight (default_dict, c, &bad_warn);
516 for (i = 0; i < n_variables; i++)
518 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)
573 output_split_file_values (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)
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);
692 /* Returns the comparison function that should be used for
693 sorting a frequency table by FRQ_SORT using VAR_TYPE
695 static hsh_compare_func *
696 get_freq_comparator (int frq_sort, int var_type)
698 /* Note that q2c generates tags beginning with 1000. */
699 switch (frq_sort | (var_type << 16))
701 case FRQ_AVALUE | (NUMERIC << 16): return compare_value_numeric_a;
702 case FRQ_AVALUE | (ALPHA << 16): return compare_value_alpha_a;
703 case FRQ_DVALUE | (NUMERIC << 16): return compare_value_numeric_d;
704 case FRQ_DVALUE | (ALPHA << 16): return compare_value_alpha_d;
705 case FRQ_AFREQ | (NUMERIC << 16): return compare_freq_numeric_a;
706 case FRQ_AFREQ | (ALPHA << 16): return compare_freq_alpha_a;
707 case FRQ_DFREQ | (NUMERIC << 16): return compare_freq_numeric_d;
708 case FRQ_DFREQ | (ALPHA << 16): return compare_freq_alpha_d;
715 /* Returns nonzero iff the value in struct freq F is non-missing
718 not_missing (const void *f_, void *v_)
720 const struct freq *f = f_;
721 struct variable *v = v_;
723 return !mv_is_value_missing (&v->miss, f->v);
726 /* Summarizes the frequency table data for variable V. */
728 postprocess_freq_tab (struct variable *v)
730 hsh_compare_func *compare;
734 struct freq *freqs, *f;
737 ft = &get_var_freqs (v)->tab;
738 assert (ft->mode == FRQM_GENERAL);
739 compare = get_freq_comparator (cmd.sort, v->type);
741 /* Extract data from hash table. */
742 count = hsh_count (ft->data);
743 data = hsh_data (ft->data);
745 /* Copy dereferenced data into freqs. */
746 freqs = xnmalloc (count, sizeof *freqs);
747 for (i = 0; i < count; i++)
749 struct freq *f = data[i];
753 /* Put data into ft. */
755 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, v);
756 ft->missing = freqs + ft->n_valid;
757 ft->n_missing = count - ft->n_valid;
760 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare, v);
761 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, v);
763 /* Summary statistics. */
764 ft->valid_cases = 0.0;
765 for(i = 0 ; i < ft->n_valid ; ++i )
768 ft->valid_cases += f->c;
772 ft->total_cases = ft->valid_cases ;
773 for(i = 0 ; i < ft->n_missing ; ++i )
776 ft->total_cases += f->c;
781 /* Frees the frequency table for variable V. */
783 cleanup_freq_tab (struct variable *v)
785 struct freq_tab *ft = &get_var_freqs (v)->tab;
786 assert (ft->mode == FRQM_GENERAL);
788 hsh_destroy (ft->data);
791 /* Parses the VARIABLES subcommand, adding to
792 {n_variables,v_variables}. */
794 frq_custom_variables (struct cmd_frequencies *cmd UNUSED, void *aux UNUSED)
797 int min = 0, max = 0;
799 size_t old_n_variables = n_variables;
803 if (token != T_ALL && (token != T_ID
804 || dict_lookup_var (default_dict, tokid) == NULL))
807 if (!parse_variables (default_dict, &v_variables, &n_variables,
808 PV_APPEND | PV_NO_SCRATCH))
811 if (!lex_match ('('))
816 if (!lex_force_int ())
818 min = lex_integer ();
820 if (!lex_force_match (','))
822 if (!lex_force_int ())
824 max = lex_integer ();
826 if (!lex_force_match (')'))
830 msg (SE, _("Upper limit of integer mode value range must be "
831 "greater than lower limit."));
836 for (i = old_n_variables; i < n_variables; i++)
838 struct variable *v = v_variables[i];
839 struct var_freqs *vf;
843 msg (SE, _("Variable %s specified multiple times on VARIABLES "
844 "subcommand."), v->name);
847 if (mode == FRQM_INTEGER && v->type != NUMERIC)
849 msg (SE, _("Integer mode specified, but %s is not a numeric "
850 "variable."), v->name);
854 vf = var_attach_aux (v, xmalloc (sizeof *vf), var_dtor_free);
856 vf->tab.valid = vf->tab.missing = NULL;
857 if (mode == FRQM_INTEGER)
861 vf->tab.vector = pool_nalloc (int_pool,
862 max - min + 1, sizeof *vf->tab.vector);
865 vf->tab.vector = NULL;
868 vf->width = v->width;
869 vf->print = v->print;
870 if (vf->width > MAX_SHORT_STRING && get_algorithm () == COMPATIBLE)
872 vf->width = MAX_SHORT_STRING;
873 vf->print.w = MAX_SHORT_STRING * (v->print.type == FMT_AHEX ? 2 : 1);
879 /* Parses the GROUPED subcommand, setting the n_grouped, grouped
880 fields of specified variables. */
882 frq_custom_grouped (struct cmd_frequencies *cmd UNUSED, void *aux UNUSED)
885 if ((token == T_ID && dict_lookup_var (default_dict, tokid) != NULL)
891 /* Max, current size of list; list itself. */
899 if (!parse_variables (default_dict, &v, &n,
900 PV_NO_DUPLICATE | PV_NUMERIC))
906 while (lex_integer ())
911 dl = pool_nrealloc (int_pool, dl, ml, sizeof *dl);
917 /* Note that nl might still be 0 and dl might still be
918 NULL. That's okay. */
919 if (!lex_match (')'))
922 msg (SE, _("`)' expected after GROUPED interval list."));
932 for (i = 0; i < n; i++)
933 if (v[i]->aux == NULL)
934 msg (SE, _("Variables %s specified on GROUPED but not on "
935 "VARIABLES."), v[i]->name);
938 struct var_freqs *vf = get_var_freqs (v[i]);
940 if (vf->groups != NULL)
941 msg (SE, _("Variables %s specified multiple times on GROUPED "
942 "subcommand."), v[i]->name);
950 if (!lex_match ('/'))
952 if ((token != T_ID || dict_lookup_var (default_dict, tokid) != NULL)
963 /* Adds X to the list of percentiles, keeping the list in proper
966 add_percentile (double x)
970 for (i = 0; i < n_percentiles; i++)
972 /* Do nothing if it's already in the list */
973 if ( fabs(x - percentiles[i].p) < DBL_EPSILON )
976 if (x < percentiles[i].p)
980 if (i >= n_percentiles || tokval != percentiles[i].p)
982 percentiles = pool_nrealloc (int_pool, percentiles,
983 n_percentiles + 1, sizeof *percentiles);
985 if (i < n_percentiles)
986 memmove (&percentiles[i + 1], &percentiles[i],
987 (n_percentiles - i) * sizeof (struct percentile) );
989 percentiles[i].p = x;
994 /* Comparison functions. */
996 /* Hash of numeric values. */
998 hash_value_numeric (const void *value_, void *foo UNUSED)
1000 const struct freq *value = value_;
1001 return hsh_hash_double (value->v[0].f);
1004 /* Hash of string values. */
1006 hash_value_alpha (const void *value_, void *v_)
1008 const struct freq *value = value_;
1009 struct variable *v = v_;
1010 struct var_freqs *vf = get_var_freqs (v);
1012 return hsh_hash_bytes (value->v[0].s, vf->width);
1015 /* Ascending numeric compare of values. */
1017 compare_value_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
1019 const struct freq *a = a_;
1020 const struct freq *b = b_;
1022 if (a->v[0].f > b->v[0].f)
1024 else if (a->v[0].f < b->v[0].f)
1030 /* Ascending string compare of values. */
1032 compare_value_alpha_a (const void *a_, const void *b_, void *v_)
1034 const struct freq *a = a_;
1035 const struct freq *b = b_;
1036 struct variable *v = v_;
1037 struct var_freqs *vf = get_var_freqs (v);
1039 return memcmp (a->v[0].s, b->v[0].s, vf->width);
1042 /* Descending numeric compare of values. */
1044 compare_value_numeric_d (const void *a, const void *b, void *foo UNUSED)
1046 return -compare_value_numeric_a (a, b, foo);
1049 /* Descending string compare of values. */
1051 compare_value_alpha_d (const void *a, const void *b, void *v)
1053 return -compare_value_alpha_a (a, b, v);
1056 /* Ascending numeric compare of frequency;
1057 secondary key on ascending numeric value. */
1059 compare_freq_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
1061 const struct freq *a = a_;
1062 const struct freq *b = b_;
1066 else if (a->c < b->c)
1069 if (a->v[0].f > b->v[0].f)
1071 else if (a->v[0].f < b->v[0].f)
1077 /* Ascending numeric compare of frequency;
1078 secondary key on ascending string value. */
1080 compare_freq_alpha_a (const void *a_, const void *b_, void *v_)
1082 const struct freq *a = a_;
1083 const struct freq *b = b_;
1084 struct variable *v = v_;
1085 struct var_freqs *vf = get_var_freqs (v);
1089 else if (a->c < b->c)
1092 return memcmp (a->v[0].s, b->v[0].s, vf->width);
1095 /* Descending numeric compare of frequency;
1096 secondary key on ascending numeric value. */
1098 compare_freq_numeric_d (const void *a_, const void *b_, void *foo UNUSED)
1100 const struct freq *a = a_;
1101 const struct freq *b = b_;
1105 else if (a->c < b->c)
1108 if (a->v[0].f > b->v[0].f)
1110 else if (a->v[0].f < b->v[0].f)
1116 /* Descending numeric compare of frequency;
1117 secondary key on ascending string value. */
1119 compare_freq_alpha_d (const void *a_, const void *b_, void *v_)
1121 const struct freq *a = a_;
1122 const struct freq *b = b_;
1123 struct variable *v = v_;
1124 struct var_freqs *vf = get_var_freqs (v);
1128 else if (a->c < b->c)
1131 return memcmp (a->v[0].s, b->v[0].s, vf->width);
1134 /* Frequency table display. */
1136 /* Sets the widths of all the columns and heights of all the rows in
1137 table T for driver D. */
1139 full_dim (struct tab_table *t, struct outp_driver *d)
1141 int lab = cmd.labels == FRQ_LABELS;
1145 t->w[0] = min (tab_natural_width (t, d, 0), d->prop_em_width * 15);
1146 for (i = lab; i < lab + 5; i++)
1147 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
1148 for (i = 0; i < t->nr; i++)
1149 t->h[i] = d->font_height;
1152 /* Displays a full frequency table for variable V. */
1154 dump_full (struct variable *v)
1157 struct var_freqs *vf;
1158 struct freq_tab *ft;
1160 struct tab_table *t;
1162 double cum_total = 0.0;
1163 double cum_freq = 0.0;
1173 static struct init vec[] =
1175 {4, 0, N_("Valid")},
1177 {1, 1, N_("Value")},
1178 {2, 1, N_("Frequency")},
1179 {3, 1, N_("Percent")},
1180 {4, 1, N_("Percent")},
1181 {5, 1, N_("Percent")},
1189 int lab = cmd.labels == FRQ_LABELS;
1191 vf = get_var_freqs (v);
1193 n_categories = ft->n_valid + ft->n_missing;
1194 t = tab_create (5 + lab, n_categories + 3, 0);
1195 tab_headers (t, 0, 0, 2, 0);
1196 tab_dim (t, full_dim);
1199 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value Label"));
1200 for (p = vec; p->s; p++)
1201 tab_text (t, p->c - (p->r ? !lab : 0), p->r,
1202 TAB_CENTER | TAT_TITLE, gettext (p->s));
1205 for (f = ft->valid; f < ft->missing; f++)
1207 double percent, valid_percent;
1211 percent = f->c / ft->total_cases * 100.0;
1212 valid_percent = f->c / ft->valid_cases * 100.0;
1213 cum_total += valid_percent;
1217 const char *label = val_labs_find (v->val_labs, f->v[0]);
1219 tab_text (t, 0, r, TAB_LEFT, label);
1222 tab_value (t, 0 + lab, r, TAB_NONE, f->v, &vf->print);
1223 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1224 tab_float (t, 2 + lab, r, TAB_NONE, percent, 5, 1);
1225 tab_float (t, 3 + lab, r, TAB_NONE, valid_percent, 5, 1);
1226 tab_float (t, 4 + lab, r, TAB_NONE, cum_total, 5, 1);
1229 for (; f < &ft->valid[n_categories]; f++)
1235 const char *label = val_labs_find (v->val_labs, f->v[0]);
1237 tab_text (t, 0, r, TAB_LEFT, label);
1240 tab_value (t, 0 + lab, r, TAB_NONE, f->v, &vf->print);
1241 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1242 tab_float (t, 2 + lab, r, TAB_NONE,
1243 f->c / ft->total_cases * 100.0, 5, 1);
1244 tab_text (t, 3 + lab, r, TAB_NONE, _("Missing"));
1248 tab_box (t, TAL_1, TAL_1,
1249 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1251 tab_hline (t, TAL_2, 0, 4 + lab, 2);
1252 tab_hline (t, TAL_2, 0, 4 + lab, r);
1253 tab_joint_text (t, 0, r, 0 + lab, r, TAB_RIGHT | TAT_TITLE, _("Total"));
1254 tab_vline (t, TAL_0, 1, r, r);
1255 tab_float (t, 1 + lab, r, TAB_NONE, cum_freq, 8, 0);
1256 tab_float (t, 2 + lab, r, TAB_NONE, 100.0, 5, 1);
1257 tab_float (t, 3 + lab, r, TAB_NONE, 100.0, 5, 1);
1259 tab_title (t, "%s: %s", v->name, v->label ? v->label : "");
1264 /* Sets the widths of all the columns and heights of all the rows in
1265 table T for driver D. */
1267 condensed_dim (struct tab_table *t, struct outp_driver *d)
1269 int cum_w = max (outp_string_width (d, _("Cum"), OUTP_PROPORTIONAL),
1270 max (outp_string_width (d, _("Cum"), OUTP_PROPORTIONAL),
1271 outp_string_width (d, "000", OUTP_PROPORTIONAL)));
1275 for (i = 0; i < 2; i++)
1276 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
1277 for (i = 2; i < 4; i++)
1279 for (i = 0; i < t->nr; i++)
1280 t->h[i] = d->font_height;
1283 /* Display condensed frequency table for variable V. */
1285 dump_condensed (struct variable *v)
1288 struct var_freqs *vf;
1289 struct freq_tab *ft;
1291 struct tab_table *t;
1293 double cum_total = 0.0;
1295 vf = get_var_freqs (v);
1297 n_categories = ft->n_valid + ft->n_missing;
1298 t = tab_create (4, n_categories + 2, 0);
1300 tab_headers (t, 0, 0, 2, 0);
1301 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value"));
1302 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Freq"));
1303 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1304 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Cum"));
1305 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1306 tab_dim (t, condensed_dim);
1309 for (f = ft->valid; f < ft->missing; f++)
1313 percent = f->c / ft->total_cases * 100.0;
1314 cum_total += f->c / ft->valid_cases * 100.0;
1316 tab_value (t, 0, r, TAB_NONE, f->v, &vf->print);
1317 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1318 tab_float (t, 2, r, TAB_NONE, percent, 3, 0);
1319 tab_float (t, 3, r, TAB_NONE, cum_total, 3, 0);
1322 for (; f < &ft->valid[n_categories]; f++)
1324 tab_value (t, 0, r, TAB_NONE, f->v, &vf->print);
1325 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1326 tab_float (t, 2, r, TAB_NONE,
1327 f->c / ft->total_cases * 100.0, 3, 0);
1331 tab_box (t, TAL_1, TAL_1,
1332 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1334 tab_hline (t, TAL_2, 0, 3, 2);
1335 tab_title (t, "%s: %s", v->name, v->label ? v->label : "");
1336 tab_columns (t, SOM_COL_DOWN, 1);
1340 /* Statistical display. */
1342 /* Calculates all the pertinent statistics for variable V, putting
1343 them in array D[]. FIXME: This could be made much more optimal. */
1345 calc_stats (struct variable *v, double d[frq_n_stats])
1347 struct freq_tab *ft = &get_var_freqs (v)->tab;
1348 double W = ft->valid_cases;
1357 double *median_value;
1359 /* Calculate percentiles. */
1361 /* If the 50th percentile was not explicitly requested then we must
1362 calculate it anyway --- it's the median */
1364 for (i = 0; i < n_percentiles; i++)
1366 if (percentiles[i].p == 0.5)
1368 median_value = &percentiles[i].value;
1373 if ( 0 == median_value )
1375 add_percentile (0.5);
1379 for (i = 0; i < n_percentiles; i++)
1381 percentiles[i].flag = 0;
1382 percentiles[i].flag2 = 0;
1386 for (idx = 0; idx < ft->n_valid; ++idx)
1388 static double prev_value = SYSMIS;
1389 f = &ft->valid[idx];
1391 for (i = 0; i < n_percentiles; i++)
1394 if ( percentiles[i].flag2 ) continue ;
1396 if ( get_algorithm() != COMPATIBLE )
1398 (ft->valid_cases - 1) * percentiles[i].p;
1401 (ft->valid_cases + 1) * percentiles[i].p - 1;
1403 if ( percentiles[i].flag )
1405 percentiles[i].x2 = f->v[0].f;
1406 percentiles[i].x1 = prev_value;
1407 percentiles[i].flag2 = 1;
1413 if ( f->c > 1 && rank - (f->c - 1) > tp )
1415 percentiles[i].x2 = percentiles[i].x1 = f->v[0].f;
1416 percentiles[i].flag2 = 1;
1420 percentiles[i].flag=1;
1426 prev_value = f->v[0].f;
1429 for (i = 0; i < n_percentiles; i++)
1431 /* Catches the case when p == 100% */
1432 if ( ! percentiles[i].flag2 )
1433 percentiles[i].x1 = percentiles[i].x2 = f->v[0].f;
1436 printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
1437 i,percentiles[i].p,percentiles[i].x1,percentiles[i].x2);
1441 for (i = 0; i < n_percentiles; i++)
1443 struct freq_tab *ft = &get_var_freqs (v)->tab;
1447 if ( get_algorithm() != COMPATIBLE )
1449 s = modf((ft->valid_cases - 1) * percentiles[i].p , &dummy);
1453 s = modf((ft->valid_cases + 1) * percentiles[i].p -1, &dummy);
1456 percentiles[i].value = percentiles[i].x1 +
1457 ( percentiles[i].x2 - percentiles[i].x1) * s ;
1459 if ( percentiles[i].p == 0.50)
1460 median_value = &percentiles[i].value;
1464 /* Calculate the mode. */
1467 for (f = ft->valid; f < ft->missing; f++)
1469 if (most_often < f->c)
1474 else if (most_often == f->c)
1476 /* A duplicate mode is undefined.
1477 FIXME: keep track of *all* the modes. */
1482 /* Calculate moments. */
1483 m = moments_create (MOMENT_KURTOSIS);
1484 for (f = ft->valid; f < ft->missing; f++)
1485 moments_pass_one (m, f->v[0].f, f->c);
1486 for (f = ft->valid; f < ft->missing; f++)
1487 moments_pass_two (m, f->v[0].f, f->c);
1488 moments_calculate (m, NULL, &d[frq_mean], &d[frq_variance],
1489 &d[frq_skew], &d[frq_kurt]);
1490 moments_destroy (m);
1492 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1493 d[frq_min] = ft->valid[0].v[0].f;
1494 d[frq_max] = ft->valid[ft->n_valid - 1].v[0].f;
1495 d[frq_mode] = X_mode;
1496 d[frq_range] = d[frq_max] - d[frq_min];
1497 d[frq_median] = *median_value;
1498 d[frq_sum] = d[frq_mean] * W;
1499 d[frq_stddev] = sqrt (d[frq_variance]);
1500 d[frq_semean] = d[frq_stddev] / sqrt (W);
1501 d[frq_seskew] = calc_seskew (W);
1502 d[frq_sekurt] = calc_sekurt (W);
1505 /* Displays a table of all the statistics requested for variable V. */
1507 dump_statistics (struct variable *v, int show_varname)
1509 struct freq_tab *ft;
1510 double stat_value[frq_n_stats];
1511 struct tab_table *t;
1514 int n_explicit_percentiles = n_percentiles;
1516 if ( implicit_50th && n_percentiles > 0 )
1519 if (v->type == ALPHA)
1521 ft = &get_var_freqs (v)->tab;
1522 if (ft->n_valid == 0)
1524 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1528 calc_stats (v, stat_value);
1530 t = tab_create (3, n_stats + n_explicit_percentiles + 2, 0);
1531 tab_dim (t, tab_natural_dimensions);
1533 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1536 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1537 tab_vline (t, TAL_GAP , 1, 0, tab_nr(t) - 1 ) ;
1539 r=2; /* N missing and N valid are always dumped */
1541 for (i = 0; i < frq_n_stats; i++)
1542 if (stats & BIT_INDEX (i))
1544 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1545 gettext (st_name[i].s10));
1546 tab_float (t, 2, r, TAB_NONE, stat_value[i], 11, 3);
1550 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1551 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1552 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1554 tab_float(t, 2, 0, TAB_NONE, ft->valid_cases, 11, 0);
1555 tab_float(t, 2, 1, TAB_NONE, ft->total_cases - ft->valid_cases, 11, 0);
1558 for (i = 0; i < n_explicit_percentiles; i++, r++)
1562 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1565 tab_float (t, 1, r, TAB_LEFT, percentiles[i].p * 100, 3, 0 );
1566 tab_float (t, 2, r, TAB_NONE, percentiles[i].value, 11, 3);
1570 tab_columns (t, SOM_COL_DOWN, 1);
1574 tab_title (t, "%s: %s", v->name, v->label);
1576 tab_title (t, "%s", v->name);
1579 tab_flags (t, SOMF_NO_TITLE);
1586 /* Create a gsl_histogram from a freq_tab */
1588 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var)
1591 double x_min = DBL_MAX;
1592 double x_max = -DBL_MAX;
1594 gsl_histogram *hist;
1595 const double bins = 11;
1597 struct hsh_iterator hi;
1598 struct hsh_table *fh = ft->data;
1601 /* Find out the extremes of the x value */
1602 for ( frq = hsh_first(fh, &hi); frq != 0; frq = hsh_next(fh, &hi) )
1604 if ( mv_is_value_missing(&var->miss, frq->v))
1607 if ( frq->v[0].f < x_min ) x_min = frq->v[0].f ;
1608 if ( frq->v[0].f > x_max ) x_max = frq->v[0].f ;
1611 hist = histogram_create(bins, x_min, x_max);
1613 for( i = 0 ; i < ft->n_valid ; ++i )
1615 frq = &ft->valid[i];
1616 gsl_histogram_accumulate(hist, frq->v[0].f, frq->c);
1623 static struct slice *
1624 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1625 const struct variable *var,
1629 /* Allocate an array of slices and fill them from the data in frq_tab
1630 n_slices will contain the number of slices allocated.
1631 The caller is responsible for freeing slices
1633 static struct slice *
1634 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1635 const struct variable *var,
1639 struct slice *slices;
1641 *n_slices = frq_tab->n_valid;
1643 slices = xnmalloc (*n_slices, sizeof *slices);
1645 for (i = 0 ; i < *n_slices ; ++i )
1647 const struct freq *frq = &frq_tab->valid[i];
1649 slices[i].label = value_to_string(frq->v, var);
1651 slices[i].magnetude = frq->c;
1661 do_piechart(const struct variable *var, const struct freq_tab *frq_tab)
1663 struct slice *slices;
1666 slices = freq_tab_to_slice_array(frq_tab, var, &n_slices);
1668 piechart_plot(var_to_string(var), slices, n_slices);