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.
27 #include <libpspp/message.h>
30 #include <gsl/gsl_histogram.h>
32 #include <libpspp/alloc.h>
33 #include <libpspp/bit-vector.h>
34 #include <data/case.h>
35 #include <libpspp/compiler.h>
36 #include <data/dictionary.h>
37 #include <libpspp/hash.h>
38 #include <libpspp/pool.h>
39 #include <language/command.h>
40 #include <language/lexer/lexer.h>
41 #include <math/moments.h>
42 #include <libpspp/message.h>
43 #include <libpspp/array.h>
44 #include <libpspp/magic.h>
45 #include <libpspp/misc.h>
46 #include <output/output.h>
47 #include <output/manager.h>
48 #include <libpspp/str.h>
49 #include <output/table.h>
50 #include <data/value-labels.h>
51 #include <data/variable.h>
52 #include <procedure.h>
53 #include <data/settings.h>
54 #include <output/charts/piechart.h>
55 #include <output/chart.h>
56 #include <output/charts/plot-hist.h>
57 #include <math/histogram.h>
60 #define _(msgid) gettext (msgid)
61 #define N_(msgid) msgid
68 format=cond:condense/onepage(*n:onepage_limit,"%s>=0")/!standard,
69 table:limit(n:limit,"%s>0")/notable/!table,
70 labels:!labels/nolabels,
71 sort:!avalue/dvalue/afreq/dfreq,
72 spaces:!single/double,
73 paging:newpage/!oldpage;
74 missing=miss:include/!exclude;
75 barchart(ba_)=:minimum(d:min),
77 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0");
78 piechart(pie_)=:minimum(d:min),
80 missing:missing/!nomissing;
81 histogram(hi_)=:minimum(d:min),
83 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
84 norm:!nonormal/normal,
85 incr:increment(d:inc,"%s>0");
86 hbar(hb_)=: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");
93 +percentiles = double list;
94 statistics[st_]=1|mean,2|semean,3|median,4|mode,5|stddev,6|variance,
95 7|kurtosis,8|skewness,9|range,10|minimum,11|maximum,12|sum,
96 13|default,14|seskewness,15|sekurtosis,all,none.
104 frq_mean = 0, frq_semean, frq_median, frq_mode, frq_stddev, frq_variance,
105 frq_kurt, frq_sekurt, frq_skew, frq_seskew, frq_range, frq_min, frq_max,
109 /* Description of a statistic. */
112 int st_indx; /* Index into a_statistics[]. */
113 const char *s10; /* Identifying string. */
116 /* Table of statistics, indexed by dsc_*. */
117 static struct frq_info st_name[frq_n_stats + 1] =
119 {FRQ_ST_MEAN, N_("Mean")},
120 {FRQ_ST_SEMEAN, N_("S.E. Mean")},
121 {FRQ_ST_MEDIAN, N_("Median")},
122 {FRQ_ST_MODE, N_("Mode")},
123 {FRQ_ST_STDDEV, N_("Std Dev")},
124 {FRQ_ST_VARIANCE, N_("Variance")},
125 {FRQ_ST_KURTOSIS, N_("Kurtosis")},
126 {FRQ_ST_SEKURTOSIS, N_("S.E. Kurt")},
127 {FRQ_ST_SKEWNESS, N_("Skewness")},
128 {FRQ_ST_SESKEWNESS, N_("S.E. Skew")},
129 {FRQ_ST_RANGE, N_("Range")},
130 {FRQ_ST_MINIMUM, N_("Minimum")},
131 {FRQ_ST_MAXIMUM, N_("Maximum")},
132 {FRQ_ST_SUM, N_("Sum")},
136 /* Percentiles to calculate. */
140 double p; /* the %ile to be calculated */
141 double value; /* the %ile's value */
142 double x1; /* The datum value <= the percentile */
143 double x2; /* The datum value >= the percentile */
145 int flag2; /* Set to 1 if this percentile value has been found */
149 static void add_percentile (double x) ;
151 static struct percentile *percentiles;
152 static int n_percentiles;
154 static int implicit_50th ;
156 /* Groups of statistics. */
158 #define frq_default \
159 (BI (frq_mean) | BI (frq_stddev) | BI (frq_min) | BI (frq_max))
161 (BI (frq_sum) | BI(frq_min) | BI(frq_max) \
162 | BI(frq_mean) | BI(frq_semean) | BI(frq_stddev) \
163 | BI(frq_variance) | BI(frq_kurt) | BI(frq_sekurt) \
164 | BI(frq_skew) | BI(frq_seskew) | BI(frq_range) \
165 | BI(frq_range) | BI(frq_mode) | BI(frq_median))
167 /* Statistics; number of statistics. */
168 static unsigned long stats;
171 /* Types of graphs. */
174 GFT_NONE, /* Don't draw graphs. */
175 GFT_BAR, /* Draw bar charts. */
176 GFT_HIST, /* Draw histograms. */
177 GFT_PIE, /* Draw piechart */
178 GFT_HBAR /* Draw bar charts or histograms at our discretion. */
181 /* Parsed command. */
182 static struct cmd_frequencies cmd;
184 /* Summary of the barchart, histogram, and hbar subcommands. */
185 /* FIXME: These should not be mututally exclusive */
186 static int chart; /* NONE/BAR/HIST/HBAR/PIE. */
187 static double min, max; /* Minimum, maximum on y axis. */
188 static int format; /* FREQ/PERCENT: Scaling of y axis. */
189 static double scale, incr; /* FIXME */
190 static int normal; /* FIXME */
192 /* Variables for which to calculate statistics. */
193 static size_t n_variables;
194 static struct variable **v_variables;
196 /* Arenas used to store semi-permanent storage. */
197 static struct pool *int_pool; /* Integer mode. */
198 static struct pool *gen_pool; /* General mode. */
200 /* Frequency tables. */
202 /* Frequency table entry. */
205 union value v; /* The value. */
206 double c; /* The number of occurrences of the value. */
209 /* Types of frequency tables. */
216 /* Entire frequency table. */
219 int mode; /* FRQM_GENERAL or FRQM_INTEGER. */
222 struct hsh_table *data; /* Undifferentiated data. */
225 double *vector; /* Frequencies proper. */
226 int min, max; /* The boundaries of the table. */
227 double out_of_range; /* Sum of weights of out-of-range values. */
228 double sysmis; /* Sum of weights of SYSMIS values. */
231 struct freq *valid; /* Valid freqs. */
232 int n_valid; /* Number of total freqs. */
234 struct freq *missing; /* Missing freqs. */
235 int n_missing; /* Number of missing freqs. */
238 double total_cases; /* Sum of weights of all cases. */
239 double valid_cases; /* Sum of weights of valid cases. */
243 /* Per-variable frequency data. */
246 /* Freqency table. */
247 struct freq_tab tab; /* Frequencies table to use. */
250 int n_groups; /* Number of groups. */
251 double *groups; /* Groups. */
254 double stat[frq_n_stats];
257 static inline struct var_freqs *
258 get_var_freqs (struct variable *v)
261 assert (v->aux != NULL);
265 static void determine_charts (void);
267 static void calc_stats (struct variable *v, double d[frq_n_stats]);
269 static void precalc (void *);
270 static bool calc (struct ccase *, void *);
271 static void postcalc (void *);
273 static void postprocess_freq_tab (struct variable *);
274 static void dump_full (struct variable *);
275 static void dump_condensed (struct variable *);
276 static void dump_statistics (struct variable *, int show_varname);
277 static void cleanup_freq_tab (struct variable *);
279 static hsh_hash_func hash_value_numeric, hash_value_alpha;
280 static hsh_compare_func compare_value_numeric_a, compare_value_alpha_a;
281 static hsh_compare_func compare_value_numeric_d, compare_value_alpha_d;
282 static hsh_compare_func compare_freq_numeric_a, compare_freq_alpha_a;
283 static hsh_compare_func compare_freq_numeric_d, compare_freq_alpha_d;
286 static void do_piechart(const struct variable *var,
287 const struct freq_tab *frq_tab);
290 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var);
294 /* Parser and outline. */
296 static int internal_cmd_frequencies (void);
299 cmd_frequencies (void)
303 int_pool = pool_create ();
304 result = internal_cmd_frequencies ();
305 pool_destroy (int_pool);
307 pool_destroy (gen_pool);
315 internal_cmd_frequencies (void)
326 if (!parse_frequencies (&cmd))
329 if (cmd.onepage_limit == NOT_LONG)
330 cmd.onepage_limit = 50;
332 /* Figure out statistics to calculate. */
334 if (cmd.a_statistics[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
335 stats |= frq_default;
336 if (cmd.a_statistics[FRQ_ST_ALL])
338 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
339 stats &= ~frq_median;
340 for (i = 0; i < frq_n_stats; i++)
341 if (cmd.a_statistics[st_name[i].st_indx])
342 stats |= BIT_INDEX (i);
343 if (stats & frq_kurt)
345 if (stats & frq_skew)
348 /* Calculate n_stats. */
350 for (i = 0; i < frq_n_stats; i++)
351 if ((stats & BIT_INDEX (i)))
356 if (chart != GFT_NONE || cmd.sbc_ntiles)
357 cmd.sort = FRQ_AVALUE;
359 /* Work out what percentiles need to be calculated */
360 if ( cmd.sbc_percentiles )
362 for ( i = 0 ; i < MAXLISTS ; ++i )
365 subc_list_double *ptl_list = &cmd.dl_percentiles[i];
366 for ( pl = 0 ; pl < subc_list_double_count(ptl_list); ++pl)
367 add_percentile(subc_list_double_at(ptl_list,pl) / 100.0 );
370 if ( cmd.sbc_ntiles )
372 for ( i = 0 ; i < cmd.sbc_ntiles ; ++i )
375 for (j = 0; j <= cmd.n_ntiles[i]; ++j )
376 add_percentile(j / (double) cmd.n_ntiles[i]);
382 ok = procedure_with_splits (precalc, calc, postcalc, NULL);
384 free_frequencies(&cmd);
386 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
389 /* Figure out which charts the user requested. */
391 determine_charts (void)
393 int count = (!!cmd.sbc_histogram) + (!!cmd.sbc_barchart) +
394 (!!cmd.sbc_hbar) + (!!cmd.sbc_piechart);
404 msg (SW, _("At most one of BARCHART, HISTOGRAM, or HBAR should be "
405 "given. HBAR will be assumed. Argument values will be "
406 "given precedence increasing along the order given."));
408 else if (cmd.sbc_histogram)
410 else if (cmd.sbc_barchart)
412 else if (cmd.sbc_piechart)
423 if (cmd.sbc_barchart)
425 if (cmd.ba_min != SYSMIS)
427 if (cmd.ba_max != SYSMIS)
429 if (cmd.ba_scale == FRQ_FREQ)
434 else if (cmd.ba_scale == FRQ_PERCENT)
436 format = FRQ_PERCENT;
441 if (cmd.sbc_histogram)
443 if (cmd.hi_min != SYSMIS)
445 if (cmd.hi_max != SYSMIS)
447 if (cmd.hi_scale == FRQ_FREQ)
452 else if (cmd.hi_scale == FRQ_PERCENT)
454 format = FRQ_PERCENT;
457 if (cmd.hi_norm != FRQ_NONORMAL )
459 if (cmd.hi_incr == FRQ_INCREMENT)
465 if (cmd.hb_min != SYSMIS)
467 if (cmd.hb_max != SYSMIS)
469 if (cmd.hb_scale == FRQ_FREQ)
474 else if (cmd.hb_scale == FRQ_PERCENT)
476 format = FRQ_PERCENT;
481 if (cmd.hb_incr == FRQ_INCREMENT)
485 if (min != SYSMIS && max != SYSMIS && min >= max)
487 msg (SE, _("MAX must be greater than or equal to MIN, if both are "
488 "specified. However, MIN was specified as %g and MAX as %g. "
489 "MIN and MAX will be ignored."), min, max);
494 /* Add data from case C to the frequency table. */
496 calc (struct ccase *c, void *aux UNUSED)
502 weight = dict_get_case_weight (default_dict, c, &bad_warn);
504 for (i = 0; i < n_variables; i++)
506 struct variable *v = v_variables[i];
507 const union value *val = case_data (c, v->fv);
508 struct freq_tab *ft = &get_var_freqs (v)->tab;
516 struct freq **fpp = (struct freq **) hsh_probe (ft->data, val);
522 struct freq *fp = *fpp = pool_alloc (gen_pool, sizeof *fp);
530 if (val->f == SYSMIS)
531 ft->sysmis += weight;
532 else if (val->f > INT_MIN+1 && val->f < INT_MAX-1)
535 if (i >= ft->min && i <= ft->max)
536 ft->vector[i - ft->min] += weight;
539 ft->out_of_range += weight;
548 /* Prepares each variable that is the target of FREQUENCIES by setting
549 up its hash table. */
551 precalc (void *aux UNUSED)
555 pool_destroy (gen_pool);
556 gen_pool = pool_create ();
558 for (i = 0; i < n_variables; i++)
560 struct variable *v = v_variables[i];
561 struct freq_tab *ft = &get_var_freqs (v)->tab;
563 if (ft->mode == FRQM_GENERAL)
566 hsh_compare_func *compare;
568 if (v->type == NUMERIC)
570 hash = hash_value_numeric;
571 compare = compare_value_numeric_a;
575 hash = hash_value_alpha;
576 compare = compare_value_alpha_a;
578 ft->data = hsh_create (16, compare, hash, NULL, v);
584 for (j = (ft->max - ft->min); j >= 0; j--)
586 ft->out_of_range = 0.0;
592 /* Finishes up with the variables after frequencies have been
593 calculated. Displays statistics, percentiles, ... */
595 postcalc (void *aux UNUSED)
599 for (i = 0; i < n_variables; i++)
601 struct variable *v = v_variables[i];
602 struct var_freqs *vf = get_var_freqs (v);
603 struct freq_tab *ft = &vf->tab;
605 int dumped_freq_tab = 1;
607 postprocess_freq_tab (v);
609 /* Frequencies tables. */
610 n_categories = ft->n_valid + ft->n_missing;
611 if (cmd.table == FRQ_TABLE
612 || (cmd.table == FRQ_LIMIT && n_categories <= cmd.limit))
622 if (n_categories > cmd.onepage_limit)
635 dump_statistics (v, !dumped_freq_tab);
639 if ( chart == GFT_HIST)
641 double d[frq_n_stats];
642 struct normal_curve norm;
643 gsl_histogram *hist ;
646 norm.N = vf->tab.valid_cases;
649 norm.mean = d[frq_mean];
650 norm.stddev = d[frq_stddev];
652 hist = freq_tab_to_hist(ft,v);
654 histogram_plot(hist, var_to_string(v), &norm, normal);
656 gsl_histogram_free(hist);
660 if ( chart == GFT_PIE)
662 do_piechart(v_variables[i], ft);
667 cleanup_freq_tab (v);
672 /* Returns the comparison function that should be used for
673 sorting a frequency table by FRQ_SORT using VAR_TYPE
675 static hsh_compare_func *
676 get_freq_comparator (int frq_sort, int var_type)
678 /* Note that q2c generates tags beginning with 1000. */
679 switch (frq_sort | (var_type << 16))
681 case FRQ_AVALUE | (NUMERIC << 16): return compare_value_numeric_a;
682 case FRQ_AVALUE | (ALPHA << 16): return compare_value_alpha_a;
683 case FRQ_DVALUE | (NUMERIC << 16): return compare_value_numeric_d;
684 case FRQ_DVALUE | (ALPHA << 16): return compare_value_alpha_d;
685 case FRQ_AFREQ | (NUMERIC << 16): return compare_freq_numeric_a;
686 case FRQ_AFREQ | (ALPHA << 16): return compare_freq_alpha_a;
687 case FRQ_DFREQ | (NUMERIC << 16): return compare_freq_numeric_d;
688 case FRQ_DFREQ | (ALPHA << 16): return compare_freq_alpha_d;
695 /* Returns nonzero iff the value in struct freq F is non-missing
698 not_missing (const void *f_, void *v_)
700 const struct freq *f = f_;
701 struct variable *v = v_;
703 return !mv_is_value_missing (&v->miss, &f->v);
706 /* Summarizes the frequency table data for variable V. */
708 postprocess_freq_tab (struct variable *v)
710 hsh_compare_func *compare;
714 struct freq *freqs, *f;
717 ft = &get_var_freqs (v)->tab;
718 assert (ft->mode == FRQM_GENERAL);
719 compare = get_freq_comparator (cmd.sort, v->type);
721 /* Extract data from hash table. */
722 count = hsh_count (ft->data);
723 data = hsh_data (ft->data);
725 /* Copy dereferenced data into freqs. */
726 freqs = xnmalloc (count, sizeof *freqs);
727 for (i = 0; i < count; i++)
729 struct freq *f = data[i];
733 /* Put data into ft. */
735 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, v);
736 ft->missing = freqs + ft->n_valid;
737 ft->n_missing = count - ft->n_valid;
740 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare, v);
741 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, v);
743 /* Summary statistics. */
744 ft->valid_cases = 0.0;
745 for(i = 0 ; i < ft->n_valid ; ++i )
748 ft->valid_cases += f->c;
752 ft->total_cases = ft->valid_cases ;
753 for(i = 0 ; i < ft->n_missing ; ++i )
756 ft->total_cases += f->c;
761 /* Frees the frequency table for variable V. */
763 cleanup_freq_tab (struct variable *v)
765 struct freq_tab *ft = &get_var_freqs (v)->tab;
766 assert (ft->mode == FRQM_GENERAL);
768 hsh_destroy (ft->data);
771 /* Parses the VARIABLES subcommand, adding to
772 {n_variables,v_variables}. */
774 frq_custom_variables (struct cmd_frequencies *cmd UNUSED)
777 int min = 0, max = 0;
779 size_t old_n_variables = n_variables;
783 if (token != T_ALL && (token != T_ID
784 || dict_lookup_var (default_dict, tokid) == NULL))
787 if (!parse_variables (default_dict, &v_variables, &n_variables,
788 PV_APPEND | PV_NO_SCRATCH))
791 if (!lex_match ('('))
796 if (!lex_force_int ())
798 min = lex_integer ();
800 if (!lex_force_match (','))
802 if (!lex_force_int ())
804 max = lex_integer ();
806 if (!lex_force_match (')'))
810 msg (SE, _("Upper limit of integer mode value range must be "
811 "greater than lower limit."));
816 for (i = old_n_variables; i < n_variables; i++)
818 struct variable *v = v_variables[i];
819 struct var_freqs *vf;
823 msg (SE, _("Variable %s specified multiple times on VARIABLES "
824 "subcommand."), v->name);
827 if (mode == FRQM_INTEGER && v->type != NUMERIC)
829 msg (SE, _("Integer mode specified, but %s is not a numeric "
830 "variable."), v->name);
834 vf = var_attach_aux (v, xmalloc (sizeof *vf), var_dtor_free);
836 vf->tab.valid = vf->tab.missing = NULL;
837 if (mode == FRQM_INTEGER)
841 vf->tab.vector = pool_nalloc (int_pool,
842 max - min + 1, sizeof *vf->tab.vector);
845 vf->tab.vector = NULL;
852 /* Parses the GROUPED subcommand, setting the n_grouped, grouped
853 fields of specified variables. */
855 frq_custom_grouped (struct cmd_frequencies *cmd UNUSED)
858 if ((token == T_ID && dict_lookup_var (default_dict, tokid) != NULL)
864 /* Max, current size of list; list itself. */
872 if (!parse_variables (default_dict, &v, &n,
873 PV_NO_DUPLICATE | PV_NUMERIC))
879 while (lex_integer ())
884 dl = pool_nrealloc (int_pool, dl, ml, sizeof *dl);
890 /* Note that nl might still be 0 and dl might still be
891 NULL. That's okay. */
892 if (!lex_match (')'))
895 msg (SE, _("`)' expected after GROUPED interval list."));
905 for (i = 0; i < n; i++)
906 if (v[i]->aux == NULL)
907 msg (SE, _("Variables %s specified on GROUPED but not on "
908 "VARIABLES."), v[i]->name);
911 struct var_freqs *vf = get_var_freqs (v[i]);
913 if (vf->groups != NULL)
914 msg (SE, _("Variables %s specified multiple times on GROUPED "
915 "subcommand."), v[i]->name);
923 if (!lex_match ('/'))
925 if ((token != T_ID || dict_lookup_var (default_dict, tokid) != NULL)
936 /* Adds X to the list of percentiles, keeping the list in proper
939 add_percentile (double x)
943 for (i = 0; i < n_percentiles; i++)
945 /* Do nothing if it's already in the list */
946 if ( fabs(x - percentiles[i].p) < DBL_EPSILON )
949 if (x < percentiles[i].p)
953 if (i >= n_percentiles || tokval != percentiles[i].p)
955 percentiles = pool_nrealloc (int_pool, percentiles,
956 n_percentiles + 1, sizeof *percentiles);
958 if (i < n_percentiles)
959 memmove (&percentiles[i + 1], &percentiles[i],
960 (n_percentiles - i) * sizeof (struct percentile) );
962 percentiles[i].p = x;
967 /* Comparison functions. */
969 /* Hash of numeric values. */
971 hash_value_numeric (const void *value_, void *foo UNUSED)
973 const struct freq *value = value_;
974 return hsh_hash_double (value->v.f);
977 /* Hash of string values. */
979 hash_value_alpha (const void *value_, void *v_)
981 const struct freq *value = value_;
982 struct variable *v = v_;
984 return hsh_hash_bytes (value->v.s, v->width);
987 /* Ascending numeric compare of values. */
989 compare_value_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
991 const struct freq *a = a_;
992 const struct freq *b = b_;
996 else if (a->v.f < b->v.f)
1002 /* Ascending string compare of values. */
1004 compare_value_alpha_a (const void *a_, const void *b_, void *v_)
1006 const struct freq *a = a_;
1007 const struct freq *b = b_;
1008 const struct variable *v = v_;
1010 return memcmp (a->v.s, b->v.s, v->width);
1013 /* Descending numeric compare of values. */
1015 compare_value_numeric_d (const void *a, const void *b, void *foo UNUSED)
1017 return -compare_value_numeric_a (a, b, foo);
1020 /* Descending string compare of values. */
1022 compare_value_alpha_d (const void *a, const void *b, void *v)
1024 return -compare_value_alpha_a (a, b, v);
1027 /* Ascending numeric compare of frequency;
1028 secondary key on ascending numeric value. */
1030 compare_freq_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
1032 const struct freq *a = a_;
1033 const struct freq *b = b_;
1037 else if (a->c < b->c)
1040 if (a->v.f > b->v.f)
1042 else if (a->v.f < b->v.f)
1048 /* Ascending numeric compare of frequency;
1049 secondary key on ascending string value. */
1051 compare_freq_alpha_a (const void *a_, const void *b_, void *v_)
1053 const struct freq *a = a_;
1054 const struct freq *b = b_;
1055 const struct variable *v = v_;
1059 else if (a->c < b->c)
1062 return memcmp (a->v.s, b->v.s, v->width);
1065 /* Descending numeric compare of frequency;
1066 secondary key on ascending numeric value. */
1068 compare_freq_numeric_d (const void *a_, const void *b_, void *foo UNUSED)
1070 const struct freq *a = a_;
1071 const struct freq *b = b_;
1075 else if (a->c < b->c)
1078 if (a->v.f > b->v.f)
1080 else if (a->v.f < b->v.f)
1086 /* Descending numeric compare of frequency;
1087 secondary key on ascending string value. */
1089 compare_freq_alpha_d (const void *a_, const void *b_, void *v_)
1091 const struct freq *a = a_;
1092 const struct freq *b = b_;
1093 const struct variable *v = v_;
1097 else if (a->c < b->c)
1100 return memcmp (a->v.s, b->v.s, v->width);
1103 /* Frequency table display. */
1105 /* Sets the widths of all the columns and heights of all the rows in
1106 table T for driver D. */
1108 full_dim (struct tab_table *t, struct outp_driver *d)
1110 int lab = cmd.labels == FRQ_LABELS;
1114 t->w[0] = min (tab_natural_width (t, d, 0), d->prop_em_width * 15);
1115 for (i = lab; i < lab + 5; i++)
1116 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
1117 for (i = 0; i < t->nr; i++)
1118 t->h[i] = d->font_height;
1121 /* Displays a full frequency table for variable V. */
1123 dump_full (struct variable *v)
1126 struct freq_tab *ft;
1128 struct tab_table *t;
1130 double cum_total = 0.0;
1131 double cum_freq = 0.0;
1141 static struct init vec[] =
1143 {4, 0, N_("Valid")},
1145 {1, 1, N_("Value")},
1146 {2, 1, N_("Frequency")},
1147 {3, 1, N_("Percent")},
1148 {4, 1, N_("Percent")},
1149 {5, 1, N_("Percent")},
1157 int lab = cmd.labels == FRQ_LABELS;
1159 ft = &get_var_freqs (v)->tab;
1160 n_categories = ft->n_valid + ft->n_missing;
1161 t = tab_create (5 + lab, n_categories + 3, 0);
1162 tab_headers (t, 0, 0, 2, 0);
1163 tab_dim (t, full_dim);
1166 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value Label"));
1167 for (p = vec; p->s; p++)
1168 tab_text (t, p->c - (p->r ? !lab : 0), p->r,
1169 TAB_CENTER | TAT_TITLE, gettext (p->s));
1172 for (f = ft->valid; f < ft->missing; f++)
1174 double percent, valid_percent;
1178 percent = f->c / ft->total_cases * 100.0;
1179 valid_percent = f->c / ft->valid_cases * 100.0;
1180 cum_total += valid_percent;
1184 const char *label = val_labs_find (v->val_labs, f->v);
1186 tab_text (t, 0, r, TAB_LEFT, label);
1189 tab_value (t, 0 + lab, r, TAB_NONE, &f->v, &v->print);
1190 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1191 tab_float (t, 2 + lab, r, TAB_NONE, percent, 5, 1);
1192 tab_float (t, 3 + lab, r, TAB_NONE, valid_percent, 5, 1);
1193 tab_float (t, 4 + lab, r, TAB_NONE, cum_total, 5, 1);
1196 for (; f < &ft->valid[n_categories]; f++)
1202 const char *label = val_labs_find (v->val_labs, f->v);
1204 tab_text (t, 0, r, TAB_LEFT, label);
1207 tab_value (t, 0 + lab, r, TAB_NONE, &f->v, &v->print);
1208 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1209 tab_float (t, 2 + lab, r, TAB_NONE,
1210 f->c / ft->total_cases * 100.0, 5, 1);
1211 tab_text (t, 3 + lab, r, TAB_NONE, _("Missing"));
1215 tab_box (t, TAL_1, TAL_1,
1216 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1218 tab_hline (t, TAL_2, 0, 4 + lab, 2);
1219 tab_hline (t, TAL_2, 0, 4 + lab, r);
1220 tab_joint_text (t, 0, r, 0 + lab, r, TAB_RIGHT | TAT_TITLE, _("Total"));
1221 tab_vline (t, TAL_0, 1, r, r);
1222 tab_float (t, 1 + lab, r, TAB_NONE, cum_freq, 8, 0);
1223 tab_float (t, 2 + lab, r, TAB_NONE, 100.0, 5, 1);
1224 tab_float (t, 3 + lab, r, TAB_NONE, 100.0, 5, 1);
1226 tab_title (t, "%s: %s", v->name, v->label ? v->label : "");
1231 /* Sets the widths of all the columns and heights of all the rows in
1232 table T for driver D. */
1234 condensed_dim (struct tab_table *t, struct outp_driver *d)
1236 int cum_w = max (outp_string_width (d, _("Cum"), OUTP_PROPORTIONAL),
1237 max (outp_string_width (d, _("Cum"), OUTP_PROPORTIONAL),
1238 outp_string_width (d, "000", OUTP_PROPORTIONAL)));
1242 for (i = 0; i < 2; i++)
1243 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
1244 for (i = 2; i < 4; i++)
1246 for (i = 0; i < t->nr; i++)
1247 t->h[i] = d->font_height;
1250 /* Display condensed frequency table for variable V. */
1252 dump_condensed (struct variable *v)
1255 struct freq_tab *ft;
1257 struct tab_table *t;
1259 double cum_total = 0.0;
1261 ft = &get_var_freqs (v)->tab;
1262 n_categories = ft->n_valid + ft->n_missing;
1263 t = tab_create (4, n_categories + 2, 0);
1265 tab_headers (t, 0, 0, 2, 0);
1266 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value"));
1267 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Freq"));
1268 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1269 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Cum"));
1270 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1271 tab_dim (t, condensed_dim);
1274 for (f = ft->valid; f < ft->missing; f++)
1278 percent = f->c / ft->total_cases * 100.0;
1279 cum_total += f->c / ft->valid_cases * 100.0;
1281 tab_value (t, 0, r, TAB_NONE, &f->v, &v->print);
1282 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1283 tab_float (t, 2, r, TAB_NONE, percent, 3, 0);
1284 tab_float (t, 3, r, TAB_NONE, cum_total, 3, 0);
1287 for (; f < &ft->valid[n_categories]; f++)
1289 tab_value (t, 0, r, TAB_NONE, &f->v, &v->print);
1290 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1291 tab_float (t, 2, r, TAB_NONE,
1292 f->c / ft->total_cases * 100.0, 3, 0);
1296 tab_box (t, TAL_1, TAL_1,
1297 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1299 tab_hline (t, TAL_2, 0, 3, 2);
1300 tab_title (t, "%s: %s", v->name, v->label ? v->label : "");
1301 tab_columns (t, SOM_COL_DOWN, 1);
1305 /* Statistical display. */
1307 /* Calculates all the pertinent statistics for variable V, putting
1308 them in array D[]. FIXME: This could be made much more optimal. */
1310 calc_stats (struct variable *v, double d[frq_n_stats])
1312 struct freq_tab *ft = &get_var_freqs (v)->tab;
1313 double W = ft->valid_cases;
1322 double *median_value;
1324 /* Calculate percentiles. */
1326 /* If the 50th percentile was not explicitly requested then we must
1327 calculate it anyway --- it's the median */
1329 for (i = 0; i < n_percentiles; i++)
1331 if (percentiles[i].p == 0.5)
1333 median_value = &percentiles[i].value;
1338 if ( 0 == median_value )
1340 add_percentile (0.5);
1344 for (i = 0; i < n_percentiles; i++)
1346 percentiles[i].flag = 0;
1347 percentiles[i].flag2 = 0;
1351 for (idx = 0; idx < ft->n_valid; ++idx)
1353 static double prev_value = SYSMIS;
1354 f = &ft->valid[idx];
1356 for (i = 0; i < n_percentiles; i++)
1359 if ( percentiles[i].flag2 ) continue ;
1361 if ( get_algorithm() != COMPATIBLE )
1363 (ft->valid_cases - 1) * percentiles[i].p;
1366 (ft->valid_cases + 1) * percentiles[i].p - 1;
1368 if ( percentiles[i].flag )
1370 percentiles[i].x2 = f->v.f;
1371 percentiles[i].x1 = prev_value;
1372 percentiles[i].flag2 = 1;
1378 if ( f->c > 1 && rank - (f->c - 1) > tp )
1380 percentiles[i].x2 = percentiles[i].x1 = f->v.f;
1381 percentiles[i].flag2 = 1;
1385 percentiles[i].flag=1;
1391 prev_value = f->v.f;
1394 for (i = 0; i < n_percentiles; i++)
1396 /* Catches the case when p == 100% */
1397 if ( ! percentiles[i].flag2 )
1398 percentiles[i].x1 = percentiles[i].x2 = f->v.f;
1401 printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
1402 i,percentiles[i].p,percentiles[i].x1,percentiles[i].x2);
1406 for (i = 0; i < n_percentiles; i++)
1408 struct freq_tab *ft = &get_var_freqs (v)->tab;
1412 if ( get_algorithm() != COMPATIBLE )
1414 s = modf((ft->valid_cases - 1) * percentiles[i].p , &dummy);
1418 s = modf((ft->valid_cases + 1) * percentiles[i].p -1, &dummy);
1421 percentiles[i].value = percentiles[i].x1 +
1422 ( percentiles[i].x2 - percentiles[i].x1) * s ;
1424 if ( percentiles[i].p == 0.50)
1425 median_value = &percentiles[i].value;
1429 /* Calculate the mode. */
1432 for (f = ft->valid; f < ft->missing; f++)
1434 if (most_often < f->c)
1439 else if (most_often == f->c)
1441 /* A duplicate mode is undefined.
1442 FIXME: keep track of *all* the modes. */
1447 /* Calculate moments. */
1448 m = moments_create (MOMENT_KURTOSIS);
1449 for (f = ft->valid; f < ft->missing; f++)
1450 moments_pass_one (m, f->v.f, f->c);
1451 for (f = ft->valid; f < ft->missing; f++)
1452 moments_pass_two (m, f->v.f, f->c);
1453 moments_calculate (m, NULL, &d[frq_mean], &d[frq_variance],
1454 &d[frq_skew], &d[frq_kurt]);
1455 moments_destroy (m);
1457 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1458 d[frq_min] = ft->valid[0].v.f;
1459 d[frq_max] = ft->valid[ft->n_valid - 1].v.f;
1460 d[frq_mode] = X_mode;
1461 d[frq_range] = d[frq_max] - d[frq_min];
1462 d[frq_median] = *median_value;
1463 d[frq_sum] = d[frq_mean] * W;
1464 d[frq_stddev] = sqrt (d[frq_variance]);
1465 d[frq_semean] = d[frq_stddev] / sqrt (W);
1466 d[frq_seskew] = calc_seskew (W);
1467 d[frq_sekurt] = calc_sekurt (W);
1470 /* Displays a table of all the statistics requested for variable V. */
1472 dump_statistics (struct variable *v, int show_varname)
1474 struct freq_tab *ft;
1475 double stat_value[frq_n_stats];
1476 struct tab_table *t;
1479 int n_explicit_percentiles = n_percentiles;
1481 if ( implicit_50th && n_percentiles > 0 )
1484 if (v->type == ALPHA)
1486 ft = &get_var_freqs (v)->tab;
1487 if (ft->n_valid == 0)
1489 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1493 calc_stats (v, stat_value);
1495 t = tab_create (3, n_stats + n_explicit_percentiles + 2, 0);
1496 tab_dim (t, tab_natural_dimensions);
1498 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1501 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1502 tab_vline (t, TAL_GAP , 1, 0, tab_nr(t) - 1 ) ;
1504 r=2; /* N missing and N valid are always dumped */
1506 for (i = 0; i < frq_n_stats; i++)
1507 if (stats & BIT_INDEX (i))
1509 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1510 gettext (st_name[i].s10));
1511 tab_float (t, 2, r, TAB_NONE, stat_value[i], 11, 3);
1515 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1516 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1517 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1519 tab_float(t, 2, 0, TAB_NONE, ft->valid_cases, 11, 0);
1520 tab_float(t, 2, 1, TAB_NONE, ft->total_cases - ft->valid_cases, 11, 0);
1523 for (i = 0; i < n_explicit_percentiles; i++, r++)
1527 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1530 tab_float (t, 1, r, TAB_LEFT, percentiles[i].p * 100, 3, 0 );
1531 tab_float (t, 2, r, TAB_NONE, percentiles[i].value, 11, 3);
1535 tab_columns (t, SOM_COL_DOWN, 1);
1539 tab_title (t, "%s: %s", v->name, v->label);
1541 tab_title (t, "%s", v->name);
1544 tab_flags (t, SOMF_NO_TITLE);
1551 /* Create a gsl_histogram from a freq_tab */
1553 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var)
1556 double x_min = DBL_MAX;
1557 double x_max = -DBL_MAX;
1559 gsl_histogram *hist;
1560 const double bins = 11;
1562 struct hsh_iterator hi;
1563 struct hsh_table *fh = ft->data;
1566 /* Find out the extremes of the x value */
1567 for ( frq = hsh_first(fh, &hi); frq != 0; frq = hsh_next(fh, &hi) )
1569 if ( mv_is_value_missing(&var->miss, &frq->v))
1572 if ( frq->v.f < x_min ) x_min = frq->v.f ;
1573 if ( frq->v.f > x_max ) x_max = frq->v.f ;
1576 hist = histogram_create(bins, x_min, x_max);
1578 for( i = 0 ; i < ft->n_valid ; ++i )
1580 frq = &ft->valid[i];
1581 gsl_histogram_accumulate(hist, frq->v.f, frq->c);
1588 static struct slice *
1589 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1590 const struct variable *var,
1594 /* Allocate an array of slices and fill them from the data in frq_tab
1595 n_slices will contain the number of slices allocated.
1596 The caller is responsible for freeing slices
1598 static struct slice *
1599 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1600 const struct variable *var,
1604 struct slice *slices;
1606 *n_slices = frq_tab->n_valid;
1608 slices = xnmalloc (*n_slices, sizeof *slices);
1610 for (i = 0 ; i < *n_slices ; ++i )
1612 const struct freq *frq = &frq_tab->valid[i];
1614 slices[i].label = value_to_string(&frq->v, var);
1616 slices[i].magnetude = frq->c;
1626 do_piechart(const struct variable *var, const struct freq_tab *frq_tab)
1628 struct slice *slices;
1631 slices = freq_tab_to_slice_array(frq_tab, var, &n_slices);
1633 piechart_plot(var_to_string(var), slices, n_slices);