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>
33 #include "bit-vector.h"
36 #include "dictionary.h"
50 #include "value-labels.h"
52 #include "procedure.h"
56 #include "plot-hist.h"
59 #define _(msgid) gettext (msgid)
60 #define N_(msgid) msgid
64 #include "debug-print.h"
69 format=cond:condense/onepage(*n:onepage_limit,"%s>=0")/!standard,
70 table:limit(n:limit,"%s>0")/notable/!table,
71 labels:!labels/nolabels,
72 sort:!avalue/dvalue/afreq/dfreq,
73 spaces:!single/double,
74 paging:newpage/!oldpage;
75 missing=miss:include/!exclude;
76 barchart(ba_)=:minimum(d:min),
78 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0");
79 piechart(pie_)=:minimum(d:min),
81 missing:missing/!nomissing;
82 histogram(hi_)=:minimum(d:min),
84 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
85 norm:!nonormal/normal,
86 incr:increment(d:inc,"%s>0");
87 hbar(hb_)=:minimum(d:min),
89 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
90 norm:!nonormal/normal,
91 incr:increment(d:inc,"%s>0");
94 +percentiles = double list;
95 statistics[st_]=1|mean,2|semean,3|median,4|mode,5|stddev,6|variance,
96 7|kurtosis,8|skewness,9|range,10|minimum,11|maximum,12|sum,
97 13|default,14|seskewness,15|sekurtosis,all,none.
105 frq_mean = 0, frq_semean, frq_median, frq_mode, frq_stddev, frq_variance,
106 frq_kurt, frq_sekurt, frq_skew, frq_seskew, frq_range, frq_min, frq_max,
110 /* Description of a statistic. */
113 int st_indx; /* Index into a_statistics[]. */
114 const char *s10; /* Identifying string. */
117 /* Table of statistics, indexed by dsc_*. */
118 static struct frq_info st_name[frq_n_stats + 1] =
120 {FRQ_ST_MEAN, N_("Mean")},
121 {FRQ_ST_SEMEAN, N_("S.E. Mean")},
122 {FRQ_ST_MEDIAN, N_("Median")},
123 {FRQ_ST_MODE, N_("Mode")},
124 {FRQ_ST_STDDEV, N_("Std Dev")},
125 {FRQ_ST_VARIANCE, N_("Variance")},
126 {FRQ_ST_KURTOSIS, N_("Kurtosis")},
127 {FRQ_ST_SEKURTOSIS, N_("S.E. Kurt")},
128 {FRQ_ST_SKEWNESS, N_("Skewness")},
129 {FRQ_ST_SESKEWNESS, N_("S.E. Skew")},
130 {FRQ_ST_RANGE, N_("Range")},
131 {FRQ_ST_MINIMUM, N_("Minimum")},
132 {FRQ_ST_MAXIMUM, N_("Maximum")},
133 {FRQ_ST_SUM, N_("Sum")},
137 /* Percentiles to calculate. */
141 double p; /* the %ile to be calculated */
142 double value; /* the %ile's value */
143 double x1; /* The datum value <= the percentile */
144 double x2; /* The datum value >= the percentile */
146 int flag2; /* Set to 1 if this percentile value has been found */
150 static void add_percentile (double x) ;
152 static struct percentile *percentiles;
153 static int n_percentiles;
155 static int implicit_50th ;
157 /* Groups of statistics. */
159 #define frq_default \
160 (BI (frq_mean) | BI (frq_stddev) | BI (frq_min) | BI (frq_max))
162 (BI (frq_sum) | BI(frq_min) | BI(frq_max) \
163 | BI(frq_mean) | BI(frq_semean) | BI(frq_stddev) \
164 | BI(frq_variance) | BI(frq_kurt) | BI(frq_sekurt) \
165 | BI(frq_skew) | BI(frq_seskew) | BI(frq_range) \
166 | BI(frq_range) | BI(frq_mode) | BI(frq_median))
168 /* Statistics; number of statistics. */
169 static unsigned long stats;
172 /* Types of graphs. */
175 GFT_NONE, /* Don't draw graphs. */
176 GFT_BAR, /* Draw bar charts. */
177 GFT_HIST, /* Draw histograms. */
178 GFT_PIE, /* Draw piechart */
179 GFT_HBAR /* Draw bar charts or histograms at our discretion. */
182 /* Parsed command. */
183 static struct cmd_frequencies cmd;
185 /* Summary of the barchart, histogram, and hbar subcommands. */
186 /* FIXME: These should not be mututally exclusive */
187 static int chart; /* NONE/BAR/HIST/HBAR/PIE. */
188 static double min, max; /* Minimum, maximum on y axis. */
189 static int format; /* FREQ/PERCENT: Scaling of y axis. */
190 static double scale, incr; /* FIXME */
191 static int normal; /* FIXME */
193 /* Variables for which to calculate statistics. */
194 static size_t n_variables;
195 static struct variable **v_variables;
197 /* Arenas used to store semi-permanent storage. */
198 static struct pool *int_pool; /* Integer mode. */
199 static struct pool *gen_pool; /* General mode. */
201 /* Frequency tables. */
203 /* Frequency table entry. */
206 union value v; /* The value. */
207 double c; /* The number of occurrences of the value. */
210 /* Types of frequency tables. */
217 /* Entire frequency table. */
220 int mode; /* FRQM_GENERAL or FRQM_INTEGER. */
223 struct hsh_table *data; /* Undifferentiated data. */
226 double *vector; /* Frequencies proper. */
227 int min, max; /* The boundaries of the table. */
228 double out_of_range; /* Sum of weights of out-of-range values. */
229 double sysmis; /* Sum of weights of SYSMIS values. */
232 struct freq *valid; /* Valid freqs. */
233 int n_valid; /* Number of total freqs. */
235 struct freq *missing; /* Missing freqs. */
236 int n_missing; /* Number of missing freqs. */
239 double total_cases; /* Sum of weights of all cases. */
240 double valid_cases; /* Sum of weights of valid cases. */
244 /* Per-variable frequency data. */
247 /* Freqency table. */
248 struct freq_tab tab; /* Frequencies table to use. */
251 int n_groups; /* Number of groups. */
252 double *groups; /* Groups. */
255 double stat[frq_n_stats];
258 static inline struct var_freqs *
259 get_var_freqs (struct variable *v)
262 assert (v->aux != NULL);
266 static void determine_charts (void);
268 static void calc_stats (struct variable *v, double d[frq_n_stats]);
270 static void precalc (void *);
271 static bool calc (struct ccase *, void *);
272 static void postcalc (void *);
274 static void postprocess_freq_tab (struct variable *);
275 static void dump_full (struct variable *);
276 static void dump_condensed (struct variable *);
277 static void dump_statistics (struct variable *, int show_varname);
278 static void cleanup_freq_tab (struct variable *);
280 static hsh_hash_func hash_value_numeric, hash_value_alpha;
281 static hsh_compare_func compare_value_numeric_a, compare_value_alpha_a;
282 static hsh_compare_func compare_value_numeric_d, compare_value_alpha_d;
283 static hsh_compare_func compare_freq_numeric_a, compare_freq_alpha_a;
284 static hsh_compare_func compare_freq_numeric_d, compare_freq_alpha_d;
287 static void do_piechart(const struct variable *var,
288 const struct freq_tab *frq_tab);
291 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var);
295 /* Parser and outline. */
297 static int internal_cmd_frequencies (void);
300 cmd_frequencies (void)
304 int_pool = pool_create ();
305 result = internal_cmd_frequencies ();
306 pool_destroy (int_pool);
308 pool_destroy (gen_pool);
316 internal_cmd_frequencies (void)
327 if (!parse_frequencies (&cmd))
330 if (cmd.onepage_limit == NOT_LONG)
331 cmd.onepage_limit = 50;
333 /* Figure out statistics to calculate. */
335 if (cmd.a_statistics[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
336 stats |= frq_default;
337 if (cmd.a_statistics[FRQ_ST_ALL])
339 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
340 stats &= ~frq_median;
341 for (i = 0; i < frq_n_stats; i++)
342 if (cmd.a_statistics[st_name[i].st_indx])
343 stats |= BIT_INDEX (i);
344 if (stats & frq_kurt)
346 if (stats & frq_skew)
349 /* Calculate n_stats. */
351 for (i = 0; i < frq_n_stats; i++)
352 if ((stats & BIT_INDEX (i)))
357 if (chart != GFT_NONE || cmd.sbc_ntiles)
358 cmd.sort = FRQ_AVALUE;
360 /* Work out what percentiles need to be calculated */
361 if ( cmd.sbc_percentiles )
363 for ( i = 0 ; i < MAXLISTS ; ++i )
366 subc_list_double *ptl_list = &cmd.dl_percentiles[i];
367 for ( pl = 0 ; pl < subc_list_double_count(ptl_list); ++pl)
368 add_percentile(subc_list_double_at(ptl_list,pl) / 100.0 );
371 if ( cmd.sbc_ntiles )
373 for ( i = 0 ; i < cmd.sbc_ntiles ; ++i )
376 for (j = 0; j <= cmd.n_ntiles[i]; ++j )
377 add_percentile(j / (double) cmd.n_ntiles[i]);
383 ok = procedure_with_splits (precalc, calc, postcalc, NULL);
385 free_frequencies(&cmd);
387 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
390 /* Figure out which charts the user requested. */
392 determine_charts (void)
394 int count = (!!cmd.sbc_histogram) + (!!cmd.sbc_barchart) +
395 (!!cmd.sbc_hbar) + (!!cmd.sbc_piechart);
405 msg (SW, _("At most one of BARCHART, HISTOGRAM, or HBAR should be "
406 "given. HBAR will be assumed. Argument values will be "
407 "given precedence increasing along the order given."));
409 else if (cmd.sbc_histogram)
411 else if (cmd.sbc_barchart)
413 else if (cmd.sbc_piechart)
424 if (cmd.sbc_barchart)
426 if (cmd.ba_min != SYSMIS)
428 if (cmd.ba_max != SYSMIS)
430 if (cmd.ba_scale == FRQ_FREQ)
435 else if (cmd.ba_scale == FRQ_PERCENT)
437 format = FRQ_PERCENT;
442 if (cmd.sbc_histogram)
444 if (cmd.hi_min != SYSMIS)
446 if (cmd.hi_max != SYSMIS)
448 if (cmd.hi_scale == FRQ_FREQ)
453 else if (cmd.hi_scale == FRQ_PERCENT)
455 format = FRQ_PERCENT;
458 if (cmd.hi_norm != FRQ_NONORMAL )
460 if (cmd.hi_incr == FRQ_INCREMENT)
466 if (cmd.hb_min != SYSMIS)
468 if (cmd.hb_max != SYSMIS)
470 if (cmd.hb_scale == FRQ_FREQ)
475 else if (cmd.hb_scale == FRQ_PERCENT)
477 format = FRQ_PERCENT;
482 if (cmd.hb_incr == FRQ_INCREMENT)
486 if (min != SYSMIS && max != SYSMIS && min >= max)
488 msg (SE, _("MAX must be greater than or equal to MIN, if both are "
489 "specified. However, MIN was specified as %g and MAX as %g. "
490 "MIN and MAX will be ignored."), min, max);
495 /* Add data from case C to the frequency table. */
497 calc (struct ccase *c, void *aux UNUSED)
503 weight = dict_get_case_weight (default_dict, c, &bad_warn);
505 for (i = 0; i < n_variables; i++)
507 struct variable *v = v_variables[i];
508 const union value *val = case_data (c, v->fv);
509 struct freq_tab *ft = &get_var_freqs (v)->tab;
517 struct freq **fpp = (struct freq **) hsh_probe (ft->data, val);
523 struct freq *fp = *fpp = pool_alloc (gen_pool, sizeof *fp);
531 if (val->f == SYSMIS)
532 ft->sysmis += weight;
533 else if (val->f > INT_MIN+1 && val->f < INT_MAX-1)
536 if (i >= ft->min && i <= ft->max)
537 ft->vector[i - ft->min] += weight;
540 ft->out_of_range += weight;
549 /* Prepares each variable that is the target of FREQUENCIES by setting
550 up its hash table. */
552 precalc (void *aux UNUSED)
556 pool_destroy (gen_pool);
557 gen_pool = pool_create ();
559 for (i = 0; i < n_variables; i++)
561 struct variable *v = v_variables[i];
562 struct freq_tab *ft = &get_var_freqs (v)->tab;
564 if (ft->mode == FRQM_GENERAL)
567 hsh_compare_func *compare;
569 if (v->type == NUMERIC)
571 hash = hash_value_numeric;
572 compare = compare_value_numeric_a;
576 hash = hash_value_alpha;
577 compare = compare_value_alpha_a;
579 ft->data = hsh_create (16, compare, hash, NULL, v);
585 for (j = (ft->max - ft->min); j >= 0; j--)
587 ft->out_of_range = 0.0;
593 /* Finishes up with the variables after frequencies have been
594 calculated. Displays statistics, percentiles, ... */
596 postcalc (void *aux UNUSED)
600 for (i = 0; i < n_variables; i++)
602 struct variable *v = v_variables[i];
603 struct var_freqs *vf = get_var_freqs (v);
604 struct freq_tab *ft = &vf->tab;
606 int dumped_freq_tab = 1;
608 postprocess_freq_tab (v);
610 /* Frequencies tables. */
611 n_categories = ft->n_valid + ft->n_missing;
612 if (cmd.table == FRQ_TABLE
613 || (cmd.table == FRQ_LIMIT && n_categories <= cmd.limit))
623 if (n_categories > cmd.onepage_limit)
636 dump_statistics (v, !dumped_freq_tab);
640 if ( chart == GFT_HIST)
642 double d[frq_n_stats];
643 struct normal_curve norm;
644 gsl_histogram *hist ;
647 norm.N = vf->tab.valid_cases;
650 norm.mean = d[frq_mean];
651 norm.stddev = d[frq_stddev];
653 hist = freq_tab_to_hist(ft,v);
655 histogram_plot(hist, var_to_string(v), &norm, normal);
657 gsl_histogram_free(hist);
661 if ( chart == GFT_PIE)
663 do_piechart(v_variables[i], ft);
668 cleanup_freq_tab (v);
673 /* Returns the comparison function that should be used for
674 sorting a frequency table by FRQ_SORT using VAR_TYPE
676 static hsh_compare_func *
677 get_freq_comparator (int frq_sort, int var_type)
679 /* Note that q2c generates tags beginning with 1000. */
680 switch (frq_sort | (var_type << 16))
682 case FRQ_AVALUE | (NUMERIC << 16): return compare_value_numeric_a;
683 case FRQ_AVALUE | (ALPHA << 16): return compare_value_alpha_a;
684 case FRQ_DVALUE | (NUMERIC << 16): return compare_value_numeric_d;
685 case FRQ_DVALUE | (ALPHA << 16): return compare_value_alpha_d;
686 case FRQ_AFREQ | (NUMERIC << 16): return compare_freq_numeric_a;
687 case FRQ_AFREQ | (ALPHA << 16): return compare_freq_alpha_a;
688 case FRQ_DFREQ | (NUMERIC << 16): return compare_freq_numeric_d;
689 case FRQ_DFREQ | (ALPHA << 16): return compare_freq_alpha_d;
696 /* Returns nonzero iff the value in struct freq F is non-missing
699 not_missing (const void *f_, void *v_)
701 const struct freq *f = f_;
702 struct variable *v = v_;
704 return !mv_is_value_missing (&v->miss, &f->v);
707 /* Summarizes the frequency table data for variable V. */
709 postprocess_freq_tab (struct variable *v)
711 hsh_compare_func *compare;
715 struct freq *freqs, *f;
718 ft = &get_var_freqs (v)->tab;
719 assert (ft->mode == FRQM_GENERAL);
720 compare = get_freq_comparator (cmd.sort, v->type);
722 /* Extract data from hash table. */
723 count = hsh_count (ft->data);
724 data = hsh_data (ft->data);
726 /* Copy dereferenced data into freqs. */
727 freqs = xnmalloc (count, sizeof *freqs);
728 for (i = 0; i < count; i++)
730 struct freq *f = data[i];
734 /* Put data into ft. */
736 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, v);
737 ft->missing = freqs + ft->n_valid;
738 ft->n_missing = count - ft->n_valid;
741 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare, v);
742 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, v);
744 /* Summary statistics. */
745 ft->valid_cases = 0.0;
746 for(i = 0 ; i < ft->n_valid ; ++i )
749 ft->valid_cases += f->c;
753 ft->total_cases = ft->valid_cases ;
754 for(i = 0 ; i < ft->n_missing ; ++i )
757 ft->total_cases += f->c;
762 /* Frees the frequency table for variable V. */
764 cleanup_freq_tab (struct variable *v)
766 struct freq_tab *ft = &get_var_freqs (v)->tab;
767 assert (ft->mode == FRQM_GENERAL);
769 hsh_destroy (ft->data);
772 /* Parses the VARIABLES subcommand, adding to
773 {n_variables,v_variables}. */
775 frq_custom_variables (struct cmd_frequencies *cmd UNUSED)
778 int min = 0, max = 0;
780 size_t old_n_variables = n_variables;
784 if (token != T_ALL && (token != T_ID
785 || dict_lookup_var (default_dict, tokid) == NULL))
788 if (!parse_variables (default_dict, &v_variables, &n_variables,
789 PV_APPEND | PV_NO_SCRATCH))
792 if (!lex_match ('('))
797 if (!lex_force_int ())
799 min = lex_integer ();
801 if (!lex_force_match (','))
803 if (!lex_force_int ())
805 max = lex_integer ();
807 if (!lex_force_match (')'))
811 msg (SE, _("Upper limit of integer mode value range must be "
812 "greater than lower limit."));
817 for (i = old_n_variables; i < n_variables; i++)
819 struct variable *v = v_variables[i];
820 struct var_freqs *vf;
824 msg (SE, _("Variable %s specified multiple times on VARIABLES "
825 "subcommand."), v->name);
828 if (mode == FRQM_INTEGER && v->type != NUMERIC)
830 msg (SE, _("Integer mode specified, but %s is not a numeric "
831 "variable."), v->name);
835 vf = var_attach_aux (v, xmalloc (sizeof *vf), var_dtor_free);
837 vf->tab.valid = vf->tab.missing = NULL;
838 if (mode == FRQM_INTEGER)
842 vf->tab.vector = pool_nalloc (int_pool,
843 max - min + 1, sizeof *vf->tab.vector);
846 vf->tab.vector = NULL;
853 /* Parses the GROUPED subcommand, setting the n_grouped, grouped
854 fields of specified variables. */
856 frq_custom_grouped (struct cmd_frequencies *cmd UNUSED)
859 if ((token == T_ID && dict_lookup_var (default_dict, tokid) != NULL)
865 /* Max, current size of list; list itself. */
873 if (!parse_variables (default_dict, &v, &n,
874 PV_NO_DUPLICATE | PV_NUMERIC))
880 while (lex_integer ())
885 dl = pool_nrealloc (int_pool, dl, ml, sizeof *dl);
891 /* Note that nl might still be 0 and dl might still be
892 NULL. That's okay. */
893 if (!lex_match (')'))
896 msg (SE, _("`)' expected after GROUPED interval list."));
906 for (i = 0; i < n; i++)
907 if (v[i]->aux == NULL)
908 msg (SE, _("Variables %s specified on GROUPED but not on "
909 "VARIABLES."), v[i]->name);
912 struct var_freqs *vf = get_var_freqs (v[i]);
914 if (vf->groups != NULL)
915 msg (SE, _("Variables %s specified multiple times on GROUPED "
916 "subcommand."), v[i]->name);
924 if (!lex_match ('/'))
926 if ((token != T_ID || dict_lookup_var (default_dict, tokid) != NULL)
937 /* Adds X to the list of percentiles, keeping the list in proper
940 add_percentile (double x)
944 for (i = 0; i < n_percentiles; i++)
946 /* Do nothing if it's already in the list */
947 if ( fabs(x - percentiles[i].p) < DBL_EPSILON )
950 if (x < percentiles[i].p)
954 if (i >= n_percentiles || tokval != percentiles[i].p)
956 percentiles = pool_nrealloc (int_pool, percentiles,
957 n_percentiles + 1, sizeof *percentiles);
959 if (i < n_percentiles)
960 memmove (&percentiles[i + 1], &percentiles[i],
961 (n_percentiles - i) * sizeof (struct percentile) );
963 percentiles[i].p = x;
968 /* Comparison functions. */
970 /* Hash of numeric values. */
972 hash_value_numeric (const void *value_, void *foo UNUSED)
974 const struct freq *value = value_;
975 return hsh_hash_double (value->v.f);
978 /* Hash of string values. */
980 hash_value_alpha (const void *value_, void *v_)
982 const struct freq *value = value_;
983 struct variable *v = v_;
985 return hsh_hash_bytes (value->v.s, v->width);
988 /* Ascending numeric compare of values. */
990 compare_value_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
992 const struct freq *a = a_;
993 const struct freq *b = b_;
997 else if (a->v.f < b->v.f)
1003 /* Ascending string compare of values. */
1005 compare_value_alpha_a (const void *a_, const void *b_, void *v_)
1007 const struct freq *a = a_;
1008 const struct freq *b = b_;
1009 const struct variable *v = v_;
1011 return memcmp (a->v.s, b->v.s, v->width);
1014 /* Descending numeric compare of values. */
1016 compare_value_numeric_d (const void *a, const void *b, void *foo UNUSED)
1018 return -compare_value_numeric_a (a, b, foo);
1021 /* Descending string compare of values. */
1023 compare_value_alpha_d (const void *a, const void *b, void *v)
1025 return -compare_value_alpha_a (a, b, v);
1028 /* Ascending numeric compare of frequency;
1029 secondary key on ascending numeric value. */
1031 compare_freq_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
1033 const struct freq *a = a_;
1034 const struct freq *b = b_;
1038 else if (a->c < b->c)
1041 if (a->v.f > b->v.f)
1043 else if (a->v.f < b->v.f)
1049 /* Ascending numeric compare of frequency;
1050 secondary key on ascending string value. */
1052 compare_freq_alpha_a (const void *a_, const void *b_, void *v_)
1054 const struct freq *a = a_;
1055 const struct freq *b = b_;
1056 const struct variable *v = v_;
1060 else if (a->c < b->c)
1063 return memcmp (a->v.s, b->v.s, v->width);
1066 /* Descending numeric compare of frequency;
1067 secondary key on ascending numeric value. */
1069 compare_freq_numeric_d (const void *a_, const void *b_, void *foo UNUSED)
1071 const struct freq *a = a_;
1072 const struct freq *b = b_;
1076 else if (a->c < b->c)
1079 if (a->v.f > b->v.f)
1081 else if (a->v.f < b->v.f)
1087 /* Descending numeric compare of frequency;
1088 secondary key on ascending string value. */
1090 compare_freq_alpha_d (const void *a_, const void *b_, void *v_)
1092 const struct freq *a = a_;
1093 const struct freq *b = b_;
1094 const struct variable *v = v_;
1098 else if (a->c < b->c)
1101 return memcmp (a->v.s, b->v.s, v->width);
1104 /* Frequency table display. */
1106 /* Sets the widths of all the columns and heights of all the rows in
1107 table T for driver D. */
1109 full_dim (struct tab_table *t, struct outp_driver *d)
1111 int lab = cmd.labels == FRQ_LABELS;
1115 t->w[0] = min (tab_natural_width (t, d, 0), d->prop_em_width * 15);
1116 for (i = lab; i < lab + 5; i++)
1117 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
1118 for (i = 0; i < t->nr; i++)
1119 t->h[i] = d->font_height;
1122 /* Displays a full frequency table for variable V. */
1124 dump_full (struct variable *v)
1127 struct freq_tab *ft;
1129 struct tab_table *t;
1131 double cum_total = 0.0;
1132 double cum_freq = 0.0;
1142 static struct init vec[] =
1144 {4, 0, N_("Valid")},
1146 {1, 1, N_("Value")},
1147 {2, 1, N_("Frequency")},
1148 {3, 1, N_("Percent")},
1149 {4, 1, N_("Percent")},
1150 {5, 1, N_("Percent")},
1158 int lab = cmd.labels == FRQ_LABELS;
1160 ft = &get_var_freqs (v)->tab;
1161 n_categories = ft->n_valid + ft->n_missing;
1162 t = tab_create (5 + lab, n_categories + 3, 0);
1163 tab_headers (t, 0, 0, 2, 0);
1164 tab_dim (t, full_dim);
1167 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value Label"));
1168 for (p = vec; p->s; p++)
1169 tab_text (t, p->c - (p->r ? !lab : 0), p->r,
1170 TAB_CENTER | TAT_TITLE, gettext (p->s));
1173 for (f = ft->valid; f < ft->missing; f++)
1175 double percent, valid_percent;
1179 percent = f->c / ft->total_cases * 100.0;
1180 valid_percent = f->c / ft->valid_cases * 100.0;
1181 cum_total += valid_percent;
1185 const char *label = val_labs_find (v->val_labs, f->v);
1187 tab_text (t, 0, r, TAB_LEFT, label);
1190 tab_value (t, 0 + lab, r, TAB_NONE, &f->v, &v->print);
1191 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1192 tab_float (t, 2 + lab, r, TAB_NONE, percent, 5, 1);
1193 tab_float (t, 3 + lab, r, TAB_NONE, valid_percent, 5, 1);
1194 tab_float (t, 4 + lab, r, TAB_NONE, cum_total, 5, 1);
1197 for (; f < &ft->valid[n_categories]; f++)
1203 const char *label = val_labs_find (v->val_labs, f->v);
1205 tab_text (t, 0, r, TAB_LEFT, label);
1208 tab_value (t, 0 + lab, r, TAB_NONE, &f->v, &v->print);
1209 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1210 tab_float (t, 2 + lab, r, TAB_NONE,
1211 f->c / ft->total_cases * 100.0, 5, 1);
1212 tab_text (t, 3 + lab, r, TAB_NONE, _("Missing"));
1216 tab_box (t, TAL_1, TAL_1,
1217 cmd.spaces == FRQ_SINGLE ? -1 : (TAL_1 | TAL_SPACING), TAL_1,
1219 tab_hline (t, TAL_2, 0, 4 + lab, 2);
1220 tab_hline (t, TAL_2, 0, 4 + lab, r);
1221 tab_joint_text (t, 0, r, 0 + lab, r, TAB_RIGHT | TAT_TITLE, _("Total"));
1222 tab_vline (t, TAL_0, 1, r, r);
1223 tab_float (t, 1 + lab, r, TAB_NONE, cum_freq, 8, 0);
1224 tab_float (t, 2 + lab, r, TAB_NONE, 100.0, 5, 1);
1225 tab_float (t, 3 + lab, r, TAB_NONE, 100.0, 5, 1);
1227 tab_title (t, 1, "%s: %s", v->name, v->label ? v->label : "");
1232 /* Sets the widths of all the columns and heights of all the rows in
1233 table T for driver D. */
1235 condensed_dim (struct tab_table *t, struct outp_driver *d)
1237 int cum_w = max (outp_string_width (d, _("Cum")),
1238 max (outp_string_width (d, _("Cum")),
1239 outp_string_width (d, "000")));
1243 for (i = 0; i < 2; i++)
1244 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
1245 for (i = 2; i < 4; i++)
1247 for (i = 0; i < t->nr; i++)
1248 t->h[i] = d->font_height;
1251 /* Display condensed frequency table for variable V. */
1253 dump_condensed (struct variable *v)
1256 struct freq_tab *ft;
1258 struct tab_table *t;
1260 double cum_total = 0.0;
1262 ft = &get_var_freqs (v)->tab;
1263 n_categories = ft->n_valid + ft->n_missing;
1264 t = tab_create (4, n_categories + 2, 0);
1266 tab_headers (t, 0, 0, 2, 0);
1267 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value"));
1268 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Freq"));
1269 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1270 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Cum"));
1271 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1272 tab_dim (t, condensed_dim);
1275 for (f = ft->valid; f < ft->missing; f++)
1279 percent = f->c / ft->total_cases * 100.0;
1280 cum_total += f->c / ft->valid_cases * 100.0;
1282 tab_value (t, 0, r, TAB_NONE, &f->v, &v->print);
1283 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1284 tab_float (t, 2, r, TAB_NONE, percent, 3, 0);
1285 tab_float (t, 3, r, TAB_NONE, cum_total, 3, 0);
1288 for (; f < &ft->valid[n_categories]; f++)
1290 tab_value (t, 0, r, TAB_NONE, &f->v, &v->print);
1291 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1292 tab_float (t, 2, r, TAB_NONE,
1293 f->c / ft->total_cases * 100.0, 3, 0);
1297 tab_box (t, TAL_1, TAL_1,
1298 cmd.spaces == FRQ_SINGLE ? -1 : (TAL_1 | TAL_SPACING), TAL_1,
1300 tab_hline (t, TAL_2, 0, 3, 2);
1301 tab_title (t, 1, "%s: %s", v->name, v->label ? v->label : "");
1302 tab_columns (t, SOM_COL_DOWN, 1);
1306 /* Statistical display. */
1308 /* Calculates all the pertinent statistics for variable V, putting
1309 them in array D[]. FIXME: This could be made much more optimal. */
1311 calc_stats (struct variable *v, double d[frq_n_stats])
1313 struct freq_tab *ft = &get_var_freqs (v)->tab;
1314 double W = ft->valid_cases;
1323 double *median_value;
1325 /* Calculate percentiles. */
1327 /* If the 50th percentile was not explicitly requested then we must
1328 calculate it anyway --- it's the median */
1330 for (i = 0; i < n_percentiles; i++)
1332 if (percentiles[i].p == 0.5)
1334 median_value = &percentiles[i].value;
1339 if ( 0 == median_value )
1341 add_percentile (0.5);
1345 for (i = 0; i < n_percentiles; i++)
1347 percentiles[i].flag = 0;
1348 percentiles[i].flag2 = 0;
1352 for (idx = 0; idx < ft->n_valid; ++idx)
1354 static double prev_value = SYSMIS;
1355 f = &ft->valid[idx];
1357 for (i = 0; i < n_percentiles; i++)
1360 if ( percentiles[i].flag2 ) continue ;
1362 if ( get_algorithm() != COMPATIBLE )
1364 (ft->valid_cases - 1) * percentiles[i].p;
1367 (ft->valid_cases + 1) * percentiles[i].p - 1;
1369 if ( percentiles[i].flag )
1371 percentiles[i].x2 = f->v.f;
1372 percentiles[i].x1 = prev_value;
1373 percentiles[i].flag2 = 1;
1379 if ( f->c > 1 && rank - (f->c - 1) > tp )
1381 percentiles[i].x2 = percentiles[i].x1 = f->v.f;
1382 percentiles[i].flag2 = 1;
1386 percentiles[i].flag=1;
1392 prev_value = f->v.f;
1395 for (i = 0; i < n_percentiles; i++)
1397 /* Catches the case when p == 100% */
1398 if ( ! percentiles[i].flag2 )
1399 percentiles[i].x1 = percentiles[i].x2 = f->v.f;
1402 printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
1403 i,percentiles[i].p,percentiles[i].x1,percentiles[i].x2);
1407 for (i = 0; i < n_percentiles; i++)
1409 struct freq_tab *ft = &get_var_freqs (v)->tab;
1413 if ( get_algorithm() != COMPATIBLE )
1415 s = modf((ft->valid_cases - 1) * percentiles[i].p , &dummy);
1419 s = modf((ft->valid_cases + 1) * percentiles[i].p -1, &dummy);
1422 percentiles[i].value = percentiles[i].x1 +
1423 ( percentiles[i].x2 - percentiles[i].x1) * s ;
1425 if ( percentiles[i].p == 0.50)
1426 median_value = &percentiles[i].value;
1430 /* Calculate the mode. */
1433 for (f = ft->valid; f < ft->missing; f++)
1435 if (most_often < f->c)
1440 else if (most_often == f->c)
1442 /* A duplicate mode is undefined.
1443 FIXME: keep track of *all* the modes. */
1448 /* Calculate moments. */
1449 m = moments_create (MOMENT_KURTOSIS);
1450 for (f = ft->valid; f < ft->missing; f++)
1451 moments_pass_one (m, f->v.f, f->c);
1452 for (f = ft->valid; f < ft->missing; f++)
1453 moments_pass_two (m, f->v.f, f->c);
1454 moments_calculate (m, NULL, &d[frq_mean], &d[frq_variance],
1455 &d[frq_skew], &d[frq_kurt]);
1456 moments_destroy (m);
1458 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1459 d[frq_min] = ft->valid[0].v.f;
1460 d[frq_max] = ft->valid[ft->n_valid - 1].v.f;
1461 d[frq_mode] = X_mode;
1462 d[frq_range] = d[frq_max] - d[frq_min];
1463 d[frq_median] = *median_value;
1464 d[frq_sum] = d[frq_mean] * W;
1465 d[frq_stddev] = sqrt (d[frq_variance]);
1466 d[frq_semean] = d[frq_stddev] / sqrt (W);
1467 d[frq_seskew] = calc_seskew (W);
1468 d[frq_sekurt] = calc_sekurt (W);
1471 /* Displays a table of all the statistics requested for variable V. */
1473 dump_statistics (struct variable *v, int show_varname)
1475 struct freq_tab *ft;
1476 double stat_value[frq_n_stats];
1477 struct tab_table *t;
1480 int n_explicit_percentiles = n_percentiles;
1482 if ( implicit_50th && n_percentiles > 0 )
1485 if (v->type == ALPHA)
1487 ft = &get_var_freqs (v)->tab;
1488 if (ft->n_valid == 0)
1490 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1494 calc_stats (v, stat_value);
1496 t = tab_create (3, n_stats + n_explicit_percentiles + 2, 0);
1497 tab_dim (t, tab_natural_dimensions);
1499 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1502 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1503 tab_vline (t, TAL_1 | TAL_SPACING , 1, 0, tab_nr(t) - 1 ) ;
1505 r=2; /* N missing and N valid are always dumped */
1507 for (i = 0; i < frq_n_stats; i++)
1508 if (stats & BIT_INDEX (i))
1510 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1511 gettext (st_name[i].s10));
1512 tab_float (t, 2, r, TAB_NONE, stat_value[i], 11, 3);
1516 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1517 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1518 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1520 tab_float(t, 2, 0, TAB_NONE, ft->valid_cases, 11, 0);
1521 tab_float(t, 2, 1, TAB_NONE, ft->total_cases - ft->valid_cases, 11, 0);
1524 for (i = 0; i < n_explicit_percentiles; i++, r++)
1528 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1531 tab_float (t, 1, r, TAB_LEFT, percentiles[i].p * 100, 3, 0 );
1532 tab_float (t, 2, r, TAB_NONE, percentiles[i].value, 11, 3);
1536 tab_columns (t, SOM_COL_DOWN, 1);
1540 tab_title (t, 1, "%s: %s", v->name, v->label);
1542 tab_title (t, 0, v->name);
1545 tab_flags (t, SOMF_NO_TITLE);
1552 /* Create a gsl_histogram from a freq_tab */
1554 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var)
1557 double x_min = DBL_MAX;
1558 double x_max = -DBL_MAX;
1560 gsl_histogram *hist;
1561 const double bins = 11;
1563 struct hsh_iterator hi;
1564 struct hsh_table *fh = ft->data;
1567 /* Find out the extremes of the x value */
1568 for ( frq = hsh_first(fh, &hi); frq != 0; frq = hsh_next(fh, &hi) )
1570 if ( mv_is_value_missing(&var->miss, &frq->v))
1573 if ( frq->v.f < x_min ) x_min = frq->v.f ;
1574 if ( frq->v.f > x_max ) x_max = frq->v.f ;
1577 hist = histogram_create(bins, x_min, x_max);
1579 for( i = 0 ; i < ft->n_valid ; ++i )
1581 frq = &ft->valid[i];
1582 gsl_histogram_accumulate(hist, frq->v.f, frq->c);
1589 static struct slice *
1590 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1591 const struct variable *var,
1595 /* Allocate an array of slices and fill them from the data in frq_tab
1596 n_slices will contain the number of slices allocated.
1597 The caller is responsible for freeing slices
1599 static struct slice *
1600 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1601 const struct variable *var,
1605 struct slice *slices;
1607 *n_slices = frq_tab->n_valid;
1609 slices = xnmalloc (*n_slices, sizeof *slices);
1611 for (i = 0 ; i < *n_slices ; ++i )
1613 const struct freq *frq = &frq_tab->valid[i];
1615 slices[i].label = value_to_string(&frq->v, var);
1617 slices[i].magnetude = frq->c;
1627 do_piechart(const struct variable *var, const struct freq_tab *frq_tab)
1629 struct slice *slices;
1632 slices = freq_tab_to_slice_array(frq_tab, var, &n_slices);
1634 piechart_plot(var_to_string(var), slices, n_slices);