1 /* PSPP - computes sample statistics.
2 Copyright (C) 1997-9, 2000 Free Software Foundation, Inc.
4 This program is free software; you can redistribute it and/or
5 modify it under the terms of the GNU General Public License as
6 published by the Free Software Foundation; either version 2 of the
7 License, or (at your option) any later version.
9 This program is distributed in the hope that it will be useful, but
10 WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
22 * Remember that histograms, bar charts need mean, stddev.
29 #include <gsl/gsl_histogram.h>
31 #include <data/case.h>
32 #include <data/casegrouper.h>
33 #include <data/casereader.h>
34 #include <data/dictionary.h>
35 #include <data/format.h>
36 #include <data/procedure.h>
37 #include <data/settings.h>
38 #include <data/value-labels.h>
39 #include <data/variable.h>
40 #include <language/command.h>
41 #include <language/dictionary/split-file.h>
42 #include <language/lexer/lexer.h>
43 #include <libpspp/alloc.h>
44 #include <libpspp/array.h>
45 #include <libpspp/bit-vector.h>
46 #include <libpspp/compiler.h>
47 #include <libpspp/hash.h>
48 #include <libpspp/magic.h>
49 #include <libpspp/message.h>
50 #include <libpspp/misc.h>
51 #include <libpspp/pool.h>
52 #include <libpspp/str.h>
53 #include <math/histogram.h>
54 #include <math/moments.h>
55 #include <output/chart.h>
56 #include <output/charts/piechart.h>
57 #include <output/charts/plot-hist.h>
58 #include <output/manager.h>
59 #include <output/output.h>
60 #include <output/table.h>
67 #define _(msgid) gettext (msgid)
68 #define N_(msgid) msgid
75 +format=cond:condense/onepage(*n:onepage_limit,"%s>=0")/!standard,
76 table:limit(n:limit,"%s>0")/notable/!table,
77 labels:!labels/nolabels,
78 sort:!avalue/dvalue/afreq/dfreq,
79 spaces:!single/double,
80 paging:newpage/!oldpage;
81 missing=miss:include/!exclude;
82 barchart(ba_)=:minimum(d:min),
84 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0");
85 piechart(pie_)=:minimum(d:min),
87 missing:missing/!nomissing;
88 histogram(hi_)=:minimum(d:min),
90 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
91 norm:!nonormal/normal,
92 incr:increment(d:inc,"%s>0");
93 hbar(hb_)=:minimum(d:min),
95 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
96 norm:!nonormal/normal,
97 incr:increment(d:inc,"%s>0");
100 +percentiles = double list;
101 +statistics[st_]=1|mean,2|semean,3|median,4|mode,5|stddev,6|variance,
102 7|kurtosis,8|skewness,9|range,10|minimum,11|maximum,12|sum,
103 13|default,14|seskewness,15|sekurtosis,all,none.
111 frq_mean = 0, frq_semean, frq_median, frq_mode, frq_stddev, frq_variance,
112 frq_kurt, frq_sekurt, frq_skew, frq_seskew, frq_range, frq_min, frq_max,
116 /* Description of a statistic. */
119 int st_indx; /* Index into a_statistics[]. */
120 const char *s10; /* Identifying string. */
123 /* Table of statistics, indexed by dsc_*. */
124 static const struct frq_info st_name[frq_n_stats + 1] =
126 {FRQ_ST_MEAN, N_("Mean")},
127 {FRQ_ST_SEMEAN, N_("S.E. Mean")},
128 {FRQ_ST_MEDIAN, N_("Median")},
129 {FRQ_ST_MODE, N_("Mode")},
130 {FRQ_ST_STDDEV, N_("Std Dev")},
131 {FRQ_ST_VARIANCE, N_("Variance")},
132 {FRQ_ST_KURTOSIS, N_("Kurtosis")},
133 {FRQ_ST_SEKURTOSIS, N_("S.E. Kurt")},
134 {FRQ_ST_SKEWNESS, N_("Skewness")},
135 {FRQ_ST_SESKEWNESS, N_("S.E. Skew")},
136 {FRQ_ST_RANGE, N_("Range")},
137 {FRQ_ST_MINIMUM, N_("Minimum")},
138 {FRQ_ST_MAXIMUM, N_("Maximum")},
139 {FRQ_ST_SUM, N_("Sum")},
143 /* Percentiles to calculate. */
147 double p; /* the %ile to be calculated */
148 double value; /* the %ile's value */
149 double x1; /* The datum value <= the percentile */
150 double x2; /* The datum value >= the percentile */
152 int flag2; /* Set to 1 if this percentile value has been found */
156 static void add_percentile (double x) ;
158 static struct percentile *percentiles;
159 static int n_percentiles;
161 static int implicit_50th ;
163 /* Groups of statistics. */
165 #define frq_default \
166 (BI (frq_mean) | BI (frq_stddev) | BI (frq_min) | BI (frq_max))
168 (BI (frq_sum) | BI(frq_min) | BI(frq_max) \
169 | BI(frq_mean) | BI(frq_semean) | BI(frq_stddev) \
170 | BI(frq_variance) | BI(frq_kurt) | BI(frq_sekurt) \
171 | BI(frq_skew) | BI(frq_seskew) | BI(frq_range) \
172 | BI(frq_range) | BI(frq_mode) | BI(frq_median))
174 /* Statistics; number of statistics. */
175 static unsigned long stats;
178 /* Types of graphs. */
181 GFT_NONE, /* Don't draw graphs. */
182 GFT_BAR, /* Draw bar charts. */
183 GFT_HIST, /* Draw histograms. */
184 GFT_PIE, /* Draw piechart */
185 GFT_HBAR /* Draw bar charts or histograms at our discretion. */
188 /* Parsed command. */
189 static struct cmd_frequencies cmd;
191 /* Summary of the barchart, histogram, and hbar subcommands. */
192 /* FIXME: These should not be mututally exclusive */
193 static int chart; /* NONE/BAR/HIST/HBAR/PIE. */
194 static double min, max; /* Minimum, maximum on y axis. */
195 static int format; /* FREQ/PERCENT: Scaling of y axis. */
196 static double scale, incr; /* FIXME */
197 static int normal; /* FIXME */
199 /* Variables for which to calculate statistics. */
200 static size_t n_variables;
201 static const struct variable **v_variables;
203 /* Arenas used to store semi-permanent storage. */
204 static struct pool *int_pool; /* Integer mode. */
205 static struct pool *gen_pool; /* General mode. */
207 /* Frequency tables. */
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];
256 /* Width and format for analysis and display.
257 This is normally the same as "width" and "print" in struct
258 variable, but in SPSS-compatible mode only the first
259 MAX_SHORT_STRING bytes of long string variables are
262 struct fmt_spec print;
265 static inline struct var_freqs *
266 get_var_freqs (const struct variable *v)
268 return var_get_aux (v);
271 static void determine_charts (void);
273 static void calc_stats (const struct variable *v, double d[frq_n_stats]);
275 static void precalc (struct casereader *, struct dataset *);
276 static void calc (const struct ccase *, const struct dataset *);
277 static void postcalc (void);
279 static void postprocess_freq_tab (const struct variable *);
280 static void dump_full (const struct variable *);
281 static void dump_condensed (const struct variable *);
282 static void dump_statistics (const struct variable *, int show_varname);
283 static void cleanup_freq_tab (const struct variable *);
285 static hsh_compare_func compare_value_numeric_a, compare_value_alpha_a;
286 static hsh_compare_func compare_value_numeric_d, compare_value_alpha_d;
287 static hsh_compare_func compare_freq_numeric_a, compare_freq_alpha_a;
288 static hsh_compare_func compare_freq_numeric_d, compare_freq_alpha_d;
291 static void do_piechart(const struct variable *var,
292 const struct freq_tab *frq_tab);
295 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var);
299 /* Parser and outline. */
301 static int internal_cmd_frequencies (struct lexer *lexer, struct dataset *ds);
304 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
308 int_pool = pool_create ();
309 result = internal_cmd_frequencies (lexer, ds);
310 pool_destroy (int_pool);
312 pool_destroy (gen_pool);
320 internal_cmd_frequencies (struct lexer *lexer, struct dataset *ds)
322 struct casegrouper *grouper;
323 struct casereader *input, *group;
333 if (!parse_frequencies (lexer, ds, &cmd, NULL))
336 if (cmd.onepage_limit == NOT_LONG)
337 cmd.onepage_limit = 50;
339 /* Figure out statistics to calculate. */
341 if (cmd.a_statistics[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
342 stats |= frq_default;
343 if (cmd.a_statistics[FRQ_ST_ALL])
345 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
346 stats &= ~frq_median;
347 for (i = 0; i < frq_n_stats; i++)
348 if (cmd.a_statistics[st_name[i].st_indx])
349 stats |= BIT_INDEX (i);
350 if (stats & frq_kurt)
352 if (stats & frq_skew)
355 /* Calculate n_stats. */
357 for (i = 0; i < frq_n_stats; i++)
358 if ((stats & BIT_INDEX (i)))
363 if (chart != GFT_NONE || cmd.sbc_ntiles)
364 cmd.sort = FRQ_AVALUE;
366 /* Work out what percentiles need to be calculated */
367 if ( cmd.sbc_percentiles )
369 for ( i = 0 ; i < MAXLISTS ; ++i )
372 subc_list_double *ptl_list = &cmd.dl_percentiles[i];
373 for ( pl = 0 ; pl < subc_list_double_count(ptl_list); ++pl)
374 add_percentile (subc_list_double_at(ptl_list, pl) / 100.0 );
377 if ( cmd.sbc_ntiles )
379 for ( i = 0 ; i < cmd.sbc_ntiles ; ++i )
382 for (j = 0; j <= cmd.n_ntiles[i]; ++j )
383 add_percentile (j / (double) cmd.n_ntiles[i]);
389 input = casereader_create_filter_weight (proc_open (ds), dataset_dict (ds),
391 grouper = casegrouper_create_splits (input, dataset_dict (ds));
392 for (; casegrouper_get_next_group (grouper, &group);
393 casereader_destroy (group))
398 for (; casereader_read (group, &c); case_destroy (&c))
402 ok = casegrouper_destroy (grouper);
403 ok = proc_commit (ds) && ok;
405 free_frequencies(&cmd);
407 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
410 /* Figure out which charts the user requested. */
412 determine_charts (void)
414 int count = (!!cmd.sbc_histogram) + (!!cmd.sbc_barchart) +
415 (!!cmd.sbc_hbar) + (!!cmd.sbc_piechart);
425 msg (SW, _("At most one of BARCHART, HISTOGRAM, or HBAR should be "
426 "given. HBAR will be assumed. Argument values will be "
427 "given precedence increasing along the order given."));
429 else if (cmd.sbc_histogram)
431 else if (cmd.sbc_barchart)
433 else if (cmd.sbc_piechart)
444 if (cmd.sbc_barchart)
446 if (cmd.ba_min != SYSMIS)
448 if (cmd.ba_max != SYSMIS)
450 if (cmd.ba_scale == FRQ_FREQ)
455 else if (cmd.ba_scale == FRQ_PERCENT)
457 format = FRQ_PERCENT;
462 if (cmd.sbc_histogram)
464 if (cmd.hi_min != SYSMIS)
466 if (cmd.hi_max != SYSMIS)
468 if (cmd.hi_scale == FRQ_FREQ)
473 else if (cmd.hi_scale == FRQ_PERCENT)
475 format = FRQ_PERCENT;
478 if (cmd.hi_norm != FRQ_NONORMAL )
480 if (cmd.hi_incr == FRQ_INCREMENT)
486 if (cmd.hb_min != SYSMIS)
488 if (cmd.hb_max != SYSMIS)
490 if (cmd.hb_scale == FRQ_FREQ)
495 else if (cmd.hb_scale == FRQ_PERCENT)
497 format = FRQ_PERCENT;
502 if (cmd.hb_incr == FRQ_INCREMENT)
506 if (min != SYSMIS && max != SYSMIS && min >= max)
508 msg (SE, _("MAX must be greater than or equal to MIN, if both are "
509 "specified. However, MIN was specified as %g and MAX as %g. "
510 "MIN and MAX will be ignored."), min, max);
515 /* Add data from case C to the frequency table. */
517 calc (const struct ccase *c, const struct dataset *ds)
519 double weight = dict_get_case_weight (dataset_dict (ds), c, NULL);
522 for (i = 0; i < n_variables; i++)
524 const struct variable *v = v_variables[i];
525 const union value *val = case_data (c, v);
526 struct var_freqs *vf = get_var_freqs (v);
527 struct freq_tab *ft = &vf->tab;
537 target.value = (union value *) val;
538 fpp = (struct freq **) hsh_probe (ft->data, &target);
541 (*fpp)->count += weight;
544 struct freq *fp = pool_alloc (gen_pool, sizeof *fp);
546 fp->value = pool_clone (gen_pool,
548 MAX (MAX_SHORT_STRING, vf->width));
555 if (val->f == SYSMIS)
556 ft->sysmis += weight;
557 else if (val->f > INT_MIN+1 && val->f < INT_MAX-1)
560 if (i >= ft->min && i <= ft->max)
561 ft->vector[i - ft->min] += weight;
564 ft->out_of_range += weight;
572 /* Prepares each variable that is the target of FREQUENCIES by setting
573 up its hash table. */
575 precalc (struct casereader *input, struct dataset *ds)
580 if (!casereader_peek (input, 0, &c))
582 output_split_file_values (ds, &c);
585 pool_destroy (gen_pool);
586 gen_pool = pool_create ();
588 for (i = 0; i < n_variables; i++)
590 const struct variable *v = v_variables[i];
591 struct freq_tab *ft = &get_var_freqs (v)->tab;
593 if (ft->mode == FRQM_GENERAL)
595 ft->data = hsh_create (16, compare_freq, hash_freq, NULL, v);
601 for (j = (ft->max - ft->min); j >= 0; j--)
603 ft->out_of_range = 0.0;
609 /* Finishes up with the variables after frequencies have been
610 calculated. Displays statistics, percentiles, ... */
616 for (i = 0; i < n_variables; i++)
618 const struct variable *v = v_variables[i];
619 struct var_freqs *vf = get_var_freqs (v);
620 struct freq_tab *ft = &vf->tab;
622 int dumped_freq_tab = 1;
624 postprocess_freq_tab (v);
626 /* Frequencies tables. */
627 n_categories = ft->n_valid + ft->n_missing;
628 if (cmd.table == FRQ_TABLE
629 || (cmd.table == FRQ_LIMIT && n_categories <= cmd.limit))
639 if (n_categories > cmd.onepage_limit)
652 dump_statistics (v, !dumped_freq_tab);
656 if ( chart == GFT_HIST)
658 double d[frq_n_stats];
659 struct normal_curve norm;
660 gsl_histogram *hist ;
663 norm.N = vf->tab.valid_cases;
666 norm.mean = d[frq_mean];
667 norm.stddev = d[frq_stddev];
669 hist = freq_tab_to_hist(ft,v);
671 histogram_plot(hist, var_to_string(v), &norm, normal);
673 gsl_histogram_free(hist);
677 if ( chart == GFT_PIE)
679 do_piechart(v_variables[i], ft);
684 cleanup_freq_tab (v);
689 /* Returns the comparison function that should be used for
690 sorting a frequency table by FRQ_SORT using VAR_TYPE
692 static hsh_compare_func *
693 get_freq_comparator (int frq_sort, enum var_type var_type)
695 bool is_numeric = var_type == VAR_NUMERIC;
699 return is_numeric ? compare_value_numeric_a : compare_value_alpha_a;
701 return is_numeric ? compare_value_numeric_d : compare_value_alpha_d;
703 return is_numeric ? compare_freq_numeric_a : compare_freq_alpha_a;
705 return is_numeric ? compare_freq_numeric_d : compare_freq_alpha_d;
711 /* Returns true iff the value in struct freq F is non-missing
714 not_missing (const void *f_, const void *v_)
716 const struct freq *f = f_;
717 const struct variable *v = v_;
719 return !var_is_value_missing (v, f->value, MV_ANY);
722 /* Summarizes the frequency table data for variable V. */
724 postprocess_freq_tab (const struct variable *v)
726 hsh_compare_func *compare;
730 struct freq *freqs, *f;
733 ft = &get_var_freqs (v)->tab;
734 assert (ft->mode == FRQM_GENERAL);
735 compare = get_freq_comparator (cmd.sort, var_get_type (v));
737 /* Extract data from hash table. */
738 count = hsh_count (ft->data);
739 data = hsh_data (ft->data);
741 /* Copy dereferenced data into freqs. */
742 freqs = xnmalloc (count, sizeof *freqs);
743 for (i = 0; i < count; i++)
745 struct freq *f = data[i];
749 /* Put data into ft. */
751 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, v);
752 ft->missing = freqs + ft->n_valid;
753 ft->n_missing = count - ft->n_valid;
756 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare, v);
757 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, v);
759 /* Summary statistics. */
760 ft->valid_cases = 0.0;
761 for(i = 0 ; i < ft->n_valid ; ++i )
764 ft->valid_cases += f->count;
768 ft->total_cases = ft->valid_cases ;
769 for(i = 0 ; i < ft->n_missing ; ++i )
772 ft->total_cases += f->count;
777 /* Frees the frequency table for variable V. */
779 cleanup_freq_tab (const struct variable *v)
781 struct freq_tab *ft = &get_var_freqs (v)->tab;
782 assert (ft->mode == FRQM_GENERAL);
784 hsh_destroy (ft->data);
787 /* Parses the VARIABLES subcommand, adding to
788 {n_variables,v_variables}. */
790 frq_custom_variables (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *aux UNUSED)
793 int min = 0, max = 0;
795 size_t old_n_variables = n_variables;
798 lex_match (lexer, '=');
799 if (lex_token (lexer) != T_ALL && (lex_token (lexer) != T_ID
800 || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) == NULL))
803 if (!parse_variables_const (lexer, dataset_dict (ds), &v_variables, &n_variables,
804 PV_APPEND | PV_NO_SCRATCH))
807 if (!lex_match (lexer, '('))
812 if (!lex_force_int (lexer))
814 min = lex_integer (lexer);
816 if (!lex_force_match (lexer, ','))
818 if (!lex_force_int (lexer))
820 max = lex_integer (lexer);
822 if (!lex_force_match (lexer, ')'))
826 msg (SE, _("Upper limit of integer mode value range must be "
827 "greater than lower limit."));
832 for (i = old_n_variables; i < n_variables; i++)
834 const struct variable *v = v_variables[i];
835 struct var_freqs *vf;
837 if (var_get_aux (v) != NULL)
839 msg (SE, _("Variable %s specified multiple times on VARIABLES "
840 "subcommand."), var_get_name (v));
843 if (mode == FRQM_INTEGER && !var_is_numeric (v))
845 msg (SE, _("Integer mode specified, but %s is not a numeric "
846 "variable."), var_get_name (v));
850 vf = var_attach_aux (v, xmalloc (sizeof *vf), var_dtor_free);
852 vf->tab.valid = vf->tab.missing = NULL;
853 if (mode == FRQM_INTEGER)
857 vf->tab.vector = pool_nalloc (int_pool,
858 max - min + 1, sizeof *vf->tab.vector);
861 vf->tab.vector = NULL;
864 vf->width = var_get_width (v);
865 vf->print = *var_get_print_format (v);
866 if (vf->width > MAX_SHORT_STRING && get_algorithm () == COMPATIBLE)
868 enum fmt_type type = var_get_print_format (v)->type;
869 vf->width = MAX_SHORT_STRING;
870 vf->print.w = MAX_SHORT_STRING * (type == FMT_AHEX ? 2 : 1);
876 /* Parses the GROUPED subcommand, setting the n_grouped, grouped
877 fields of specified variables. */
879 frq_custom_grouped (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *aux UNUSED)
881 lex_match (lexer, '=');
882 if ((lex_token (lexer) == T_ID && dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
883 || lex_token (lexer) == T_ID)
888 /* Max, current size of list; list itself. */
894 const struct variable **v;
896 if (!parse_variables_const (lexer, dataset_dict (ds), &v, &n,
897 PV_NO_DUPLICATE | PV_NUMERIC))
899 if (lex_match (lexer, '('))
903 while (lex_integer (lexer))
908 dl = pool_nrealloc (int_pool, dl, ml, sizeof *dl);
910 dl[nl++] = lex_tokval (lexer);
912 lex_match (lexer, ',');
914 /* Note that nl might still be 0 and dl might still be
915 NULL. That's okay. */
916 if (!lex_match (lexer, ')'))
919 msg (SE, _("`)' expected after GROUPED interval list."));
929 for (i = 0; i < n; i++)
930 if (var_get_aux (v[i]) == NULL)
931 msg (SE, _("Variables %s specified on GROUPED but not on "
932 "VARIABLES."), var_get_name (v[i]));
935 struct var_freqs *vf = get_var_freqs (v[i]);
937 if (vf->groups != NULL)
938 msg (SE, _("Variables %s specified multiple times on GROUPED "
939 "subcommand."), var_get_name (v[i]));
947 if (!lex_match (lexer, '/'))
949 if ((lex_token (lexer) != T_ID || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
950 && lex_token (lexer) != T_ALL)
952 lex_put_back (lexer, '/');
960 /* Adds X to the list of percentiles, keeping the list in proper
963 add_percentile (double x)
967 for (i = 0; i < n_percentiles; i++)
969 /* Do nothing if it's already in the list */
970 if ( fabs(x - percentiles[i].p) < DBL_EPSILON )
973 if (x < percentiles[i].p)
977 if (i >= n_percentiles || x != percentiles[i].p)
979 percentiles = pool_nrealloc (int_pool, percentiles,
980 n_percentiles + 1, sizeof *percentiles);
982 if (i < n_percentiles)
983 memmove (&percentiles[i + 1], &percentiles[i],
984 (n_percentiles - i) * sizeof (struct percentile) );
986 percentiles[i].p = x;
991 /* Comparison functions. */
993 /* Ascending numeric compare of values. */
995 compare_value_numeric_a (const void *a_, const void *b_, const void *aux UNUSED)
997 const struct freq *a = a_;
998 const struct freq *b = b_;
1000 if (a->value[0].f > b->value[0].f)
1002 else if (a->value[0].f < b->value[0].f)
1008 /* Ascending string compare of values. */
1010 compare_value_alpha_a (const void *a_, const void *b_, const void *v_)
1012 const struct freq *a = a_;
1013 const struct freq *b = b_;
1014 const struct variable *v = v_;
1015 struct var_freqs *vf = get_var_freqs (v);
1017 return memcmp (a->value[0].s, b->value[0].s, vf->width);
1020 /* Descending numeric compare of values. */
1022 compare_value_numeric_d (const void *a, const void *b, const void *aux UNUSED)
1024 return -compare_value_numeric_a (a, b, aux);
1027 /* Descending string compare of values. */
1029 compare_value_alpha_d (const void *a, const void *b, const void *v)
1031 return -compare_value_alpha_a (a, b, v);
1034 /* Ascending numeric compare of frequency;
1035 secondary key on ascending numeric value. */
1037 compare_freq_numeric_a (const void *a_, const void *b_, const void *aux UNUSED)
1039 const struct freq *a = a_;
1040 const struct freq *b = b_;
1042 if (a->count > b->count)
1044 else if (a->count < b->count)
1047 if (a->value[0].f > b->value[0].f)
1049 else if (a->value[0].f < b->value[0].f)
1055 /* Ascending numeric compare of frequency;
1056 secondary key on ascending string value. */
1058 compare_freq_alpha_a (const void *a_, const void *b_, const void *v_)
1060 const struct freq *a = a_;
1061 const struct freq *b = b_;
1062 const struct variable *v = v_;
1063 struct var_freqs *vf = get_var_freqs (v);
1065 if (a->count > b->count)
1067 else if (a->count < b->count)
1070 return memcmp (a->value[0].s, b->value[0].s, vf->width);
1073 /* Descending numeric compare of frequency;
1074 secondary key on ascending numeric value. */
1076 compare_freq_numeric_d (const void *a_, const void *b_, const void *aux UNUSED)
1078 const struct freq *a = a_;
1079 const struct freq *b = b_;
1081 if (a->count > b->count)
1083 else if (a->count < b->count)
1086 if (a->value[0].f > b->value[0].f)
1088 else if (a->value[0].f < b->value[0].f)
1094 /* Descending numeric compare of frequency;
1095 secondary key on ascending string value. */
1097 compare_freq_alpha_d (const void *a_, const void *b_, const void *v_)
1099 const struct freq *a = a_;
1100 const struct freq *b = b_;
1101 const struct variable *v = v_;
1102 struct var_freqs *vf = get_var_freqs (v);
1104 if (a->count > b->count)
1106 else if (a->count < b->count)
1109 return memcmp (a->value[0].s, b->value[0].s, vf->width);
1112 /* Frequency table display. */
1114 /* Sets the widths of all the columns and heights of all the rows in
1115 table T for driver D. */
1117 full_dim (struct tab_table *t, struct outp_driver *d)
1122 if (cmd.labels == FRQ_LABELS)
1124 t->w[0] = MIN (tab_natural_width (t, d, 0), d->prop_em_width * 15);
1129 for (;i < columns; i++)
1130 t->w[i] = MAX (tab_natural_width (t, d, i), d->prop_em_width * 8);
1132 for (i = 0; i < t->nr; i++)
1133 t->h[i] = d->font_height;
1136 /* Displays a full frequency table for variable V. */
1138 dump_full (const struct variable *v)
1141 struct var_freqs *vf;
1142 struct freq_tab *ft;
1144 struct tab_table *t;
1146 double cum_total = 0.0;
1147 double cum_freq = 0.0;
1155 const struct init *p;
1157 static const struct init vec[] =
1159 {4, 0, N_("Valid")},
1161 {1, 1, N_("Value")},
1162 {2, 1, N_("Frequency")},
1163 {3, 1, N_("Percent")},
1164 {4, 1, N_("Percent")},
1165 {5, 1, N_("Percent")},
1173 const bool lab = (cmd.labels == FRQ_LABELS);
1175 vf = get_var_freqs (v);
1177 n_categories = ft->n_valid + ft->n_missing;
1178 t = tab_create (5 + lab, n_categories + 3, 0);
1179 tab_headers (t, 0, 0, 2, 0);
1180 tab_dim (t, full_dim);
1183 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value Label"));
1185 for (p = vec; p->s; p++)
1186 tab_text (t, lab ? p->c : p->c - 1, p->r,
1187 TAB_CENTER | TAT_TITLE, gettext (p->s));
1190 for (f = ft->valid; f < ft->missing; f++)
1192 double percent, valid_percent;
1194 cum_freq += f->count;
1196 percent = f->count / ft->total_cases * 100.0;
1197 valid_percent = f->count / ft->valid_cases * 100.0;
1198 cum_total += valid_percent;
1202 const char *label = var_lookup_value_label (v, &f->value[0]);
1204 tab_text (t, 0, r, TAB_LEFT, label);
1207 tab_value (t, 0 + lab, r, TAB_NONE, f->value, &vf->print);
1208 tab_float (t, 1 + lab, r, TAB_NONE, f->count, 8, 0);
1209 tab_float (t, 2 + lab, r, TAB_NONE, percent, 5, 1);
1210 tab_float (t, 3 + lab, r, TAB_NONE, valid_percent, 5, 1);
1211 tab_float (t, 4 + lab, r, TAB_NONE, cum_total, 5, 1);
1214 for (; f < &ft->valid[n_categories]; f++)
1216 cum_freq += f->count;
1220 const char *label = var_lookup_value_label (v, &f->value[0]);
1222 tab_text (t, 0, r, TAB_LEFT, label);
1225 tab_value (t, 0 + lab, r, TAB_NONE, f->value, &vf->print);
1226 tab_float (t, 1 + lab, r, TAB_NONE, f->count, 8, 0);
1227 tab_float (t, 2 + lab, r, TAB_NONE,
1228 f->count / ft->total_cases * 100.0, 5, 1);
1229 tab_text (t, 3 + lab, r, TAB_NONE, _("Missing"));
1233 tab_box (t, TAL_1, TAL_1,
1234 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1236 tab_hline (t, TAL_2, 0, 4 + lab, 2);
1237 tab_hline (t, TAL_2, 0, 4 + lab, r);
1238 tab_joint_text (t, 0, r, 0 + lab, r, TAB_RIGHT | TAT_TITLE, _("Total"));
1239 tab_vline (t, TAL_0, 1, r, r);
1240 tab_float (t, 1 + lab, r, TAB_NONE, cum_freq, 8, 0);
1241 tab_float (t, 2 + lab, r, TAB_NONE, 100.0, 5, 1);
1242 tab_float (t, 3 + lab, r, TAB_NONE, 100.0, 5, 1);
1244 tab_title (t, "%s", var_to_string (v));
1248 /* Sets the widths of all the columns and heights of all the rows in
1249 table T for driver D. */
1251 condensed_dim (struct tab_table *t, struct outp_driver *d)
1253 int cum_w = MAX (outp_string_width (d, _("Cum"), OUTP_PROPORTIONAL),
1254 MAX (outp_string_width (d, _("Cum"), OUTP_PROPORTIONAL),
1255 outp_string_width (d, "000", OUTP_PROPORTIONAL)));
1259 for (i = 0; i < 2; i++)
1260 t->w[i] = MAX (tab_natural_width (t, d, i), d->prop_em_width * 8);
1261 for (i = 2; i < 4; i++)
1263 for (i = 0; i < t->nr; i++)
1264 t->h[i] = d->font_height;
1267 /* Display condensed frequency table for variable V. */
1269 dump_condensed (const struct variable *v)
1272 struct var_freqs *vf;
1273 struct freq_tab *ft;
1275 struct tab_table *t;
1277 double cum_total = 0.0;
1279 vf = get_var_freqs (v);
1281 n_categories = ft->n_valid + ft->n_missing;
1282 t = tab_create (4, n_categories + 2, 0);
1284 tab_headers (t, 0, 0, 2, 0);
1285 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value"));
1286 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Freq"));
1287 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1288 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Cum"));
1289 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1290 tab_dim (t, condensed_dim);
1293 for (f = ft->valid; f < ft->missing; f++)
1297 percent = f->count / ft->total_cases * 100.0;
1298 cum_total += f->count / ft->valid_cases * 100.0;
1300 tab_value (t, 0, r, TAB_NONE, f->value, &vf->print);
1301 tab_float (t, 1, r, TAB_NONE, f->count, 8, 0);
1302 tab_float (t, 2, r, TAB_NONE, percent, 3, 0);
1303 tab_float (t, 3, r, TAB_NONE, cum_total, 3, 0);
1306 for (; f < &ft->valid[n_categories]; f++)
1308 tab_value (t, 0, r, TAB_NONE, f->value, &vf->print);
1309 tab_float (t, 1, r, TAB_NONE, f->count, 8, 0);
1310 tab_float (t, 2, r, TAB_NONE,
1311 f->count / ft->total_cases * 100.0, 3, 0);
1315 tab_box (t, TAL_1, TAL_1,
1316 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1318 tab_hline (t, TAL_2, 0, 3, 2);
1319 tab_title (t, "%s", var_to_string (v));
1320 tab_columns (t, SOM_COL_DOWN, 1);
1324 /* Statistical display. */
1326 /* Calculates all the pertinent statistics for variable V, putting
1327 them in array D[]. FIXME: This could be made much more optimal. */
1329 calc_stats (const struct variable *v, double d[frq_n_stats])
1331 struct freq_tab *ft = &get_var_freqs (v)->tab;
1332 double W = ft->valid_cases;
1341 double *median_value;
1343 /* Calculate percentiles. */
1345 /* If the 50th percentile was not explicitly requested then we must
1346 calculate it anyway --- it's the median */
1348 for (i = 0; i < n_percentiles; i++)
1350 if (percentiles[i].p == 0.5)
1352 median_value = &percentiles[i].value;
1357 if ( 0 == median_value )
1359 add_percentile (0.5);
1363 for (i = 0; i < n_percentiles; i++)
1365 percentiles[i].flag = 0;
1366 percentiles[i].flag2 = 0;
1370 for (idx = 0; idx < ft->n_valid; ++idx)
1372 static double prev_value = SYSMIS;
1373 f = &ft->valid[idx];
1375 for (i = 0; i < n_percentiles; i++)
1378 if ( percentiles[i].flag2 ) continue ;
1380 if ( get_algorithm() != COMPATIBLE )
1382 (ft->valid_cases - 1) * percentiles[i].p;
1385 (ft->valid_cases + 1) * percentiles[i].p - 1;
1387 if ( percentiles[i].flag )
1389 percentiles[i].x2 = f->value[0].f;
1390 percentiles[i].x1 = prev_value;
1391 percentiles[i].flag2 = 1;
1397 if ( f->count > 1 && rank - (f->count - 1) > tp )
1399 percentiles[i].x2 = percentiles[i].x1 = f->value[0].f;
1400 percentiles[i].flag2 = 1;
1404 percentiles[i].flag=1;
1410 prev_value = f->value[0].f;
1413 for (i = 0; i < n_percentiles; i++)
1415 /* Catches the case when p == 100% */
1416 if ( ! percentiles[i].flag2 )
1417 percentiles[i].x1 = percentiles[i].x2 = f->value[0].f;
1420 printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
1421 i,percentiles[i].p,percentiles[i].x1,percentiles[i].x2);
1425 for (i = 0; i < n_percentiles; i++)
1427 struct freq_tab *ft = &get_var_freqs (v)->tab;
1431 if ( get_algorithm() != COMPATIBLE )
1433 s = modf((ft->valid_cases - 1) * percentiles[i].p , &dummy);
1437 s = modf((ft->valid_cases + 1) * percentiles[i].p -1, &dummy);
1440 percentiles[i].value = percentiles[i].x1 +
1441 ( percentiles[i].x2 - percentiles[i].x1) * s ;
1443 if ( percentiles[i].p == 0.50)
1444 median_value = &percentiles[i].value;
1448 /* Calculate the mode. */
1451 for (f = ft->valid; f < ft->missing; f++)
1453 if (most_often < f->count)
1455 most_often = f->count;
1456 X_mode = f->value[0].f;
1458 else if (most_often == f->count)
1460 /* A duplicate mode is undefined.
1461 FIXME: keep track of *all* the modes. */
1466 /* Calculate moments. */
1467 m = moments_create (MOMENT_KURTOSIS);
1468 for (f = ft->valid; f < ft->missing; f++)
1469 moments_pass_one (m, f->value[0].f, f->count);
1470 for (f = ft->valid; f < ft->missing; f++)
1471 moments_pass_two (m, f->value[0].f, f->count);
1472 moments_calculate (m, NULL, &d[frq_mean], &d[frq_variance],
1473 &d[frq_skew], &d[frq_kurt]);
1474 moments_destroy (m);
1476 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1477 d[frq_min] = ft->valid[0].value[0].f;
1478 d[frq_max] = ft->valid[ft->n_valid - 1].value[0].f;
1479 d[frq_mode] = X_mode;
1480 d[frq_range] = d[frq_max] - d[frq_min];
1481 d[frq_median] = *median_value;
1482 d[frq_sum] = d[frq_mean] * W;
1483 d[frq_stddev] = sqrt (d[frq_variance]);
1484 d[frq_semean] = d[frq_stddev] / sqrt (W);
1485 d[frq_seskew] = calc_seskew (W);
1486 d[frq_sekurt] = calc_sekurt (W);
1489 /* Displays a table of all the statistics requested for variable V. */
1491 dump_statistics (const struct variable *v, int show_varname)
1493 struct freq_tab *ft;
1494 double stat_value[frq_n_stats];
1495 struct tab_table *t;
1498 int n_explicit_percentiles = n_percentiles;
1500 if ( implicit_50th && n_percentiles > 0 )
1503 if (var_is_alpha (v))
1505 ft = &get_var_freqs (v)->tab;
1506 if (ft->n_valid == 0)
1508 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1512 calc_stats (v, stat_value);
1514 t = tab_create (3, n_stats + n_explicit_percentiles + 2, 0);
1515 tab_dim (t, tab_natural_dimensions);
1517 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1520 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1521 tab_vline (t, TAL_GAP , 1, 0, tab_nr(t) - 1 ) ;
1523 r=2; /* N missing and N valid are always dumped */
1525 for (i = 0; i < frq_n_stats; i++)
1526 if (stats & BIT_INDEX (i))
1528 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1529 gettext (st_name[i].s10));
1530 tab_float (t, 2, r, TAB_NONE, stat_value[i], 11, 3);
1534 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1535 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1536 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1538 tab_float(t, 2, 0, TAB_NONE, ft->valid_cases, 11, 0);
1539 tab_float(t, 2, 1, TAB_NONE, ft->total_cases - ft->valid_cases, 11, 0);
1542 for (i = 0; i < n_explicit_percentiles; i++, r++)
1546 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1549 tab_float (t, 1, r, TAB_LEFT, percentiles[i].p * 100, 3, 0 );
1550 tab_float (t, 2, r, TAB_NONE, percentiles[i].value, 11, 3);
1554 tab_columns (t, SOM_COL_DOWN, 1);
1556 tab_title (t, "%s", var_to_string (v));
1558 tab_flags (t, SOMF_NO_TITLE);
1565 /* Create a gsl_histogram from a freq_tab */
1567 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var)
1570 double x_min = DBL_MAX;
1571 double x_max = -DBL_MAX;
1573 gsl_histogram *hist;
1574 const double bins = 11;
1576 struct hsh_iterator hi;
1577 struct hsh_table *fh = ft->data;
1580 /* Find out the extremes of the x value */
1581 for ( frq = hsh_first(fh, &hi); frq != 0; frq = hsh_next(fh, &hi) )
1583 if (var_is_value_missing(var, frq->value, MV_ANY))
1586 if ( frq->value[0].f < x_min ) x_min = frq->value[0].f ;
1587 if ( frq->value[0].f > x_max ) x_max = frq->value[0].f ;
1590 hist = histogram_create(bins, x_min, x_max);
1592 for( i = 0 ; i < ft->n_valid ; ++i )
1594 frq = &ft->valid[i];
1595 gsl_histogram_accumulate(hist, frq->value[0].f, frq->count);
1602 static struct slice *
1603 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1604 const struct variable *var,
1608 /* Allocate an array of slices and fill them from the data in frq_tab
1609 n_slices will contain the number of slices allocated.
1610 The caller is responsible for freeing slices
1612 static struct slice *
1613 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1614 const struct variable *var,
1618 struct slice *slices;
1620 *n_slices = frq_tab->n_valid;
1622 slices = xnmalloc (*n_slices, sizeof *slices);
1624 for (i = 0 ; i < *n_slices ; ++i )
1626 const struct freq *frq = &frq_tab->valid[i];
1628 slices[i].label = var_get_value_name (var, frq->value);
1629 slices[i].magnetude = frq->count;
1639 do_piechart(const struct variable *var, const struct freq_tab *frq_tab)
1641 struct slice *slices;
1644 slices = freq_tab_to_slice_array(frq_tab, var, &n_slices);
1646 piechart_plot(var_to_string(var), slices, n_slices);