1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU 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, see <http://www.gnu.org/licenses/>. */
20 * Remember that histograms, bar charts need mean, stddev.
27 #include <gsl/gsl_histogram.h>
29 #include <data/case.h>
30 #include <data/casegrouper.h>
31 #include <data/casereader.h>
32 #include <data/dictionary.h>
33 #include <data/format.h>
34 #include <data/procedure.h>
35 #include <data/settings.h>
36 #include <data/value-labels.h>
37 #include <data/variable.h>
38 #include <language/command.h>
39 #include <language/dictionary/split-file.h>
40 #include <language/lexer/lexer.h>
41 #include <libpspp/alloc.h>
42 #include <libpspp/array.h>
43 #include <libpspp/bit-vector.h>
44 #include <libpspp/compiler.h>
45 #include <libpspp/hash.h>
46 #include <libpspp/magic.h>
47 #include <libpspp/message.h>
48 #include <libpspp/misc.h>
49 #include <libpspp/pool.h>
50 #include <libpspp/str.h>
51 #include <math/histogram.h>
52 #include <math/moments.h>
53 #include <output/chart.h>
54 #include <output/charts/piechart.h>
55 #include <output/charts/plot-hist.h>
56 #include <output/manager.h>
57 #include <output/output.h>
58 #include <output/table.h>
65 #define _(msgid) gettext (msgid)
66 #define N_(msgid) msgid
73 +format=cond:condense/onepage(*n:onepage_limit,"%s>=0")/!standard,
74 table:limit(n:limit,"%s>0")/notable/!table,
75 labels:!labels/nolabels,
76 sort:!avalue/dvalue/afreq/dfreq,
77 spaces:!single/double,
78 paging:newpage/!oldpage;
79 missing=miss:include/!exclude;
80 barchart(ba_)=:minimum(d:min),
82 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0");
83 piechart(pie_)=:minimum(d:min),
85 missing:missing/!nomissing;
86 histogram(hi_)=: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");
91 hbar(hb_)=:minimum(d:min),
93 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
94 norm:!nonormal/normal,
95 incr:increment(d:inc,"%s>0");
98 +percentiles = double list;
99 +statistics[st_]=1|mean,2|semean,3|median,4|mode,5|stddev,6|variance,
100 7|kurtosis,8|skewness,9|range,10|minimum,11|maximum,12|sum,
101 13|default,14|seskewness,15|sekurtosis,all,none.
109 frq_mean = 0, frq_semean, frq_median, frq_mode, frq_stddev, frq_variance,
110 frq_kurt, frq_sekurt, frq_skew, frq_seskew, frq_range, frq_min, frq_max,
114 /* Description of a statistic. */
117 int st_indx; /* Index into a_statistics[]. */
118 const char *s10; /* Identifying string. */
121 /* Table of statistics, indexed by dsc_*. */
122 static const struct frq_info st_name[frq_n_stats + 1] =
124 {FRQ_ST_MEAN, N_("Mean")},
125 {FRQ_ST_SEMEAN, N_("S.E. Mean")},
126 {FRQ_ST_MEDIAN, N_("Median")},
127 {FRQ_ST_MODE, N_("Mode")},
128 {FRQ_ST_STDDEV, N_("Std Dev")},
129 {FRQ_ST_VARIANCE, N_("Variance")},
130 {FRQ_ST_KURTOSIS, N_("Kurtosis")},
131 {FRQ_ST_SEKURTOSIS, N_("S.E. Kurt")},
132 {FRQ_ST_SKEWNESS, N_("Skewness")},
133 {FRQ_ST_SESKEWNESS, N_("S.E. Skew")},
134 {FRQ_ST_RANGE, N_("Range")},
135 {FRQ_ST_MINIMUM, N_("Minimum")},
136 {FRQ_ST_MAXIMUM, N_("Maximum")},
137 {FRQ_ST_SUM, N_("Sum")},
141 /* Percentiles to calculate. */
145 double p; /* the %ile to be calculated */
146 double value; /* the %ile's value */
147 double x1; /* The datum value <= the percentile */
148 double x2; /* The datum value >= the percentile */
150 int flag2; /* Set to 1 if this percentile value has been found */
154 static void add_percentile (double x) ;
156 static struct percentile *percentiles;
157 static int n_percentiles;
159 static int implicit_50th ;
161 /* Groups of statistics. */
163 #define frq_default \
164 (BI (frq_mean) | BI (frq_stddev) | BI (frq_min) | BI (frq_max))
166 (BI (frq_sum) | BI(frq_min) | BI(frq_max) \
167 | BI(frq_mean) | BI(frq_semean) | BI(frq_stddev) \
168 | BI(frq_variance) | BI(frq_kurt) | BI(frq_sekurt) \
169 | BI(frq_skew) | BI(frq_seskew) | BI(frq_range) \
170 | BI(frq_range) | BI(frq_mode) | BI(frq_median))
172 /* Statistics; number of statistics. */
173 static unsigned long stats;
176 /* Types of graphs. */
179 GFT_NONE, /* Don't draw graphs. */
180 GFT_BAR, /* Draw bar charts. */
181 GFT_HIST, /* Draw histograms. */
182 GFT_PIE, /* Draw piechart */
183 GFT_HBAR /* Draw bar charts or histograms at our discretion. */
186 /* Parsed command. */
187 static struct cmd_frequencies cmd;
189 /* Summary of the barchart, histogram, and hbar subcommands. */
190 /* FIXME: These should not be mututally exclusive */
191 static int chart; /* NONE/BAR/HIST/HBAR/PIE. */
192 static double min, max; /* Minimum, maximum on y axis. */
193 static int format; /* FREQ/PERCENT: Scaling of y axis. */
194 static double scale, incr; /* FIXME */
195 static int normal; /* FIXME */
197 /* Variables for which to calculate statistics. */
198 static size_t n_variables;
199 static const struct variable **v_variables;
201 /* Arenas used to store semi-permanent storage. */
202 static struct pool *int_pool; /* Integer mode. */
203 static struct pool *gen_pool; /* General mode. */
205 /* Frequency tables. */
207 /* Types of frequency tables. */
214 /* Entire frequency table. */
217 int mode; /* FRQM_GENERAL or FRQM_INTEGER. */
220 struct hsh_table *data; /* Undifferentiated data. */
223 double *vector; /* Frequencies proper. */
224 int min, max; /* The boundaries of the table. */
225 double out_of_range; /* Sum of weights of out-of-range values. */
226 double sysmis; /* Sum of weights of SYSMIS values. */
229 struct freq *valid; /* Valid freqs. */
230 int n_valid; /* Number of total freqs. */
232 struct freq *missing; /* Missing freqs. */
233 int n_missing; /* Number of missing freqs. */
236 double total_cases; /* Sum of weights of all cases. */
237 double valid_cases; /* Sum of weights of valid cases. */
241 /* Per-variable frequency data. */
244 /* Freqency table. */
245 struct freq_tab tab; /* Frequencies table to use. */
248 int n_groups; /* Number of groups. */
249 double *groups; /* Groups. */
252 double stat[frq_n_stats];
254 /* Width and format for analysis and display.
255 This is normally the same as "width" and "print" in struct
256 variable, but in SPSS-compatible mode only the first
257 MAX_SHORT_STRING bytes of long string variables are
260 struct fmt_spec print;
263 static inline struct var_freqs *
264 get_var_freqs (const struct variable *v)
266 return var_get_aux (v);
269 static void determine_charts (void);
271 static void calc_stats (const struct variable *v, double d[frq_n_stats]);
273 static void precalc (struct casereader *, struct dataset *);
274 static void calc (const struct ccase *, const struct dataset *);
275 static void postcalc (void);
277 static void postprocess_freq_tab (const struct variable *);
278 static void dump_full (const struct variable *);
279 static void dump_condensed (const struct variable *);
280 static void dump_statistics (const struct variable *, int show_varname);
281 static void cleanup_freq_tab (const struct variable *);
283 static hsh_compare_func compare_value_numeric_a, compare_value_alpha_a;
284 static hsh_compare_func compare_value_numeric_d, compare_value_alpha_d;
285 static hsh_compare_func compare_freq_numeric_a, compare_freq_alpha_a;
286 static hsh_compare_func compare_freq_numeric_d, compare_freq_alpha_d;
289 static void do_piechart(const struct variable *var,
290 const struct freq_tab *frq_tab);
293 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var);
297 /* Parser and outline. */
299 static int internal_cmd_frequencies (struct lexer *lexer, struct dataset *ds);
302 cmd_frequencies (struct lexer *lexer, struct dataset *ds)
306 int_pool = pool_create ();
307 result = internal_cmd_frequencies (lexer, ds);
308 pool_destroy (int_pool);
310 pool_destroy (gen_pool);
318 internal_cmd_frequencies (struct lexer *lexer, struct dataset *ds)
320 struct casegrouper *grouper;
321 struct casereader *input, *group;
331 if (!parse_frequencies (lexer, ds, &cmd, NULL))
334 if (cmd.onepage_limit == NOT_LONG)
335 cmd.onepage_limit = 50;
337 /* Figure out statistics to calculate. */
339 if (cmd.a_statistics[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
340 stats |= frq_default;
341 if (cmd.a_statistics[FRQ_ST_ALL])
343 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
344 stats &= ~frq_median;
345 for (i = 0; i < frq_n_stats; i++)
346 if (cmd.a_statistics[st_name[i].st_indx])
347 stats |= BIT_INDEX (i);
348 if (stats & frq_kurt)
350 if (stats & frq_skew)
353 /* Calculate n_stats. */
355 for (i = 0; i < frq_n_stats; i++)
356 if ((stats & BIT_INDEX (i)))
361 if (chart != GFT_NONE || cmd.sbc_ntiles)
362 cmd.sort = FRQ_AVALUE;
364 /* Work out what percentiles need to be calculated */
365 if ( cmd.sbc_percentiles )
367 for ( i = 0 ; i < MAXLISTS ; ++i )
370 subc_list_double *ptl_list = &cmd.dl_percentiles[i];
371 for ( pl = 0 ; pl < subc_list_double_count(ptl_list); ++pl)
372 add_percentile (subc_list_double_at(ptl_list, pl) / 100.0 );
375 if ( cmd.sbc_ntiles )
377 for ( i = 0 ; i < cmd.sbc_ntiles ; ++i )
380 for (j = 0; j <= cmd.n_ntiles[i]; ++j )
381 add_percentile (j / (double) cmd.n_ntiles[i]);
387 input = casereader_create_filter_weight (proc_open (ds), dataset_dict (ds),
389 grouper = casegrouper_create_splits (input, dataset_dict (ds));
390 for (; casegrouper_get_next_group (grouper, &group);
391 casereader_destroy (group))
396 for (; casereader_read (group, &c); case_destroy (&c))
400 ok = casegrouper_destroy (grouper);
401 ok = proc_commit (ds) && ok;
403 free_frequencies(&cmd);
405 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
408 /* Figure out which charts the user requested. */
410 determine_charts (void)
412 int count = (!!cmd.sbc_histogram) + (!!cmd.sbc_barchart) +
413 (!!cmd.sbc_hbar) + (!!cmd.sbc_piechart);
423 msg (SW, _("At most one of BARCHART, HISTOGRAM, or HBAR should be "
424 "given. HBAR will be assumed. Argument values will be "
425 "given precedence increasing along the order given."));
427 else if (cmd.sbc_histogram)
429 else if (cmd.sbc_barchart)
431 else if (cmd.sbc_piechart)
442 if (cmd.sbc_barchart)
444 if (cmd.ba_min != SYSMIS)
446 if (cmd.ba_max != SYSMIS)
448 if (cmd.ba_scale == FRQ_FREQ)
453 else if (cmd.ba_scale == FRQ_PERCENT)
455 format = FRQ_PERCENT;
460 if (cmd.sbc_histogram)
462 if (cmd.hi_min != SYSMIS)
464 if (cmd.hi_max != SYSMIS)
466 if (cmd.hi_scale == FRQ_FREQ)
471 else if (cmd.hi_scale == FRQ_PERCENT)
473 format = FRQ_PERCENT;
476 if (cmd.hi_norm != FRQ_NONORMAL )
478 if (cmd.hi_incr == FRQ_INCREMENT)
484 if (cmd.hb_min != SYSMIS)
486 if (cmd.hb_max != SYSMIS)
488 if (cmd.hb_scale == FRQ_FREQ)
493 else if (cmd.hb_scale == FRQ_PERCENT)
495 format = FRQ_PERCENT;
500 if (cmd.hb_incr == FRQ_INCREMENT)
504 if (min != SYSMIS && max != SYSMIS && min >= max)
506 msg (SE, _("MAX must be greater than or equal to MIN, if both are "
507 "specified. However, MIN was specified as %g and MAX as %g. "
508 "MIN and MAX will be ignored."), min, max);
513 /* Add data from case C to the frequency table. */
515 calc (const struct ccase *c, const struct dataset *ds)
517 double weight = dict_get_case_weight (dataset_dict (ds), c, NULL);
520 for (i = 0; i < n_variables; i++)
522 const struct variable *v = v_variables[i];
523 const union value *val = case_data (c, v);
524 struct var_freqs *vf = get_var_freqs (v);
525 struct freq_tab *ft = &vf->tab;
535 target.value = (union value *) val;
536 fpp = (struct freq **) hsh_probe (ft->data, &target);
539 (*fpp)->count += weight;
542 struct freq *fp = pool_alloc (gen_pool, sizeof *fp);
544 fp->value = pool_clone (gen_pool,
546 MAX (MAX_SHORT_STRING, vf->width));
553 if (val->f == SYSMIS)
554 ft->sysmis += weight;
555 else if (val->f > INT_MIN+1 && val->f < INT_MAX-1)
558 if (i >= ft->min && i <= ft->max)
559 ft->vector[i - ft->min] += weight;
562 ft->out_of_range += weight;
570 /* Prepares each variable that is the target of FREQUENCIES by setting
571 up its hash table. */
573 precalc (struct casereader *input, struct dataset *ds)
578 if (!casereader_peek (input, 0, &c))
580 output_split_file_values (ds, &c);
583 pool_destroy (gen_pool);
584 gen_pool = pool_create ();
586 for (i = 0; i < n_variables; i++)
588 const struct variable *v = v_variables[i];
589 struct freq_tab *ft = &get_var_freqs (v)->tab;
591 if (ft->mode == FRQM_GENERAL)
593 ft->data = hsh_create (16, compare_freq, hash_freq, NULL, v);
599 for (j = (ft->max - ft->min); j >= 0; j--)
601 ft->out_of_range = 0.0;
607 /* Finishes up with the variables after frequencies have been
608 calculated. Displays statistics, percentiles, ... */
614 for (i = 0; i < n_variables; i++)
616 const struct variable *v = v_variables[i];
617 struct var_freqs *vf = get_var_freqs (v);
618 struct freq_tab *ft = &vf->tab;
620 int dumped_freq_tab = 1;
622 postprocess_freq_tab (v);
624 /* Frequencies tables. */
625 n_categories = ft->n_valid + ft->n_missing;
626 if (cmd.table == FRQ_TABLE
627 || (cmd.table == FRQ_LIMIT && n_categories <= cmd.limit))
637 if (n_categories > cmd.onepage_limit)
650 dump_statistics (v, !dumped_freq_tab);
654 if ( chart == GFT_HIST)
656 double d[frq_n_stats];
657 struct normal_curve norm;
658 gsl_histogram *hist ;
661 norm.N = vf->tab.valid_cases;
664 norm.mean = d[frq_mean];
665 norm.stddev = d[frq_stddev];
667 hist = freq_tab_to_hist(ft,v);
669 histogram_plot(hist, var_to_string(v), &norm, normal);
671 gsl_histogram_free(hist);
675 if ( chart == GFT_PIE)
677 do_piechart(v_variables[i], ft);
682 cleanup_freq_tab (v);
687 /* Returns the comparison function that should be used for
688 sorting a frequency table by FRQ_SORT using VAR_TYPE
690 static hsh_compare_func *
691 get_freq_comparator (int frq_sort, enum var_type var_type)
693 bool is_numeric = var_type == VAR_NUMERIC;
697 return is_numeric ? compare_value_numeric_a : compare_value_alpha_a;
699 return is_numeric ? compare_value_numeric_d : compare_value_alpha_d;
701 return is_numeric ? compare_freq_numeric_a : compare_freq_alpha_a;
703 return is_numeric ? compare_freq_numeric_d : compare_freq_alpha_d;
709 /* Returns true iff the value in struct freq F is non-missing
712 not_missing (const void *f_, const void *v_)
714 const struct freq *f = f_;
715 const struct variable *v = v_;
717 return !var_is_value_missing (v, f->value, MV_ANY);
720 /* Summarizes the frequency table data for variable V. */
722 postprocess_freq_tab (const struct variable *v)
724 hsh_compare_func *compare;
728 struct freq *freqs, *f;
731 ft = &get_var_freqs (v)->tab;
732 assert (ft->mode == FRQM_GENERAL);
733 compare = get_freq_comparator (cmd.sort, var_get_type (v));
735 /* Extract data from hash table. */
736 count = hsh_count (ft->data);
737 data = hsh_data (ft->data);
739 /* Copy dereferenced data into freqs. */
740 freqs = xnmalloc (count, sizeof *freqs);
741 for (i = 0; i < count; i++)
743 struct freq *f = data[i];
747 /* Put data into ft. */
749 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, v);
750 ft->missing = freqs + ft->n_valid;
751 ft->n_missing = count - ft->n_valid;
754 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare, v);
755 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, v);
757 /* Summary statistics. */
758 ft->valid_cases = 0.0;
759 for(i = 0 ; i < ft->n_valid ; ++i )
762 ft->valid_cases += f->count;
766 ft->total_cases = ft->valid_cases ;
767 for(i = 0 ; i < ft->n_missing ; ++i )
770 ft->total_cases += f->count;
775 /* Frees the frequency table for variable V. */
777 cleanup_freq_tab (const struct variable *v)
779 struct freq_tab *ft = &get_var_freqs (v)->tab;
780 assert (ft->mode == FRQM_GENERAL);
782 hsh_destroy (ft->data);
785 /* Parses the VARIABLES subcommand, adding to
786 {n_variables,v_variables}. */
788 frq_custom_variables (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *aux UNUSED)
791 int min = 0, max = 0;
793 size_t old_n_variables = n_variables;
796 lex_match (lexer, '=');
797 if (lex_token (lexer) != T_ALL && (lex_token (lexer) != T_ID
798 || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) == NULL))
801 if (!parse_variables_const (lexer, dataset_dict (ds), &v_variables, &n_variables,
802 PV_APPEND | PV_NO_SCRATCH))
805 if (!lex_match (lexer, '('))
810 if (!lex_force_int (lexer))
812 min = lex_integer (lexer);
814 if (!lex_force_match (lexer, ','))
816 if (!lex_force_int (lexer))
818 max = lex_integer (lexer);
820 if (!lex_force_match (lexer, ')'))
824 msg (SE, _("Upper limit of integer mode value range must be "
825 "greater than lower limit."));
830 for (i = old_n_variables; i < n_variables; i++)
832 const struct variable *v = v_variables[i];
833 struct var_freqs *vf;
835 if (var_get_aux (v) != NULL)
837 msg (SE, _("Variable %s specified multiple times on VARIABLES "
838 "subcommand."), var_get_name (v));
841 if (mode == FRQM_INTEGER && !var_is_numeric (v))
843 msg (SE, _("Integer mode specified, but %s is not a numeric "
844 "variable."), var_get_name (v));
848 vf = var_attach_aux (v, xmalloc (sizeof *vf), var_dtor_free);
850 vf->tab.valid = vf->tab.missing = NULL;
851 if (mode == FRQM_INTEGER)
855 vf->tab.vector = pool_nalloc (int_pool,
856 max - min + 1, sizeof *vf->tab.vector);
859 vf->tab.vector = NULL;
862 vf->width = var_get_width (v);
863 vf->print = *var_get_print_format (v);
864 if (vf->width > MAX_SHORT_STRING && get_algorithm () == COMPATIBLE)
866 enum fmt_type type = var_get_print_format (v)->type;
867 vf->width = MAX_SHORT_STRING;
868 vf->print.w = MAX_SHORT_STRING * (type == FMT_AHEX ? 2 : 1);
874 /* Parses the GROUPED subcommand, setting the n_grouped, grouped
875 fields of specified variables. */
877 frq_custom_grouped (struct lexer *lexer, struct dataset *ds, struct cmd_frequencies *cmd UNUSED, void *aux UNUSED)
879 lex_match (lexer, '=');
880 if ((lex_token (lexer) == T_ID && dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
881 || lex_token (lexer) == T_ID)
886 /* Max, current size of list; list itself. */
892 const struct variable **v;
894 if (!parse_variables_const (lexer, dataset_dict (ds), &v, &n,
895 PV_NO_DUPLICATE | PV_NUMERIC))
897 if (lex_match (lexer, '('))
901 while (lex_integer (lexer))
906 dl = pool_nrealloc (int_pool, dl, ml, sizeof *dl);
908 dl[nl++] = lex_tokval (lexer);
910 lex_match (lexer, ',');
912 /* Note that nl might still be 0 and dl might still be
913 NULL. That's okay. */
914 if (!lex_match (lexer, ')'))
917 msg (SE, _("`)' expected after GROUPED interval list."));
927 for (i = 0; i < n; i++)
928 if (var_get_aux (v[i]) == NULL)
929 msg (SE, _("Variables %s specified on GROUPED but not on "
930 "VARIABLES."), var_get_name (v[i]));
933 struct var_freqs *vf = get_var_freqs (v[i]);
935 if (vf->groups != NULL)
936 msg (SE, _("Variables %s specified multiple times on GROUPED "
937 "subcommand."), var_get_name (v[i]));
945 if (!lex_match (lexer, '/'))
947 if ((lex_token (lexer) != T_ID || dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) != NULL)
948 && lex_token (lexer) != T_ALL)
950 lex_put_back (lexer, '/');
958 /* Adds X to the list of percentiles, keeping the list in proper
961 add_percentile (double x)
965 for (i = 0; i < n_percentiles; i++)
967 /* Do nothing if it's already in the list */
968 if ( fabs(x - percentiles[i].p) < DBL_EPSILON )
971 if (x < percentiles[i].p)
975 if (i >= n_percentiles || x != percentiles[i].p)
977 percentiles = pool_nrealloc (int_pool, percentiles,
978 n_percentiles + 1, sizeof *percentiles);
980 if (i < n_percentiles)
981 memmove (&percentiles[i + 1], &percentiles[i],
982 (n_percentiles - i) * sizeof (struct percentile) );
984 percentiles[i].p = x;
989 /* Comparison functions. */
991 /* Ascending numeric compare of values. */
993 compare_value_numeric_a (const void *a_, const void *b_, const void *aux UNUSED)
995 const struct freq *a = a_;
996 const struct freq *b = b_;
998 if (a->value[0].f > b->value[0].f)
1000 else if (a->value[0].f < b->value[0].f)
1006 /* Ascending string compare of values. */
1008 compare_value_alpha_a (const void *a_, const void *b_, const void *v_)
1010 const struct freq *a = a_;
1011 const struct freq *b = b_;
1012 const struct variable *v = v_;
1013 struct var_freqs *vf = get_var_freqs (v);
1015 return memcmp (a->value[0].s, b->value[0].s, vf->width);
1018 /* Descending numeric compare of values. */
1020 compare_value_numeric_d (const void *a, const void *b, const void *aux UNUSED)
1022 return -compare_value_numeric_a (a, b, aux);
1025 /* Descending string compare of values. */
1027 compare_value_alpha_d (const void *a, const void *b, const void *v)
1029 return -compare_value_alpha_a (a, b, v);
1032 /* Ascending numeric compare of frequency;
1033 secondary key on ascending numeric value. */
1035 compare_freq_numeric_a (const void *a_, const void *b_, const void *aux UNUSED)
1037 const struct freq *a = a_;
1038 const struct freq *b = b_;
1040 if (a->count > b->count)
1042 else if (a->count < b->count)
1045 if (a->value[0].f > b->value[0].f)
1047 else if (a->value[0].f < b->value[0].f)
1053 /* Ascending numeric compare of frequency;
1054 secondary key on ascending string value. */
1056 compare_freq_alpha_a (const void *a_, const void *b_, const void *v_)
1058 const struct freq *a = a_;
1059 const struct freq *b = b_;
1060 const struct variable *v = v_;
1061 struct var_freqs *vf = get_var_freqs (v);
1063 if (a->count > b->count)
1065 else if (a->count < b->count)
1068 return memcmp (a->value[0].s, b->value[0].s, vf->width);
1071 /* Descending numeric compare of frequency;
1072 secondary key on ascending numeric value. */
1074 compare_freq_numeric_d (const void *a_, const void *b_, const void *aux UNUSED)
1076 const struct freq *a = a_;
1077 const struct freq *b = b_;
1079 if (a->count > b->count)
1081 else if (a->count < b->count)
1084 if (a->value[0].f > b->value[0].f)
1086 else if (a->value[0].f < b->value[0].f)
1092 /* Descending numeric compare of frequency;
1093 secondary key on ascending string value. */
1095 compare_freq_alpha_d (const void *a_, const void *b_, const void *v_)
1097 const struct freq *a = a_;
1098 const struct freq *b = b_;
1099 const struct variable *v = v_;
1100 struct var_freqs *vf = get_var_freqs (v);
1102 if (a->count > b->count)
1104 else if (a->count < b->count)
1107 return memcmp (a->value[0].s, b->value[0].s, vf->width);
1110 /* Frequency table display. */
1112 /* Sets the widths of all the columns and heights of all the rows in
1113 table T for driver D. */
1115 full_dim (struct tab_table *t, struct outp_driver *d)
1120 if (cmd.labels == FRQ_LABELS)
1122 t->w[0] = MIN (tab_natural_width (t, d, 0), d->prop_em_width * 15);
1127 for (;i < columns; i++)
1128 t->w[i] = MAX (tab_natural_width (t, d, i), d->prop_em_width * 8);
1130 for (i = 0; i < t->nr; i++)
1131 t->h[i] = d->font_height;
1134 /* Displays a full frequency table for variable V. */
1136 dump_full (const struct variable *v)
1139 struct var_freqs *vf;
1140 struct freq_tab *ft;
1142 struct tab_table *t;
1144 double cum_total = 0.0;
1145 double cum_freq = 0.0;
1153 const struct init *p;
1155 static const struct init vec[] =
1157 {4, 0, N_("Valid")},
1159 {1, 1, N_("Value")},
1160 {2, 1, N_("Frequency")},
1161 {3, 1, N_("Percent")},
1162 {4, 1, N_("Percent")},
1163 {5, 1, N_("Percent")},
1171 const bool lab = (cmd.labels == FRQ_LABELS);
1173 vf = get_var_freqs (v);
1175 n_categories = ft->n_valid + ft->n_missing;
1176 t = tab_create (5 + lab, n_categories + 3, 0);
1177 tab_headers (t, 0, 0, 2, 0);
1178 tab_dim (t, full_dim);
1181 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value Label"));
1183 for (p = vec; p->s; p++)
1184 tab_text (t, lab ? p->c : p->c - 1, p->r,
1185 TAB_CENTER | TAT_TITLE, gettext (p->s));
1188 for (f = ft->valid; f < ft->missing; f++)
1190 double percent, valid_percent;
1192 cum_freq += f->count;
1194 percent = f->count / ft->total_cases * 100.0;
1195 valid_percent = f->count / ft->valid_cases * 100.0;
1196 cum_total += valid_percent;
1200 const char *label = var_lookup_value_label (v, &f->value[0]);
1202 tab_text (t, 0, r, TAB_LEFT, label);
1205 tab_value (t, 0 + lab, r, TAB_NONE, f->value, &vf->print);
1206 tab_float (t, 1 + lab, r, TAB_NONE, f->count, 8, 0);
1207 tab_float (t, 2 + lab, r, TAB_NONE, percent, 5, 1);
1208 tab_float (t, 3 + lab, r, TAB_NONE, valid_percent, 5, 1);
1209 tab_float (t, 4 + lab, r, TAB_NONE, cum_total, 5, 1);
1212 for (; f < &ft->valid[n_categories]; f++)
1214 cum_freq += f->count;
1218 const char *label = var_lookup_value_label (v, &f->value[0]);
1220 tab_text (t, 0, r, TAB_LEFT, label);
1223 tab_value (t, 0 + lab, r, TAB_NONE, f->value, &vf->print);
1224 tab_float (t, 1 + lab, r, TAB_NONE, f->count, 8, 0);
1225 tab_float (t, 2 + lab, r, TAB_NONE,
1226 f->count / ft->total_cases * 100.0, 5, 1);
1227 tab_text (t, 3 + lab, r, TAB_NONE, _("Missing"));
1231 tab_box (t, TAL_1, TAL_1,
1232 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1234 tab_hline (t, TAL_2, 0, 4 + lab, 2);
1235 tab_hline (t, TAL_2, 0, 4 + lab, r);
1236 tab_joint_text (t, 0, r, 0 + lab, r, TAB_RIGHT | TAT_TITLE, _("Total"));
1237 tab_vline (t, TAL_0, 1, r, r);
1238 tab_float (t, 1 + lab, r, TAB_NONE, cum_freq, 8, 0);
1239 tab_float (t, 2 + lab, r, TAB_NONE, 100.0, 5, 1);
1240 tab_float (t, 3 + lab, r, TAB_NONE, 100.0, 5, 1);
1242 tab_title (t, "%s", var_to_string (v));
1246 /* Sets the widths of all the columns and heights of all the rows in
1247 table T for driver D. */
1249 condensed_dim (struct tab_table *t, struct outp_driver *d)
1251 int cum_w = MAX (outp_string_width (d, _("Cum"), OUTP_PROPORTIONAL),
1252 MAX (outp_string_width (d, _("Cum"), OUTP_PROPORTIONAL),
1253 outp_string_width (d, "000", OUTP_PROPORTIONAL)));
1257 for (i = 0; i < 2; i++)
1258 t->w[i] = MAX (tab_natural_width (t, d, i), d->prop_em_width * 8);
1259 for (i = 2; i < 4; i++)
1261 for (i = 0; i < t->nr; i++)
1262 t->h[i] = d->font_height;
1265 /* Display condensed frequency table for variable V. */
1267 dump_condensed (const struct variable *v)
1270 struct var_freqs *vf;
1271 struct freq_tab *ft;
1273 struct tab_table *t;
1275 double cum_total = 0.0;
1277 vf = get_var_freqs (v);
1279 n_categories = ft->n_valid + ft->n_missing;
1280 t = tab_create (4, n_categories + 2, 0);
1282 tab_headers (t, 0, 0, 2, 0);
1283 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value"));
1284 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Freq"));
1285 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1286 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Cum"));
1287 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1288 tab_dim (t, condensed_dim);
1291 for (f = ft->valid; f < ft->missing; f++)
1295 percent = f->count / ft->total_cases * 100.0;
1296 cum_total += f->count / ft->valid_cases * 100.0;
1298 tab_value (t, 0, r, TAB_NONE, f->value, &vf->print);
1299 tab_float (t, 1, r, TAB_NONE, f->count, 8, 0);
1300 tab_float (t, 2, r, TAB_NONE, percent, 3, 0);
1301 tab_float (t, 3, r, TAB_NONE, cum_total, 3, 0);
1304 for (; f < &ft->valid[n_categories]; f++)
1306 tab_value (t, 0, r, TAB_NONE, f->value, &vf->print);
1307 tab_float (t, 1, r, TAB_NONE, f->count, 8, 0);
1308 tab_float (t, 2, r, TAB_NONE,
1309 f->count / ft->total_cases * 100.0, 3, 0);
1313 tab_box (t, TAL_1, TAL_1,
1314 cmd.spaces == FRQ_SINGLE ? -1 : TAL_GAP, TAL_1,
1316 tab_hline (t, TAL_2, 0, 3, 2);
1317 tab_title (t, "%s", var_to_string (v));
1318 tab_columns (t, SOM_COL_DOWN, 1);
1322 /* Statistical display. */
1324 /* Calculates all the pertinent statistics for variable V, putting
1325 them in array D[]. FIXME: This could be made much more optimal. */
1327 calc_stats (const struct variable *v, double d[frq_n_stats])
1329 struct freq_tab *ft = &get_var_freqs (v)->tab;
1330 double W = ft->valid_cases;
1339 double *median_value;
1341 /* Calculate percentiles. */
1343 /* If the 50th percentile was not explicitly requested then we must
1344 calculate it anyway --- it's the median */
1346 for (i = 0; i < n_percentiles; i++)
1348 if (percentiles[i].p == 0.5)
1350 median_value = &percentiles[i].value;
1355 if ( 0 == median_value )
1357 add_percentile (0.5);
1361 for (i = 0; i < n_percentiles; i++)
1363 percentiles[i].flag = 0;
1364 percentiles[i].flag2 = 0;
1368 for (idx = 0; idx < ft->n_valid; ++idx)
1370 static double prev_value = SYSMIS;
1371 f = &ft->valid[idx];
1373 for (i = 0; i < n_percentiles; i++)
1376 if ( percentiles[i].flag2 ) continue ;
1378 if ( get_algorithm() != COMPATIBLE )
1380 (ft->valid_cases - 1) * percentiles[i].p;
1383 (ft->valid_cases + 1) * percentiles[i].p - 1;
1385 if ( percentiles[i].flag )
1387 percentiles[i].x2 = f->value[0].f;
1388 percentiles[i].x1 = prev_value;
1389 percentiles[i].flag2 = 1;
1395 if ( f->count > 1 && rank - (f->count - 1) > tp )
1397 percentiles[i].x2 = percentiles[i].x1 = f->value[0].f;
1398 percentiles[i].flag2 = 1;
1402 percentiles[i].flag=1;
1408 prev_value = f->value[0].f;
1411 for (i = 0; i < n_percentiles; i++)
1413 /* Catches the case when p == 100% */
1414 if ( ! percentiles[i].flag2 )
1415 percentiles[i].x1 = percentiles[i].x2 = f->value[0].f;
1418 printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
1419 i,percentiles[i].p,percentiles[i].x1,percentiles[i].x2);
1423 for (i = 0; i < n_percentiles; i++)
1425 struct freq_tab *ft = &get_var_freqs (v)->tab;
1429 if ( get_algorithm() != COMPATIBLE )
1431 s = modf((ft->valid_cases - 1) * percentiles[i].p , &dummy);
1435 s = modf((ft->valid_cases + 1) * percentiles[i].p -1, &dummy);
1438 percentiles[i].value = percentiles[i].x1 +
1439 ( percentiles[i].x2 - percentiles[i].x1) * s ;
1441 if ( percentiles[i].p == 0.50)
1442 median_value = &percentiles[i].value;
1446 /* Calculate the mode. */
1449 for (f = ft->valid; f < ft->missing; f++)
1451 if (most_often < f->count)
1453 most_often = f->count;
1454 X_mode = f->value[0].f;
1456 else if (most_often == f->count)
1458 /* A duplicate mode is undefined.
1459 FIXME: keep track of *all* the modes. */
1464 /* Calculate moments. */
1465 m = moments_create (MOMENT_KURTOSIS);
1466 for (f = ft->valid; f < ft->missing; f++)
1467 moments_pass_one (m, f->value[0].f, f->count);
1468 for (f = ft->valid; f < ft->missing; f++)
1469 moments_pass_two (m, f->value[0].f, f->count);
1470 moments_calculate (m, NULL, &d[frq_mean], &d[frq_variance],
1471 &d[frq_skew], &d[frq_kurt]);
1472 moments_destroy (m);
1474 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1475 d[frq_min] = ft->valid[0].value[0].f;
1476 d[frq_max] = ft->valid[ft->n_valid - 1].value[0].f;
1477 d[frq_mode] = X_mode;
1478 d[frq_range] = d[frq_max] - d[frq_min];
1479 d[frq_median] = *median_value;
1480 d[frq_sum] = d[frq_mean] * W;
1481 d[frq_stddev] = sqrt (d[frq_variance]);
1482 d[frq_semean] = d[frq_stddev] / sqrt (W);
1483 d[frq_seskew] = calc_seskew (W);
1484 d[frq_sekurt] = calc_sekurt (W);
1487 /* Displays a table of all the statistics requested for variable V. */
1489 dump_statistics (const struct variable *v, int show_varname)
1491 struct freq_tab *ft;
1492 double stat_value[frq_n_stats];
1493 struct tab_table *t;
1496 int n_explicit_percentiles = n_percentiles;
1498 if ( implicit_50th && n_percentiles > 0 )
1501 if (var_is_alpha (v))
1503 ft = &get_var_freqs (v)->tab;
1504 if (ft->n_valid == 0)
1506 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1510 calc_stats (v, stat_value);
1512 t = tab_create (3, n_stats + n_explicit_percentiles + 2, 0);
1513 tab_dim (t, tab_natural_dimensions);
1515 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1518 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1519 tab_vline (t, TAL_GAP , 1, 0, tab_nr(t) - 1 ) ;
1521 r=2; /* N missing and N valid are always dumped */
1523 for (i = 0; i < frq_n_stats; i++)
1524 if (stats & BIT_INDEX (i))
1526 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1527 gettext (st_name[i].s10));
1528 tab_float (t, 2, r, TAB_NONE, stat_value[i], 11, 3);
1532 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1533 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1534 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1536 tab_float(t, 2, 0, TAB_NONE, ft->valid_cases, 11, 0);
1537 tab_float(t, 2, 1, TAB_NONE, ft->total_cases - ft->valid_cases, 11, 0);
1540 for (i = 0; i < n_explicit_percentiles; i++, r++)
1544 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1547 tab_float (t, 1, r, TAB_LEFT, percentiles[i].p * 100, 3, 0 );
1548 tab_float (t, 2, r, TAB_NONE, percentiles[i].value, 11, 3);
1552 tab_columns (t, SOM_COL_DOWN, 1);
1554 tab_title (t, "%s", var_to_string (v));
1556 tab_flags (t, SOMF_NO_TITLE);
1563 /* Create a gsl_histogram from a freq_tab */
1565 freq_tab_to_hist(const struct freq_tab *ft, const struct variable *var)
1568 double x_min = DBL_MAX;
1569 double x_max = -DBL_MAX;
1571 gsl_histogram *hist;
1572 const double bins = 11;
1574 struct hsh_iterator hi;
1575 struct hsh_table *fh = ft->data;
1578 /* Find out the extremes of the x value */
1579 for ( frq = hsh_first(fh, &hi); frq != 0; frq = hsh_next(fh, &hi) )
1581 if (var_is_value_missing(var, frq->value, MV_ANY))
1584 if ( frq->value[0].f < x_min ) x_min = frq->value[0].f ;
1585 if ( frq->value[0].f > x_max ) x_max = frq->value[0].f ;
1588 hist = histogram_create(bins, x_min, x_max);
1590 for( i = 0 ; i < ft->n_valid ; ++i )
1592 frq = &ft->valid[i];
1593 gsl_histogram_accumulate(hist, frq->value[0].f, frq->count);
1600 static struct slice *
1601 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1602 const struct variable *var,
1606 /* Allocate an array of slices and fill them from the data in frq_tab
1607 n_slices will contain the number of slices allocated.
1608 The caller is responsible for freeing slices
1610 static struct slice *
1611 freq_tab_to_slice_array(const struct freq_tab *frq_tab,
1612 const struct variable *var,
1616 struct slice *slices;
1618 *n_slices = frq_tab->n_valid;
1620 slices = xnmalloc (*n_slices, sizeof *slices);
1622 for (i = 0 ; i < *n_slices ; ++i )
1624 const struct freq *frq = &frq_tab->valid[i];
1626 slices[i].label = var_get_value_name (var, frq->value);
1627 slices[i].magnetude = frq->count;
1637 do_piechart(const struct variable *var, const struct freq_tab *frq_tab)
1639 struct slice *slices;
1642 slices = freq_tab_to_slice_array(frq_tab, var, &n_slices);
1644 piechart_plot(var_to_string(var), slices, n_slices);