*/
#include <config.h>
-#include <assert.h>
+#include "error.h"
#include <math.h>
#include <stdlib.h>
#include "alloc.h"
#include "bitvector.h"
+#include "case.h"
#include "hash.h"
#include "pool.h"
#include "command.h"
#include "lexer.h"
+#include "moments.h"
#include "error.h"
#include "algorithm.h"
#include "magic.h"
#include "misc.h"
-#include "stats.h"
#include "output.h"
#include "som.h"
#include "str.h"
#include "value-labels.h"
#include "var.h"
#include "vfm.h"
+#include "settings.h"
#include "debug-print.h"
};
/* Percentiles to calculate. */
-static double *percentiles;
-static double *percentile_values;
+
+struct percentile
+{
+ double p; /* the %ile to be calculated */
+ double value; /* the %ile's value */
+ double x1; /* The datum value <= the percentile */
+ double x2; /* The datum value >= the percentile */
+ int flag;
+ int flag2; /* Set to 1 if this percentile value has been found */
+};
+
+static struct percentile *percentiles;
static int n_percentiles;
+static int implicit_50th ;
+
/* Groups of statistics. */
#define BI BIT_INDEX
#define frq_default \
static void determine_charts (void);
-static void precalc (void);
-static int calc (struct ccase *);
-static void postcalc (void);
+static void precalc (void *);
+static int calc (struct ccase *, void *);
+static void postcalc (void *);
static void postprocess_freq_tab (struct variable *);
static void dump_full (struct variable *);
int i;
n_percentiles = 0;
- percentile_values = NULL;
percentiles = NULL;
n_variables = 0;
for (i = 0; i < dict_get_var_cnt (default_dict); i++)
dict_get_var(default_dict, i)->p.frq.used = 0;
- lex_match_id ("FREQUENCIES");
if (!parse_frequencies (&cmd))
return CMD_FAILURE;
cmd.sort = FRQ_AVALUE;
/* Do it! */
- procedure (precalc, calc, postcalc);
+ procedure_with_splits (precalc, calc, postcalc, NULL);
return CMD_SUCCESS;
}
/* Add data from case C to the frequency table. */
static int
-calc (struct ccase *c)
+calc (struct ccase *c, void *aux UNUSED)
{
double weight;
int i;
+ int bad_warn = 1;
- weight = dict_get_case_weight (default_dict, c);
+ weight = dict_get_case_weight (default_dict, c, &bad_warn);
for (i = 0; i < n_variables; i++)
{
struct variable *v = v_variables[i];
- union value *val = &c->data[v->fv];
+ const union value *val = case_data (c, v->fv);
struct freq_tab *ft = &v->p.frq.tab;
switch (v->p.frq.tab.mode)
/* Prepares each variable that is the target of FREQUENCIES by setting
up its hash table. */
static void
-precalc (void)
+precalc (void *aux UNUSED)
{
int i;
/* Finishes up with the variables after frequencies have been
calculated. Displays statistics, percentiles, ... */
static void
-postcalc (void)
+postcalc (void *aux UNUSED)
{
int i;
case FRQ_DFREQ | (ALPHA << 16): return compare_freq_alpha_d;
default: assert (0);
}
+
+ return 0;
}
+/* Returns nonzero iff the value in struct freq F is non-missing
+ for variable V. */
static int
not_missing (const void *f_, void *v_)
{
return !is_missing (&f->v, v);
}
+/* Summarizes the frequency table data for variable V. */
static void
-postprocess_freq_tab (struct variable * v)
+postprocess_freq_tab (struct variable *v)
{
hsh_compare_func *compare;
struct freq_tab *ft;
data = hsh_data (ft->data);
/* Copy dereferenced data into freqs. */
- freqs = xmalloc (count* sizeof *freqs);
+ freqs = xmalloc (count * sizeof *freqs);
for (i = 0; i < count; i++)
{
struct freq *f = data[i];
sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, v);
/* Summary statistics. */
- ft->total_cases = ft->valid_cases = 0.0;
- for (f = ft->valid; f < ft->valid + ft->n_valid; f++)
+ ft->valid_cases = 0.0;
+ for(i = 0 ; i < ft->n_valid ; ++i )
{
- ft->total_cases += f->c;
+ f = &ft->valid[i];
+ ft->valid_cases += f->c;
+
+ }
- if ((v->type != NUMERIC || f->v.f != SYSMIS)
- && (cmd.miss != FRQ_EXCLUDE || !is_user_missing (&f->v, v)))
- ft->valid_cases += f->c;
+ ft->total_cases = ft->valid_cases ;
+ for(i = 0 ; i < ft->n_missing ; ++i )
+ {
+ f = &ft->missing[i];
+ ft->total_cases += f->c;
}
+
}
+/* Frees the frequency table for variable V. */
static void
cleanup_freq_tab (struct variable *v)
{
assert (v->p.frq.tab.mode == FRQM_GENERAL);
free (v->p.frq.tab.valid);
+ hsh_destroy (v->p.frq.tab.data);
}
/* Parses the VARIABLES subcommand, adding to
{n_variables,v_variables}. */
static int
-frq_custom_variables (struct cmd_frequencies *cmd unused)
+frq_custom_variables (struct cmd_frequencies *cmd UNUSED)
{
int mode;
- int min, max;
+ int min = 0, max = 0;
int old_n_variables = n_variables;
int i;
/* Parses the GROUPED subcommand, setting the frq.{n_grouped,grouped}
fields of specified variables. */
static int
-frq_custom_grouped (struct cmd_frequencies *cmd unused)
+frq_custom_grouped (struct cmd_frequencies *cmd UNUSED)
{
lex_match ('=');
if ((token == T_ID && dict_lookup_var (default_dict, tokid) != NULL)
return 0;
}
}
- else
- nl = 0;
+ else
+ {
+ nl = 0;
+ dl = NULL;
+ }
for (i = 0; i < n; i++)
{
int i;
for (i = 0; i < n_percentiles; i++)
- if (x <= percentiles[i])
+ if (x <= percentiles[i].p)
break;
- if (i >= n_percentiles || tokval != percentiles[i])
+
+ if (i >= n_percentiles || tokval != percentiles[i].p)
{
percentiles
= pool_realloc (int_pool, percentiles,
- (n_percentiles + 1) * sizeof *percentiles);
- percentile_values
- = pool_realloc (int_pool, percentile_values,
- (n_percentiles + 1) * sizeof *percentile_values);
- if (i < n_percentiles)
- {
+ (n_percentiles + 1) * sizeof (struct percentile ));
+
+ if (i < n_percentiles)
memmove (&percentiles[i + 1], &percentiles[i],
- (n_percentiles - i) * sizeof *percentiles);
- memmove (&percentile_values[i + 1], &percentile_values[i],
- (n_percentiles - i) * sizeof *percentile_values);
- }
- percentiles[i] = x;
+ (n_percentiles - i) * sizeof (struct percentile) );
+
+ percentiles[i].p = x;
n_percentiles++;
}
}
/* Parses the PERCENTILES subcommand, adding user-specified
percentiles to the list. */
static int
-frq_custom_percentiles (struct cmd_frequencies *cmd unused)
+frq_custom_percentiles (struct cmd_frequencies *cmd UNUSED)
{
lex_match ('=');
if (token != T_NUM)
do
{
- if (tokval <= 0 || tokval >= 100)
+ if (tokval < 0 || tokval > 100)
{
- msg (SE, _("Percentiles must be greater than "
- "0 and less than 100."));
+ msg (SE, _("Percentiles must be between 0 and 100."));
return 0;
}
/* Parses the NTILES subcommand, adding the percentiles that
correspond to the specified evenly-distributed ntiles. */
static int
-frq_custom_ntiles (struct cmd_frequencies *cmd unused)
+frq_custom_ntiles (struct cmd_frequencies *cmd UNUSED)
{
int i;
/* Hash of numeric values. */
static unsigned
-hash_value_numeric (const void *value_, void *foo unused)
+hash_value_numeric (const void *value_, void *foo UNUSED)
{
const struct freq *value = value_;
return hsh_hash_double (value->v.f);
/* Hash of string values. */
static unsigned
-hash_value_alpha (const void *value_, void *len_)
+hash_value_alpha (const void *value_, void *v_)
{
const struct freq *value = value_;
- int *len = len_;
+ struct variable *v = v_;
- return hsh_hash_bytes (value->v.s, *len);
+ return hsh_hash_bytes (value->v.s, v->width);
}
/* Ascending numeric compare of values. */
static int
-compare_value_numeric_a (const void *a_, const void *b_, void *foo unused)
+compare_value_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
{
const struct freq *a = a_;
const struct freq *b = b_;
/* Descending numeric compare of values. */
static int
-compare_value_numeric_d (const void *a, const void *b, void *foo unused)
+compare_value_numeric_d (const void *a, const void *b, void *foo UNUSED)
{
return -compare_value_numeric_a (a, b, foo);
}
/* Ascending numeric compare of frequency;
secondary key on ascending numeric value. */
static int
-compare_freq_numeric_a (const void *a_, const void *b_, void *foo unused)
+compare_freq_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
{
const struct freq *a = a_;
const struct freq *b = b_;
- if (a->v.c > b->v.c)
+ if (a->c > b->c)
return 1;
- else if (a->v.c < b->v.c)
+ else if (a->c < b->c)
return -1;
if (a->v.f > b->v.f)
const struct freq *b = b_;
const struct variable *v = v_;
- if (a->v.c > b->v.c)
+ if (a->c > b->c)
return 1;
- else if (a->v.c < b->v.c)
+ else if (a->c < b->c)
return -1;
else
return memcmp (a->v.s, b->v.s, v->width);
/* Descending numeric compare of frequency;
secondary key on ascending numeric value. */
static int
-compare_freq_numeric_d (const void *a_, const void *b_, void *foo unused)
+compare_freq_numeric_d (const void *a_, const void *b_, void *foo UNUSED)
{
const struct freq *a = a_;
const struct freq *b = b_;
- if (a->v.c > b->v.c)
+ if (a->c > b->c)
return -1;
- else if (a->v.c < b->v.c)
+ else if (a->c < b->c)
return 1;
if (a->v.f > b->v.f)
const struct freq *b = b_;
const struct variable *v = v_;
- if (a->v.c > b->v.c)
+ if (a->c > b->c)
return -1;
- else if (a->v.c < b->v.c)
+ else if (a->c < b->c)
return 1;
else
return memcmp (a->v.s, b->v.s, v->width);
struct freq *f;
struct tab_table *t;
int r;
- double cum_percent = 0.0;
+ double cum_total = 0.0;
double cum_freq = 0.0;
struct init
percent = f->c / v->p.frq.tab.total_cases * 100.0;
valid_percent = f->c / v->p.frq.tab.valid_cases * 100.0;
- cum_percent += valid_percent;
+ cum_total += valid_percent;
if (lab)
{
tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
tab_float (t, 2 + lab, r, TAB_NONE, percent, 5, 1);
tab_float (t, 3 + lab, r, TAB_NONE, valid_percent, 5, 1);
- tab_float (t, 4 + lab, r, TAB_NONE, cum_percent, 5, 1);
+ tab_float (t, 4 + lab, r, TAB_NONE, cum_total, 5, 1);
r++;
}
for (; f < &v->p.frq.tab.valid[n_categories]; f++)
struct freq *f;
struct tab_table *t;
int r;
- double cum_percent = 0.0;
+ double cum_total = 0.0;
n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
t = tab_create (4, n_categories + 2, 0);
double percent;
percent = f->c / v->p.frq.tab.total_cases * 100.0;
- cum_percent += f->c / v->p.frq.tab.valid_cases * 100.0;
+ cum_total += f->c / v->p.frq.tab.valid_cases * 100.0;
tab_value (t, 0, r, TAB_NONE, &f->v, &v->print);
tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
tab_float (t, 2, r, TAB_NONE, percent, 3, 0);
- tab_float (t, 3, r, TAB_NONE, cum_percent, 3, 0);
+ tab_float (t, 3, r, TAB_NONE, cum_total, 3, 0);
r++;
}
for (; f < &v->p.frq.tab.valid[n_categories]; f++)
calc_stats (struct variable * v, double d[frq_n_stats])
{
double W = v->p.frq.tab.valid_cases;
- double X_bar, X_mode, M2, M3, M4;
- struct freq *f;
+ struct moments *m;
+ struct freq *f=0;
int most_often;
+ double X_mode;
- double cum_percent;
+ double rank;
int i = 0;
- double previous_value;
-
- /* Calculate the mean. */
- X_bar = 0.0;
- for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
- X_bar += f->v.f * f->c;
- X_bar /= W;
+ int idx;
+ double *median_value;
/* Calculate percentiles. */
- cum_percent = 0;
- previous_value = SYSMIS;
- for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
+
+ /* If the 50th percentile was not explicitly requested then we must
+ calculate it anyway --- it's the median */
+ median_value = 0 ;
+ for (i = 0; i < n_percentiles; i++)
{
- cum_percent += f->c / v->p.frq.tab.valid_cases;
- for (; i < n_percentiles; i++)
+ if (percentiles[i].p == 0.5)
+ {
+ median_value = &percentiles[i].value;
+ break;
+ }
+ }
+
+ if ( 0 == median_value )
+ {
+ add_percentile (0.5);
+ implicit_50th = 1;
+ }
+
+ for (i = 0; i < n_percentiles; i++)
+ {
+ percentiles[i].flag = 0;
+ percentiles[i].flag2 = 0;
+ }
+
+ rank = 0;
+ for (idx = 0; idx < v->p.frq.tab.n_valid; ++idx)
+ {
+ static double prev_value = SYSMIS;
+ f = &v->p.frq.tab.valid[idx];
+ rank += f->c ;
+ for (i = 0; i < n_percentiles; i++)
{
- if (cum_percent <= percentiles[i])
- break;
+ double tp;
+ if ( percentiles[i].flag2 ) continue ;
+
+ if ( get_algorithm() != COMPATIBLE )
+ tp =
+ (v->p.frq.tab.valid_cases - 1) * percentiles[i].p;
+ else
+ tp =
+ (v->p.frq.tab.valid_cases + 1) * percentiles[i].p - 1;
+
+ if ( percentiles[i].flag )
+ {
+ percentiles[i].x2 = f->v.f;
+ percentiles[i].x1 = prev_value;
+ percentiles[i].flag2 = 1;
+ continue;
+ }
+
+ if (rank > tp )
+ {
+ if ( f->c > 1 && rank - (f->c - 1) > tp )
+ {
+ percentiles[i].x2 = percentiles[i].x1 = f->v.f;
+ percentiles[i].flag2 = 1;
+ }
+ else
+ {
+ percentiles[i].flag=1;
+ }
- percentile_values[i] = previous_value;
+ continue;
+ }
}
- previous_value = f->v.f;
+ prev_value = f->v.f;
+ }
+
+ for (i = 0; i < n_percentiles; i++)
+ {
+ /* Catches the case when p == 100% */
+ if ( ! percentiles[i].flag2 )
+ percentiles[i].x1 = percentiles[i].x2 = f->v.f;
+
+ /*
+ printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
+ i,percentiles[i].p,percentiles[i].x1,percentiles[i].x2);
+ */
+ }
+
+ for (i = 0; i < n_percentiles; i++)
+ {
+ struct freq_tab *ft = &v->p.frq.tab;
+ double s;
+
+ double dummy;
+ if ( get_algorithm() != COMPATIBLE )
+ {
+ s = modf((ft->valid_cases - 1) * percentiles[i].p , &dummy);
+ }
+ else
+ {
+ s = modf((ft->valid_cases + 1) * percentiles[i].p -1, &dummy);
+ }
+
+ percentiles[i].value = percentiles[i].x1 +
+ ( percentiles[i].x2 - percentiles[i].x1) * s ;
+
+ if ( percentiles[i].p == 0.50)
+ median_value = &percentiles[i].value;
}
+
/* Calculate the mode. */
most_often = -1;
X_mode = SYSMIS;
}
}
- /* Calculate moments about the mean. */
- M2 = M3 = M4 = 0.0;
+ /* Calculate moments. */
+ m = moments_create (MOMENT_KURTOSIS);
for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
- {
- double dev = f->v.f - X_bar;
- double tmp;
- tmp = dev * dev;
- M2 += f->c * tmp;
- tmp *= dev;
- M3 += f->c * tmp;
- tmp *= dev;
- M4 += f->c * tmp;
- }
-
+ moments_pass_one (m, f->v.f, f->c);
+ for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
+ moments_pass_two (m, f->v.f, f->c);
+ moments_calculate (m, NULL, &d[frq_mean], &d[frq_variance],
+ &d[frq_skew], &d[frq_kurt]);
+ moments_destroy (m);
+
/* Formulas below are taken from _SPSS Statistical Algorithms_. */
d[frq_min] = v->p.frq.tab.valid[0].v.f;
d[frq_max] = v->p.frq.tab.valid[v->p.frq.tab.n_valid - 1].v.f;
d[frq_mode] = X_mode;
d[frq_range] = d[frq_max] - d[frq_min];
- d[frq_median] = SYSMIS;
- d[frq_mean] = X_bar;
- d[frq_sum] = X_bar * W;
- d[frq_variance] = M2 / (W - 1);
+ d[frq_median] = *median_value;
+ d[frq_sum] = d[frq_mean] * W;
d[frq_stddev] = sqrt (d[frq_variance]);
d[frq_semean] = d[frq_stddev] / sqrt (W);
- if (W >= 3.0 && d[frq_variance] > 0)
- {
- double S = d[frq_stddev];
- d[frq_skew] = (W * M3 / ((W - 1.0) * (W - 2.0) * S * S * S));
- d[frq_seskew] = sqrt (6.0 * W * (W - 1.0)
- / ((W - 2.0) * (W + 1.0) * (W + 3.0)));
- }
- else
- {
- d[frq_skew] = d[frq_seskew] = SYSMIS;
- }
- if (W >= 4.0 && d[frq_variance] > 0)
- {
- double S2 = d[frq_variance];
- double SE_g1 = d[frq_seskew];
-
- d[frq_kurt] = ((W * (W + 1.0) * M4 - 3.0 * M2 * M2 * (W - 1.0))
- / ((W - 1.0) * (W - 2.0) * (W - 3.0) * S2 * S2));
- d[frq_sekurt] = sqrt ((4.0 * (W * W - 1.0) * SE_g1 * SE_g1)
- / ((W - 3.0) * (W + 5.0)));
- }
- else
- {
- d[frq_kurt] = d[frq_sekurt] = SYSMIS;
- }
+ d[frq_seskew] = calc_seskew (W);
+ d[frq_sekurt] = calc_sekurt (W);
}
/* Displays a table of all the statistics requested for variable V. */
struct tab_table *t;
int i, r;
+ int n_explicit_percentiles = n_percentiles;
+
+ if ( implicit_50th && n_percentiles > 0 )
+ --n_percentiles;
+
if (v->type == ALPHA)
return;
if (v->p.frq.tab.n_valid == 0)
}
calc_stats (v, stat_value);
- t = tab_create (2, n_stats + n_percentiles, 0);
+ t = tab_create (3, n_stats + n_explicit_percentiles + 2, 0);
tab_dim (t, tab_natural_dimensions);
- tab_vline (t, TAL_1 | TAL_SPACING, 1, 0, n_stats - 1);
- for (i = r = 0; i < frq_n_stats; i++)
+
+ tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
+
+
+ tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
+ tab_vline (t, TAL_1 | TAL_SPACING , 1, 0, tab_nr(t) - 1 ) ;
+
+ r=2; /* N missing and N valid are always dumped */
+
+ for (i = 0; i < frq_n_stats; i++)
if (stats & BIT_INDEX (i))
{
tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
gettext (st_name[i].s10));
- tab_float (t, 1, r, TAB_NONE, stat_value[i], 11, 3);
+ tab_float (t, 2, r, TAB_NONE, stat_value[i], 11, 3);
r++;
}
- for (i = 0; i < n_percentiles; i++, r++)
- {
- struct string ds;
+ tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
+ tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
+ tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
+
+ tab_float(t, 2, 0, TAB_NONE, v->p.frq.tab.valid_cases, 11, 0);
+ tab_float(t, 2, 1, TAB_NONE,
+ v->p.frq.tab.total_cases - v->p.frq.tab.valid_cases, 11, 0);
- ds_init (gen_pool, &ds, 20);
- ds_printf (&ds, "%s %d", _("Percentile"), (int) (percentiles[i] * 100));
- tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, ds.string);
- tab_float (t, 1, r, TAB_NONE, percentile_values[i], 11, 3);
+ for (i = 0; i < n_explicit_percentiles; i++, r++)
+ {
+ if ( i == 0 )
+ {
+ tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
+ }
+
+ tab_float (t, 1, r, TAB_LEFT, percentiles[i].p * 100, 3, 0 );
+ tab_float (t, 2, r, TAB_NONE, percentiles[i].value, 11, 3);
- ds_destroy (&ds);
}
tab_columns (t, SOM_COL_DOWN, 1);
}
else
tab_flags (t, SOMF_NO_TITLE);
-
+
+
tab_submit (t);
}
/*
projects / pspp-builds.git / blobdiff