#include "hash.h"
#include "algorithm.h"
#include "alloc.h"
+#include "moments.h"
+#include "percentiles.h"
#include <stdlib.h>
#include <math.h>
#include <float.h>
#include <assert.h>
-
+#include <chart.h>
void
-metrics_precalc(struct metrics *fs)
+metrics_precalc(struct metrics *m)
{
- fs->n = 0;
- fs->ssq = 0;
- fs->sum = 0;
- fs->min = DBL_MAX;
- fs->max = -DBL_MAX;
+ assert (m) ;
+
+ m->n_missing = 0;
+
+ m->min = DBL_MAX;
+ m->max = -DBL_MAX;
+
+
+ m->moments = moments1_create(MOMENT_KURTOSIS);
- fs->ordered_data = hsh_create(20,
+ m->ordered_data = hsh_create(20,
(hsh_compare_func *) compare_values,
(hsh_hash_func *) hash_value,
- 0,
+ (hsh_free_func *) weighted_value_free,
(void *) 0);
}
+
+/* Include val in the calculation for the metrics.
+ If val is null, then treat it as MISSING
+*/
void
-metrics_calc(struct metrics *fs, const union value *val, double weight)
+metrics_calc(struct metrics *fs, const union value *val,
+ double weight, int case_no)
{
-
-
struct weighted_value **wv;
- const double x = val->f;
+ double x;
- fs->n += weight;
- fs->ssq += x * x * weight;
- fs->sum += x * weight;
+ if ( ! val )
+ {
+ fs->n_missing += weight;
+ return ;
+ }
+
+ x = val->f;
+
+ moments1_add(fs->moments, x, weight);
+
if ( x < fs->min) fs->min = x;
if ( x > fs->max) fs->max = x;
if ( *wv )
{
/* If this value has already been seen, then simply
- increase its weight */
+ increase its weight and push a new case number */
+
+ struct case_node *cn;
assert( (*wv)->v.f == val->f );
(*wv)->w += weight;
+
+ cn = xmalloc( sizeof (struct case_node) ) ;
+ cn->next = (*wv)->case_nos ;
+ cn->num = case_no;
+
+ (*wv)->case_nos = cn;
}
else
{
- *wv = xmalloc( sizeof (struct weighted_value) );
+ struct case_node *cn;
+
+ *wv = weighted_value_create();
(*wv)->v = *val;
(*wv)->w = weight;
- hsh_insert(fs->ordered_data,(void *) *wv);
+
+ cn = xmalloc( sizeof (struct case_node) ) ;
+ cn->next=0;
+ cn->num = case_no;
+ (*wv)->case_nos = cn;
+
}
}
void
-metrics_postcalc(struct metrics *fs)
+metrics_postcalc(struct metrics *m)
{
- double sample_var;
double cc = 0.0;
double tc ;
int k1, k2 ;
int i;
int j = 1;
- struct weighted_value **data;
+ moments1_calculate (m->moments, &m->n, &m->mean, &m->var,
+ &m->skewness, &m->kurtosis);
- int n_data;
-
- fs->mean = fs->sum / fs->n;
+ moments1_destroy (m->moments);
- sample_var = ( fs->ssq / fs->n - fs->mean * fs->mean );
-
- fs->var = fs->n * sample_var / ( fs->n - 1) ;
- fs->stddev = sqrt(fs->var);
+ m->stddev = sqrt(m->var);
/* FIXME: Check this is correct ???
Shouldn't we use the sample variance ??? */
- fs->stderr = sqrt (fs->var / fs->n) ;
+ m->se_mean = sqrt (m->var / m->n) ;
- data = (struct weighted_value **) hsh_data(fs->ordered_data);
- n_data = hsh_count(fs->ordered_data);
- fs->wv = xmalloc ( sizeof (struct weighted_value) * n_data);
- for ( i = 0 ; i < n_data ; ++i )
- fs->wv[i] = *(data[i]);
+ m->wvp = (struct weighted_value **) hsh_sort(m->ordered_data);
+ m->n_data = hsh_count(m->ordered_data);
- sort (fs->wv, n_data, sizeof (struct weighted_value) ,
- (algo_compare_func *) compare_values, 0);
+ m->histogram = histogram_create(10, m->min, m->max);
+ for ( i = 0 ; i < m->n_data ; ++i )
+ {
+ struct weighted_value **wv = (m->wvp) ;
+ gsl_histogram_accumulate(m->histogram, wv[i]->v.f, wv[i]->w);
+ }
-
- tc = fs->n * 0.05 ;
+ /* Trimmed mean calculation */
+ if ( m->n_data <= 1 )
+ {
+ m->trimmed_mean = m->mean;
+ return;
+ }
+
+ tc = m->n * 0.05 ;
k1 = -1;
k2 = -1;
-
- for ( i = 0 ; i < n_data ; ++i )
+ for ( i = 0 ; i < m->n_data ; ++i )
{
- cc += fs->wv[i].w;
- fs->wv[i].cc = cc;
+ cc += m->wvp[i]->w;
+ m->wvp[i]->cc = cc;
- fs->wv[i].rank = j + (fs->wv[i].w - 1) / 2.0 ;
+ m->wvp[i]->rank = j + (m->wvp[i]->w - 1) / 2.0 ;
- j += fs->wv[i].w;
+ j += m->wvp[i]->w;
if ( cc < tc )
k1 = i;
-
}
- k2 = n_data;
- for ( i = n_data -1 ; i >= 0; --i )
+
+
+ k2 = m->n_data;
+ for ( i = m->n_data -1 ; i >= 0; --i )
{
- if ( tc > fs->n - fs->wv[i].cc)
+ if ( tc > m->n - m->wvp[i]->cc)
k2 = i;
}
- fs->trimmed_mean = 0;
+ /* Calculate the percentiles */
+ ptiles(m->ptile_hash, m->wvp, m->n_data, m->n, m->ptile_alg);
+
+ tukey_hinges(m->wvp, m->n_data, m->n, m->hinge);
+
+ /* Special case here */
+ if ( k1 + 1 == k2 )
+ {
+ m->trimmed_mean = m->wvp[k2]->v.f;
+ return;
+ }
+
+ m->trimmed_mean = 0;
for ( i = k1 + 2 ; i <= k2 - 1 ; ++i )
{
- fs->trimmed_mean += fs->wv[i].v.f * fs->wv[i].w;
+ m->trimmed_mean += m->wvp[i]->v.f * m->wvp[i]->w;
}
- fs->trimmed_mean += (fs->n - fs->wv[k2 - 1].cc - tc) * fs->wv[k2].v.f ;
- fs->trimmed_mean += (fs->wv[k1 + 1].cc - tc) * fs->wv[k1 + 1].v.f ;
- fs->trimmed_mean /= 0.9 * fs->n ;
+ m->trimmed_mean += (m->n - m->wvp[k2 - 1]->cc - tc) * m->wvp[k2]->v.f ;
+ m->trimmed_mean += (m->wvp[k1 + 1]->cc - tc) * m->wvp[k1 + 1]->v.f ;
+ m->trimmed_mean /= 0.9 * m->n ;
}
-/* Functions for hashes */
+struct weighted_value *
+weighted_value_create(void)
+{
+ struct weighted_value *wv;
+ wv = xmalloc (sizeof (struct weighted_value ));
+
+ wv->cc = 0;
+ wv->case_nos = 0;
+
+ return wv;
+}
void
-free_factor_stats(struct factor_statistics *f, int width UNUSED)
+weighted_value_free(struct weighted_value *wv)
{
- free (f);
+ struct case_node *cn = wv->case_nos;
+
+ while(cn)
+ {
+ struct case_node *next = cn->next;
+
+ free(cn);
+ cn = next;
+ }
+
+ free(wv);
+
}
-int
-compare_indep_values(const struct factor_statistics *f1,
- const struct factor_statistics *f2,
- int width)
+
+
+
+
+/* Create a factor statistics object with for N dependent vars
+ and ID as the value of the independent variable */
+struct factor_statistics *
+create_factor_statistics (int n, union value *id0, union value *id1)
{
- return compare_values(f1->id, f2->id, width);
+ struct factor_statistics *f;
+
+ f = xmalloc( sizeof ( struct factor_statistics ));
+
+ f->id[0] = *id0;
+ f->id[1] = *id1;
+ f->m = xmalloc( sizeof ( struct metrics ) * n ) ;
+
+ return f;
}
-unsigned
-hash_indep_value(const struct factor_statistics *f, int width)
+void
+factor_statistics_free(struct factor_statistics *f)
{
- return hash_value(f->id, width);
+ hsh_destroy(f->m->ordered_data);
+ gsl_histogram_free(f->m->histogram);
+ free(f->m) ;
+ free(f);
+}
+
+
+
+
+int
+factor_statistics_compare(const struct factor_statistics *f0,
+ const struct factor_statistics *f1, int width)
+{
+
+ int cmp0;
+
+ assert(f0);
+ assert(f1);
+
+ cmp0 = compare_values(&f0->id[0], &f1->id[0], width);
+
+ if ( cmp0 != 0 )
+ return cmp0;
+
+
+ if ( ( f0->id[1].f == SYSMIS ) && (f1->id[1].f != SYSMIS) )
+ return 1;
+
+ if ( ( f0->id[1].f != SYSMIS ) && (f1->id[1].f == SYSMIS) )
+ return -1;
+
+ return compare_values(&f0->id[1], &f1->id[1], width);
+
+}
+
+unsigned int
+factor_statistics_hash(const struct factor_statistics *f, int width)
+{
+
+ unsigned int h;
+
+ h = hash_value(&f->id[0], width);
+
+ if ( f->id[1].f != SYSMIS )
+ h += hash_value(&f->id[1], width);
+
+
+ return h;
+
}
+
projects / pspp-builds.git / blobdiff