X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Ffactor_stats.c;h=16e1930d58188723db0ae4a7f64c3c9c9f083447;hb=f1cd7ca88d074b671844ef073b364e069672ce66;hp=cb2197ada76e1b32bd365def88d9e73c165dbb39;hpb=9c01f251cf0e5b5eb3899fc7c62cc595f3d48511;p=pspp-builds.git

diff --git a/src/factor_stats.c b/src/factor_stats.c
index cb2197ad..16e1930d 100644
--- a/src/factor_stats.c
+++ b/src/factor_stats.c
@@ -35,7 +35,10 @@ Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
 void
 metrics_precalc(struct metrics *fs)
 {
+  assert (fs) ;
+
   fs->n = 0;
+  fs->n_missing = 0;
   fs->ssq = 0;
   fs->sum = 0;
   fs->min = DBL_MAX;
@@ -44,18 +47,29 @@ metrics_precalc(struct metrics *fs)
   fs->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;
   
+  if ( ! val ) 
+    {
+      fs->n_missing += weight;
+      return ;
+    }
+
+  x = val->f;
   fs->n    += weight;
   fs->ssq  += x * x * weight;
   fs->sum  += x * weight;
@@ -69,23 +83,38 @@ metrics_calc(struct metrics *fs, const union value *val, double weight)
   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;
@@ -99,44 +128,52 @@ metrics_postcalc(struct metrics *fs)
 
   int n_data;
   
-  fs->mean = fs->sum / fs->n;
+  m->mean = m->sum / m->n;
 
-  sample_var = ( fs->ssq / fs->n  - fs->mean * fs->mean );
+  sample_var = ( m->ssq / m->n  - m->mean * m->mean );
 
-  fs->var  = fs->n * sample_var / ( fs->n - 1) ;
-  fs->stddev = sqrt(fs->var);
+  m->var  = m->n * sample_var / ( m->n - 1) ;
+  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->stderr = sqrt (m->var / m->n) ;
+
+  data = (struct weighted_value **) hsh_data(m->ordered_data);
+  n_data = hsh_count(m->ordered_data);
 
-  data = (struct weighted_value **) hsh_data(fs->ordered_data);
-  n_data = hsh_count(fs->ordered_data);
+  if ( n_data == 0 ) 
+    {
+      m->trimmed_mean = m->mean;
+      return;
+    }
 
-  fs->wv = xmalloc ( sizeof (struct weighted_value) * n_data);
+
+  m->wv = xmalloc(sizeof(struct weighted_value ) * n_data);
 
   for ( i = 0 ; i < n_data ; ++i )
-    fs->wv[i] = *(data[i]);
+      m->wv[i] = *(data[i]);
 
-  sort (fs->wv, n_data, sizeof (struct weighted_value) , 
+  sort (m->wv, n_data, sizeof (struct weighted_value) , 
 	(algo_compare_func *) compare_values, 0);
 
-
   
-  tc = fs->n * 0.05 ;
+  /* Trimmed mean calculation */
+
+  tc = m->n * 0.05 ;
   k1 = -1;
   k2 = -1;
 
 
   for ( i = 0 ; i < n_data ; ++i ) 
     {
-      cc += fs->wv[i].w;
-      fs->wv[i].cc = cc;
+      cc += m->wv[i].w;
+      m->wv[i].cc = cc;
 
-      fs->wv[i].rank = j + (fs->wv[i].w - 1) / 2.0 ;
+      m->wv[i].rank = j + (m->wv[i].w - 1) / 2.0 ;
       
-      j += fs->wv[i].w;
+      j += m->wv[i].w;
       
       if ( cc < tc ) 
 	k1 = i;
@@ -146,44 +183,127 @@ metrics_postcalc(struct metrics *fs)
   k2 = n_data;
   for ( i = n_data -1  ; i >= 0; --i ) 
     {
-      if ( tc > fs->n - fs->wv[i].cc) 
+      if ( tc > m->n - m->wv[i].cc) 
 	k2 = i;
     }
 
 
-  fs->trimmed_mean = 0;
+  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->wv[i].v.f * m->wv[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->wv[k2 - 1].cc - tc) * m->wv[k2].v.f ;
+  m->trimmed_mean += (m->wv[k1 + 1].cc - tc) * m->wv[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)
+{
+  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;
+}
+
+
+void
+factor_statistics_free(struct factor_statistics *f)
 {
-  return compare_values(f1->id, f2->id, width);
+  free(f->m) ; 
+
+  free(f);
 }
 
 
-unsigned 
-hash_indep_value(const struct factor_statistics *f, int width)
+
+
+
+
+int 
+factor_statistics_compare(const struct factor_statistics *f0,
+	                  const struct factor_statistics *f1, void *aux)
+{
+
+  int cmp0;
+
+  assert(f0);
+  assert(f1);
+
+  cmp0 = compare_values(&f0->id[0], &f1->id[0], aux);
+
+  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], aux);
+  
+}
+
+unsigned int 
+factor_statistics_hash(const struct factor_statistics *f, void *aux)
 {
-  return hash_value(f->id, width);
+  
+  unsigned int h;
+
+  h = hash_value(&f->id[0], aux);
+  
+  if ( f->id[1].f != SYSMIS )
+    h += hash_value(&f->id[1], aux);
+
+
+  return h;
+
 }
+