#include "gl/xalloc.h"
+#define CATEGORICALS_DEBUG 0
+
#define EFFECTS_CODING 1
struct value_node
int n_vals; /* Number of values for this variable */
};
-#if 0
-static void
-dump_interaction (const struct interaction *iact)
-{
- struct string str = DS_EMPTY_INITIALIZER;
- interaction_to_string (iact, &str);
- printf ("Interaction: %s\n", ds_cstr (&str));
- ds_destroy (&str);
-}
-#endif
-
-
static struct variable_node *
lookup_variable (const struct hmap *map, const struct variable *var, unsigned int hash)
{
struct interact_params
{
- /* A map indexed by a interaction_value */
+ /* A map of cases indexed by a interaction_value */
struct hmap ivmap;
- const struct interaction *iact;
+ struct interaction *iact;
int base_subscript_short;
int base_subscript_long;
/* A map to enable the lookup of variables indexed by subscript.
This map considers only the N - 1 of the N variables.
- */
+ */
int *reverse_variable_map_short;
/* Like the above, but uses all N variables */
struct pool *pool;
- /* Missing values to be excluded */
- enum mv_class exclude;
+ /* Missing values in the dependent varirable to be excluded */
+ enum mv_class dep_excl;
+
+ /* Missing values in the factor variables to be excluded */
+ enum mv_class fctr_excl;
const void *aux1;
void *aux2;
+ bool sane;
+
const struct payload *payload;
};
-#if 0
static void
categoricals_dump (const struct categoricals *cat)
{
- int i;
-
- printf ("Reverse Variable Map (short):\n");
- for (i = 0; i < cat->df_sum; ++i)
+ if (CATEGORICALS_DEBUG)
{
- printf (" %d", cat->reverse_variable_map_short[i]);
- }
- printf ("\n");
+ int i;
- printf ("Reverse Variable Map (long):\n");
- for (i = 0; i < cat->n_cats_total; ++i)
- {
- printf (" %d", cat->reverse_variable_map_long[i]);
- }
- printf ("\n");
+ printf ("Reverse Variable Map (short):\n");
+ for (i = 0; i < cat->df_sum; ++i)
+ {
+ printf (" %d", cat->reverse_variable_map_short[i]);
+ }
+ printf ("\n");
+ printf ("Reverse Variable Map (long):\n");
+ for (i = 0; i < cat->n_cats_total; ++i)
+ {
+ printf (" %d", cat->reverse_variable_map_long[i]);
+ }
+ printf ("\n");
- printf ("Number of interactions %d\n", cat->n_iap);
- for (i = 0 ; i < cat->n_iap; ++i)
- {
- int v;
- struct string str;
- const struct interact_params *iap = &cat->iap[i];
- const struct interaction *iact = iap->iact;
- ds_init_empty (&str);
- interaction_to_string (iact, &str);
+ printf ("Number of interactions %d\n", cat->n_iap);
+ for (i = 0 ; i < cat->n_iap; ++i)
+ {
+ int v;
+ struct string str;
+ const struct interact_params *iap = &cat->iap[i];
+ const struct interaction *iact = iap->iact;
- printf ("\nInteraction: %s (n: %d; df: %d ); ", ds_cstr (&str), iap->n_cats, iap->df);
- ds_destroy (&str);
- printf ("Base subscript: %d\n", iap->base_subscript_short);
+ ds_init_empty (&str);
+ interaction_to_string (iact, &str);
- printf ("\t(");
- for (v = 0; v < hmap_count (&iap->ivmap); ++v)
- {
- int vv;
- const struct interaction_value *iv = iap->reverse_interaction_value_map[v];
-
- if (v > 0) printf (" ");
- printf ("{");
- for (vv = 0; vv < iact->n_vars; ++vv)
+ printf ("\nInteraction: %s (n: %d); ", ds_cstr (&str), iap->n_cats);
+ ds_destroy (&str);
+ printf ("Base subscript: %d\n", iap->base_subscript_short);
+
+ printf ("\t(");
+ for (v = 0; v < hmap_count (&iap->ivmap); ++v)
{
- const struct variable *var = iact->vars[vv];
- const union value *val = case_data (iv->ccase, var);
+ int vv;
+ const struct interaction_value *iv = iap->reverse_interaction_value_map[v];
+
+ if (v > 0) printf (" ");
+ printf ("{");
+ for (vv = 0; vv < iact->n_vars; ++vv)
+ {
+ const struct variable *var = iact->vars[vv];
+ const union value *val = case_data (iv->ccase, var);
- printf ("%g", val->f);
- if (vv < iact->n_vars - 1)
- printf (", ");
+ printf ("%g", val->f);
+ if (vv < iact->n_vars - 1)
+ printf (", ");
+ }
+ printf ("}");
}
- printf ("}");
+ printf (")\n");
}
- printf (")\n");
}
}
-#endif
void
categoricals_destroy (struct categoricals *cat)
/* Interate over each interaction value, and unref any cases that we reffed */
HMAP_FOR_EACH (iv, struct interaction_value, node, &cat->iap[i].ivmap)
{
-#if 0
- if (cat->payload)
- cat->payload->destroy (cat->aux1, iv->user_data);
-#endif
+ if (cat->payload && cat->payload->destroy)
+ cat->payload->destroy (cat->aux1, cat->aux2, iv->user_data);
case_unref (iv->ccase);
}
free (cat->iap[i].enc_sum);
free (cat->iap[i].df_prod);
hmap_destroy (&cat->iap[i].ivmap);
+ interaction_destroy (cat->iap[i].iact);
}
/* Interate over each variable and delete its value map */
return iv;
}
+bool
+categoricals_sane (const struct categoricals *cat)
+{
+ return cat->sane;
+}
struct categoricals *
categoricals_create (struct interaction *const*inter, size_t n_inter,
- const struct variable *wv, enum mv_class exclude)
+ const struct variable *wv, enum mv_class dep_excl, enum mv_class fctr_excl)
{
size_t i;
struct categoricals *cat = xmalloc (sizeof *cat);
cat->reverse_variable_map_short = NULL;
cat->reverse_variable_map_long = NULL;
cat->pool = pool_create ();
- cat->exclude = exclude;
+ cat->dep_excl = dep_excl;
+ cat->fctr_excl = fctr_excl;
cat->payload = NULL;
cat->aux2 = NULL;
+ cat->sane = false;
cat->iap = pool_calloc (cat->pool, cat->n_iap, sizeof *cat->iap);
{
int i;
struct variable_node *vn = NULL;
- const double weight = cat->wv ? case_data (c, cat->wv)->f : 1.0;
+ double weight;
+
+ if (NULL == cat)
+ return;
+
+ weight = cat->wv ? case_data (c, cat->wv)->f : 1.0;
assert (NULL == cat->reverse_variable_map_short);
assert (NULL == cat->reverse_variable_map_long);
size_t hash;
struct interaction_value *node;
- if ( interaction_case_is_missing (iact, c, cat->exclude))
+ if ( interaction_case_is_missing (iact, c, cat->fctr_excl))
continue;
hash = interaction_case_hash (iact, c, 0);
size_t
categoricals_df_total (const struct categoricals *cat)
{
+ if (NULL == cat)
+ return 0;
+
return cat->df_sum;
}
/* This function must be called *before* any call to categoricals_get_*_by subscript and
*after* all calls to categoricals_update */
-bool
+void
categoricals_done (const struct categoricals *cat_)
{
/* Implementation Note: Whilst this function is O(n) in cat->n_cats_total, in most
int i;
int idx_short = 0;
int idx_long = 0;
+
+ if (NULL == cat)
+ return;
+
cat->df_sum = 0;
cat->n_cats_total = 0;
int df = 1;
const struct interaction *iact = cat->iap[i].iact;
- cat->iap[i].df_prod = xcalloc (iact->n_vars, sizeof (int));
+ cat->iap[i].df_prod = iact->n_vars ? xcalloc (iact->n_vars, sizeof (int)) : NULL;
cat->iap[i].n_cats = 1;
struct variable_node *vn = lookup_variable (&cat->varmap, var, hash_pointer (var, 0));
if (hmap_count (&vn->valmap) == 0)
- return false;
+ {
+ cat->sane = false;
+ return;
+ }
cat->iap[i].df_prod[v] = df * (hmap_count (&vn->valmap) - 1);
- df = cat->iap[i].df_prod[v];
+ df = cat->iap[i].df_prod[v];
cat->iap[i].n_cats *= hmap_count (&vn->valmap);
}
- cat->df_sum += cat->iap[i].df_prod [v - 1];
+ assert (v == iact->n_vars);
+ if (v > 0)
+ cat->df_sum += cat->iap[i].df_prod [v - 1];
+
cat->n_cats_total += cat->iap[i].n_cats;
}
iap->reverse_interaction_value_map[ii] = NULL;
/* Populate the reverse variable maps. */
- for (ii = 0; ii < iap->df_prod [iap->iact->n_vars - 1]; ++ii)
- cat->reverse_variable_map_short[idx_short++] = i;
+ if (iap->df_prod)
+ {
+ for (ii = 0; ii < iap->df_prod [iap->iact->n_vars - 1]; ++ii)
+ cat->reverse_variable_map_short[idx_short++] = i;
+ }
for (ii = 0; ii < iap->n_cats; ++ii)
cat->reverse_variable_map_long[idx_long++] = i;
assert (cat->n_vars <= cat->n_iap);
- // categoricals_dump (cat);
+ categoricals_dump (cat);
/* Tally up the sums for all the encodings */
for (i = 0 ; i < cat->n_iap; ++i)
struct interact_params *iap = &cat->iap[i];
const struct interaction *iact = iap->iact;
- const int df = iap->df_prod [iact->n_vars - 1];
+ const int df = iap->df_prod ? iap->df_prod [iact->n_vars - 1] : 0;
iap->enc_sum = xcalloc (df, sizeof (*(iap->enc_sum)));
const double bin = categoricals_get_code_for_case (cat, x, iv->ccase); \
iap->enc_sum [x - iap->base_subscript_short] += bin * iv->cc;
}
+ if (cat->payload && cat->payload->calculate)
+ cat->payload->calculate (cat->aux1, cat->aux2, iv->user_data);
}
}
- return true;
+ cat->sane = true;
}
for that subscript */
double
categoricals_get_code_for_case (const struct categoricals *cat, int subscript,
- const struct ccase *c)
+ const struct ccase *c)
{
const struct interaction *iact = categoricals_get_interaction_by_subscript (cat, subscript);
double dfp = 1.0;
for (v = 0; v < iact->n_vars; ++v)
- {
- const struct variable *var = iact->vars[v];
+ {
+ const struct variable *var = iact->vars[v];
- const union value *val = case_data (c, var);
- const int width = var_get_width (var);
- const struct variable_node *vn = lookup_variable (&cat->varmap, var, hash_pointer (var, 0));
+ const union value *val = case_data (c, var);
+ const int width = var_get_width (var);
+ const struct variable_node *vn = lookup_variable (&cat->varmap, var, hash_pointer (var, 0));
- const unsigned int hash = value_hash (val, width, 0);
- const struct value_node *valn = lookup_value (&vn->valmap, val, hash, width);
+ const unsigned int hash = value_hash (val, width, 0);
+ const struct value_node *valn = lookup_value (&vn->valmap, val, hash, width);
- double bin = 1.0;
+ double bin = 1.0;
- const double df = iap->df_prod[v] / dfp;
+ const double df = iap->df_prod[v] / dfp;
- /* Translate the subscript into an index for the individual variable */
- const int index = ((subscript - base_index) % iap->df_prod[v] ) / dfp;
- dfp = iap->df_prod [v];
+ /* Translate the subscript into an index for the individual variable */
+ const int index = ((subscript - base_index) % iap->df_prod[v] ) / dfp;
+ dfp = iap->df_prod [v];
#if EFFECTS_CODING
- if ( valn->index == df )
- bin = -1.0;
- else
+ if ( valn->index == df )
+ bin = -1.0;
+ else
#endif
- if ( valn->index != index )
- bin = 0;
+ if ( valn->index != index )
+ bin = 0;
- result *= bin;
- }
+ result *= bin;
+ }
return result;
}
const struct ccase *
categoricals_get_case_by_category_real (const struct categoricals *cat, int iact, int n)
{
+ const struct interaction_value *vn;
+
const struct interact_params *vp = &cat->iap[iact];
- const struct interaction_value *vn = vp->reverse_interaction_value_map [n];
+
+ if ( n >= hmap_count (&vp->ivmap))
+ return NULL;
+
+ vn = vp->reverse_interaction_value_map [n];
return vn->ccase;
}
categoricals_get_user_data_by_category_real (const struct categoricals *cat, int iact, int n)
{
const struct interact_params *vp = &cat->iap[iact];
- const struct interaction_value *iv = vp->reverse_interaction_value_map [n];
+ const struct interaction_value *iv ;
+
+ if ( n >= hmap_count (&vp->ivmap))
+ return NULL;
+
+ iv = vp->reverse_interaction_value_map [n];
return iv->user_data;
}