/* PSPP - a program for statistical analysis.
- Copyright (C) 2009, 2010, 2011 Free Software Foundation, Inc.
+ Copyright (C) 2009, 2010, 2011, 2012 Free Software Foundation, Inc.
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
#include "gl/xalloc.h"
+#define CATEGORICALS_DEBUG 0
+
+#define EFFECTS_CODING 1
+
struct value_node
{
struct hmap_node node; /* Node in hash map. */
union value val; /* The value */
- void *user_data; /* A pointer to data which the caller can store stuff */
+ int index; /* A zero based unique index for this value */
};
struct interaction_value
{
- struct hmap_node node; /* Node in hash map. */
+ struct hmap_node node; /* Node in hash map */
- struct ccase *ccase; /* A case (probably the first in the dataset) which matches this value */
+ struct ccase *ccase; /* A case (probably the first in the dataset) which matches this value */
- /* Total of the weights of cases matching this interaction */
- double cc;
+ double cc; /* Total of the weights of cases matching this interaction */
+
+ void *user_data; /* A pointer to data which the caller can store stuff */
};
static struct value_node *
const struct variable *var; /* The variable */
struct hmap valmap; /* A map of value nodes */
+ int n_vals; /* Number of values for this variable */
};
-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);
-}
-
-
static struct variable_node *
lookup_variable (const struct hmap *map, const struct variable *var, unsigned int hash)
{
/* The number of distinct values of this interaction */
int n_cats;
- /* The degrees of freedom for this interaction */
- int df;
+ /* An array of integers df_n * df_{n-1} * df_{n-2} ...
+ These are the products of the degrees of freedom for the current
+ variable and all preceeding variables */
+ int *df_prod;
+
+ double *enc_sum;
/* A map of interaction_values indexed by subscript */
struct interaction_value **reverse_interaction_value_map;
};
-#if 0
/* Comparison function to sort the reverse_value_map in ascending order */
static int
-compare_value_node (const void *vn1_, const void *vn2_, const void *aux)
+compare_interaction_value_3way (const void *vn1_, const void *vn2_, const void *aux)
{
- const struct value_node * const *vn1 = vn1_;
- const struct value_node * const *vn2 = vn2_;
- const struct interact_params *vp = aux;
+ const struct interaction_value *const *vn1p = vn1_;
+ const struct interaction_value *const *vn2p = vn2_;
- return interaction_case_cmp_3way (vp->iact, (*vn1)->ccase, (*vn2)->ccase);
+ const struct interact_params *iap = aux;
+
+ return interaction_case_cmp_3way (iap->iact, (*vn1p)->ccase, (*vn2p)->ccase);
}
-#endif
struct categoricals
{
/* 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;
-
- /* Function to be called on each update */
- update_func *update;
+ /* Missing values in the dependent varirable to be excluded */
+ enum mv_class dep_excl;
- /* Function specified by the caller to create user_data */
- user_data_create_func *user_data_create;
+ /* Missing values in the factor variables to be excluded */
+ enum mv_class fctr_excl;
- /* Auxilliary data to be passed to update and user_data_create_func*/
- void *aux1;
+ const void *aux1;
void *aux2;
-};
+ const struct payload *payload;
+};
static void
categoricals_dump (const struct categoricals *cat)
{
- int i;
-#if 1
- 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 ("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 ("Reverse Variable Map (short):\n");
+ for (i = 0; i < cat->df_sum; ++i)
+ {
+ printf (" %d", cat->reverse_variable_map_short[i]);
+ }
+ printf ("\n");
- ds_init_empty (&str);
- interaction_to_string (iact, &str);
+ 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 ("\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);
- printf ("\t(");
- for (v = 0; v < hmap_count (&iap->ivmap); ++v)
+ printf ("Number of interactions %d\n", cat->n_iap);
+ for (i = 0 ; i < cat->n_iap; ++i)
{
- 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)
+ 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 ("\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)
{
+ struct variable_node *vn = NULL;
+ int i;
+ if (NULL == cat)
+ return;
+ for (i = 0; i < cat->n_iap; ++i)
+ {
+ struct interaction_value *iv = NULL;
+ /* 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
+ 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 */
+ HMAP_FOR_EACH (vn, struct variable_node, node, &cat->varmap)
+ {
+ hmap_destroy (&vn->valmap);
+ }
+
+ hmap_destroy (&cat->varmap);
+
+ pool_destroy (cat->pool);
+
free (cat);
}
lookup_case (const struct hmap *map, const struct interaction *iact, const struct ccase *c)
{
struct interaction_value *iv = NULL;
- size_t hash = interaction_case_hash (iact, c);
+ size_t hash = interaction_case_hash (iact, c, 0);
HMAP_FOR_EACH_WITH_HASH (iv, struct interaction_value, node, hash, map)
{
struct categoricals *
categoricals_create (struct interaction *const*inter, size_t n_inter,
- const struct variable *wv, enum mv_class exclude,
- user_data_create_func *udf,
- update_func *update, void *aux1, void *aux2
- )
+ 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->update = update;
- cat->user_data_create = udf;
-
- cat->aux1 = aux1;
- cat->aux2 = aux2;
+ cat->dep_excl = dep_excl;
+ cat->fctr_excl = fctr_excl;
+ cat->payload = NULL;
+ cat->aux2 = NULL;
cat->iap = pool_calloc (cat->pool, cat->n_iap, sizeof *cat->iap);
{
vn = pool_malloc (cat->pool, sizeof *vn);
vn->var = var;
+ vn->n_vals = 0;
hmap_init (&vn->valmap);
hmap_insert (&cat->varmap, &vn->node, hash);
if (valn == NULL)
{
valn = pool_malloc (cat->pool, sizeof *valn);
+ valn->index = vn->n_vals++;
value_init (&valn->val, width);
value_copy (&valn->val, val, width);
hmap_insert (&vn->valmap, &valn->node, hash);
}
}
-
for (i = 0 ; i < cat->n_iap; ++i)
{
const struct interaction *iact = cat->iap[i].iact;
- // if ( interaction_case_is_missing (iact, c, cat->exclude))
- // continue;
+ size_t hash;
+ struct interaction_value *node;
+
+ if ( interaction_case_is_missing (iact, c, cat->fctr_excl))
+ continue;
- size_t hash = interaction_case_hash (iact, c);
- struct interaction_value *node = lookup_case (&cat->iap[i].ivmap, iact, c);
+ hash = interaction_case_hash (iact, c, 0);
+ node = lookup_case (&cat->iap[i].ivmap, iact, c);
if ( NULL == node)
{
hmap_insert (&cat->iap[i].ivmap, &node->node, hash);
- // if (cat->user_data_create)
- // node->user_data = cat->user_data_create (cat->aux1, cat->aux2);
+ if (cat->payload)
+ {
+ node->user_data = cat->payload->create (cat->aux1, cat->aux2);
+ }
}
else
{
}
cat->iap[i].cc += weight;
- // if (cat->update)
- // cat->update (node->user_data, cat->exclude, cat->wv, NULL, c, cat->aux1, cat->aux2);
+ if (cat->payload)
+ {
+ double weight = cat->wv ? case_data (c, cat->wv)->f : 1.0;
+ cat->payload->update (cat->aux1, cat->aux2, node->user_data, c, weight);
+ }
}
}
size_t
categoricals_df (const struct categoricals *cat, size_t n)
{
- return cat->iap[n].df;
+ const struct interact_params *iap = &cat->iap[n];
+ return iap->df_prod[iap->iact->n_vars - 1];
}
size_t
categoricals_n_total (const struct categoricals *cat)
{
+ if (!categoricals_is_complete (cat))
+ return 0;
+
return cat->n_cats_total;
}
return cat->df_sum;
}
+bool
+categoricals_is_complete (const struct categoricals *cat)
+{
+ return (NULL != cat->reverse_variable_map_short);
+}
+
+
/* This function must be called *before* any call to categoricals_get_*_by subscript and
*after* all calls to categoricals_update */
-void
+bool
categoricals_done (const struct categoricals *cat_)
{
/* Implementation Note: Whilst this function is O(n) in cat->n_cats_total, in most
/* Calculate the degrees of freedom, and the number of categories */
for (i = 0 ; i < cat->n_iap; ++i)
{
+ int df = 1;
const struct interaction *iact = cat->iap[i].iact;
- cat->iap[i].df = 1;
- cat->iap[i].n_cats = 1;
+ cat->iap[i].df_prod = iact->n_vars ? xcalloc (iact->n_vars, sizeof (int)) : NULL;
+ cat->iap[i].n_cats = 1;
+
for (v = 0 ; v < iact->n_vars; ++v)
{
const struct variable *var = iact->vars[v];
struct variable_node *vn = lookup_variable (&cat->varmap, var, hash_pointer (var, 0));
- cat->iap[i].df *= hmap_count (&vn->valmap) - 1;
+ if (hmap_count (&vn->valmap) == 0)
+ return false;
+
+ cat->iap[i].df_prod[v] = df * (hmap_count (&vn->valmap) - 1);
+ df = cat->iap[i].df_prod[v];
+
cat->iap[i].n_cats *= hmap_count (&vn->valmap);
}
- cat->df_sum += cat->iap[i].df;
+ 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;
}
HMAP_FOR_EACH (ivn, struct interaction_value, node, &iap->ivmap)
{
iap->reverse_interaction_value_map[x++] = ivn;
+
}
-#if 0
+ assert (x <= iap->n_cats);
+
/* For some purposes (eg CONTRASTS in ONEWAY) the values need to be sorted */
- sort (vp->reverse_interaction_value_map, vp->n_cats, sizeof (const struct interaction_value *),
- compare_value_node, vp);
-#endif
+ sort (iap->reverse_interaction_value_map, x, sizeof (*iap->reverse_interaction_value_map),
+ compare_interaction_value_3way, iap);
+
+ /* Fill the remaining values with null */
+ for (ii = x ; ii < iap->n_cats; ++ii)
+ iap->reverse_interaction_value_map[ii] = NULL;
/* Populate the reverse variable maps. */
- for (ii = 0; ii < iap->df; ++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);
+
+ /* Tally up the sums for all the encodings */
+ for (i = 0 ; i < cat->n_iap; ++i)
+ {
+ int x, y;
+ struct interact_params *iap = &cat->iap[i];
+ const struct interaction *iact = iap->iact;
+
+ const int df = iap->df_prod ? iap->df_prod [iact->n_vars - 1] : 0;
+
+ iap->enc_sum = xcalloc (df, sizeof (*(iap->enc_sum)));
+
+ for (y = 0; y < hmap_count (&iap->ivmap); ++y)
+ {
+ struct interaction_value *iv = iap->reverse_interaction_value_map[y];
+ for (x = iap->base_subscript_short; x < iap->base_subscript_short + df ;++x)
+ {
+ 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->destroy)
+ cat->payload->destroy (cat->aux1, cat->aux2, iv->user_data);
+ }
+ }
+
+ return true;
}
static int
reverse_variable_lookup_long (const struct categoricals *cat, int subscript)
{
- printf ("%s\n", __FUNCTION__);
assert (cat->reverse_variable_map_long);
assert (subscript >= 0);
assert (subscript < cat->n_cats_total);
return cat->iap[index].iact;
}
-
-/* Return the case corresponding to SUBSCRIPT */
-static const struct ccase *
-categoricals_get_case_by_subscript (const struct categoricals *cat, int subscript)
-{
- int vindex = reverse_variable_lookup_short (cat, subscript);
- const struct interact_params *vp = &cat->iap[vindex];
- const struct interaction_value *vn = vp->reverse_interaction_value_map [subscript - vp->base_subscript_short];
- return vn->ccase;
-}
-
-
double
categoricals_get_weight_by_subscript (const struct categoricals *cat, int subscript)
{
int vindex = reverse_variable_lookup_short (cat, subscript);
const struct interact_params *vp = &cat->iap[vindex];
- const struct interaction_value *iv = vp->reverse_interaction_value_map [subscript - vp->base_subscript_short];
-
- return iv->cc;
+ return vp->enc_sum[subscript - vp->base_subscript_short];
}
/* Returns unity if the value in case C at SUBSCRIPT is equal to the category
for that subscript */
double
-categoricals_get_binary_by_subscript (const struct categoricals *cat, int subscript,
- const struct ccase *c)
+categoricals_get_code_for_case (const struct categoricals *cat, int subscript,
+ const struct ccase *c)
{
const struct interaction *iact = categoricals_get_interaction_by_subscript (cat, subscript);
- const struct ccase *c2 = categoricals_get_case_by_subscript (cat, subscript);
+ const int i = reverse_variable_lookup_short (cat, subscript);
+
+ const int base_index = cat->iap[i].base_subscript_short;
+
+ int v;
+ double result = 1.0;
+
+ const struct interact_params *iap = &cat->iap[i];
- return interaction_case_equal (iact, c, c2);
+ double dfp = 1.0;
+ for (v = 0; v < iact->n_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 unsigned int hash = value_hash (val, width, 0);
+ const struct value_node *valn = lookup_value (&vn->valmap, val, hash, width);
+
+ double bin = 1.0;
+
+ 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];
+
+#if EFFECTS_CODING
+ if ( valn->index == df )
+ bin = -1.0;
+ else
+#endif
+ if ( valn->index != index )
+ bin = 0;
+
+ result *= bin;
+ }
+
+ return result;
}
}
+/* Return a case containing the set of values corresponding to
+ the Nth Category of the IACTth interaction */
+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];
+
+ if ( n >= hmap_count (&vp->ivmap))
+ return NULL;
+
+ vn = vp->reverse_interaction_value_map [n];
+
+ return vn->ccase;
+}
+
+/* Return a the user data corresponding to the Nth Category of the IACTth interaction. */
+void *
+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 ;
+
+ if ( n >= hmap_count (&vp->ivmap))
+ return NULL;
+
+ iv = vp->reverse_interaction_value_map [n];
+
+ return iv->user_data;
+}
+
+
/* Return a case containing the set of values corresponding to SUBSCRIPT */
const struct ccase *
{
int vindex = reverse_variable_lookup_long (cat, subscript);
const struct interact_params *vp = &cat->iap[vindex];
- // const struct interaction_value *vn = vp->reverse_interaction_value_map [subscript - vp->base_subscript_long];
+ const struct interaction_value *vn = vp->reverse_interaction_value_map [subscript - vp->base_subscript_long];
- // return vn->ccase;
- return NULL;
+ return vn->ccase;
}
void *
int vindex = reverse_variable_lookup_long (cat, subscript);
const struct interact_params *vp = &cat->iap[vindex];
- //const struct value_node *vn = vp->reverse_interaction_value_map [subscript - vp->base_subscript_long];
- //return vn->user_data;
- assert (0);
- return NULL;
+ const struct interaction_value *iv = vp->reverse_interaction_value_map [subscript - vp->base_subscript_long];
+ return iv->user_data;
+}
+
+
+\f
+
+void
+categoricals_set_payload (struct categoricals *cat, const struct payload *p,
+ const void *aux1, void *aux2)
+{
+ cat->payload = p;
+ cat->aux1 = aux1;
+ cat->aux2 = aux2;
}