/* PSPP - a program for statistical analysis.
- Copyright (C) 2009, 2010, 2011, 2012 Free Software Foundation, Inc.
+ Copyright (C) 2009, 2010, 2011, 2012, 2014 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 "math/categoricals.h"
#include "math/interaction.h"
+#include <float.h>
#include <stdio.h>
#include "data/case.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 */
-
int index; /* A zero based unique index for this value */
};
-struct interaction_value
-{
- struct hmap_node node; /* Node in hash map */
-
- struct ccase *ccase; /* A case (probably the first in the dataset) which matches this value */
-
- 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 *
-lookup_value (const struct hmap *map, const union value *val, unsigned int hash, int width)
+lookup_value (const struct hmap *map, const union value *val,
+ unsigned int hash, int width)
{
- struct value_node *vn = NULL;
+ struct value_node *vn;
HMAP_FOR_EACH_WITH_HASH (vn, struct value_node, node, hash, map)
- {
- if (value_equal (&vn->val, val, width))
- break;
- }
-
- return vn;
+ if (value_equal (&vn->val, val, width))
+ return vn;
+ return NULL;
}
+/* A variable used in a categoricals object. */
struct variable_node
+ {
+ struct hmap_node node; /* In struct categorical's 'varmap'. */
+ const struct variable *var; /* The variable. */
+ struct hmap valmap; /* Contains "struct value_node"s. */
+ union value *values; /* Values in valmap, as a sorted array. */
+ };
+
+static int
+compare_value_node_3way (const void *vn1_, const void *vn2_, const void *aux)
{
- struct hmap_node node; /* Node in hash map. */
- const struct variable *var; /* The variable */
+ const struct value_node *const *vn1p = vn1_;
+ const struct value_node *const *vn2p = vn2_;
- struct hmap valmap; /* A map of value nodes */
- int n_vals; /* Number of values for this variable */
-};
+ const struct variable_node *vn = aux;
-static struct variable_node *
-lookup_variable (const struct hmap *map, const struct variable *var, unsigned int hash)
-{
- struct variable_node *vn = NULL;
- HMAP_FOR_EACH_WITH_HASH (vn, struct variable_node, node, hash, map)
- {
- if (vn->var == var)
- break;
-
- fprintf (stderr, "Warning: Hash table collision\n");
- }
-
- return vn;
+ return value_compare_3way (&(*vn1p)->val, &(*vn2p)->val,
+ var_get_width (vn->var));
}
+static struct variable_node *
+lookup_variable (const struct hmap *map, const struct variable *var)
+{
+ struct variable_node *vn;
+ HMAP_FOR_EACH_WITH_HASH (vn, struct variable_node, node,
+ hash_pointer (var, 0), map)
+ if (vn->var == var)
+ return vn;
+ return NULL;
+}
struct interact_params
{
- /* A map indexed by a interaction_value */
- struct hmap ivmap;
+ /* The interaction, and an array with iact->n_vars elements such that
+ varnodes[x] points to the variable_node for iact->vars[x]. */
+ const struct interaction *iact;
+ struct variable_node **varnodes;
- struct interaction *iact;
+ /* An example of each interaction that appears in the data, like a frequency
+ table for 'iact'. By construction, the number of elements must be less
+ than or equal to 'n_cats'.
- int base_subscript_short;
- int base_subscript_long;
+ categoricals_update() updates 'ivmap' case-by-case, then
+ categoricals_done() dumps 'ivmap' into 'ivs' and sorts it. */
+ struct hmap ivmap; /* Contains "struct interaction_value"s. */
+ struct interaction_value **ivs;
- /* The number of distinct values of this interaction */
+ int base_df;
+ int base_cats;
+
+ /* Product of hmap_count(&varnodes[*]->valmap), that is, the maximum number
+ of distinct values of this interaction. */
int n_cats;
- /* 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;
+ /* Product of degrees of freedom of all the variables. */
+ int df_prod;
double *enc_sum;
- /* A map of interaction_values indexed by subscript */
- struct interaction_value **reverse_interaction_value_map;
-
+ /* Sum of ivs[*]->cc. */
double cc;
};
+struct interaction_value
+ {
+ struct hmap_node node; /* In struct interact_params's ivmap. */
+ struct ccase *ccase; /* A case representative of the interaction. */
+ double cc; /* Total weight of cases for this interaction. */
+ void *user_data;
+ };
-/* Comparison function to sort the reverse_value_map in ascending order */
static int
-compare_interaction_value_3way (const void *vn1_, const void *vn2_, const void *aux)
+compare_interaction_value_3way (const void *vn1_, const void *vn2_,
+ const void *aux)
{
const struct interaction_value *const *vn1p = vn1_;
const struct interaction_value *const *vn2p = vn2_;
/* The weight variable */
const struct variable *wv;
- /* An array of interact_params */
- struct interact_params *iap;
-
- /* Map whose members are the union of the variables which comprise IAP */
- struct hmap varmap;
-
- /* The size of IAP. (ie, the number of interactions involved.) */
+ struct interact_params *iap; /* Interaction parameters. */
size_t n_iap;
- /* The number of categorical variables which contain entries.
- In the absence of missing values, this will be equal to N_IAP */
- size_t n_vars;
-
- size_t df_sum;
+ /* Contains a "struct variable_node" for each variable in 'iap'. */
+ struct hmap varmap;
/* 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 */
- int *reverse_variable_map_long;
+ int *df_to_iact; /* 'df_sum' elements. */
+ size_t df_sum;
+ /* Like the above, but uses all N variables. */
+ int *cat_to_iact; /* 'n_cats_total' elements. */
size_t n_cats_total;
struct pool *pool;
- /* 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 struct payload *payload;
};
+
+bool
+categoricals_isbalanced (const struct categoricals *cat)
+{
+ for (int i = 0 ; i < cat->n_iap; ++i)
+ {
+ int v;
+ const struct interact_params *iap = &cat->iap[i];
+
+ double oval = -1.0;
+ for (v = 0; v < hmap_count (&iap->ivmap); ++v)
+ {
+ const struct interaction_value *iv = iap->ivs[v];
+ if (oval == -1.0)
+ oval = iv->cc;
+ if (oval != iv->cc)
+ return false;
+ }
+ }
+ return true;
+}
+
+
static void
categoricals_dump (const struct categoricals *cat)
{
{
int i;
- printf ("Reverse Variable Map (short):\n");
+ printf ("df to interaction map:\n");
for (i = 0; i < cat->df_sum; ++i)
- {
- printf (" %d", cat->reverse_variable_map_short[i]);
- }
+ printf (" %d", cat->df_to_iact[i]);
printf ("\n");
- printf ("Reverse Variable Map (long):\n");
+ printf ("Category to interaction map:\n");
for (i = 0; i < cat->n_cats_total; ++i)
- {
- printf (" %d", cat->reverse_variable_map_long[i]);
- }
+ printf (" %d", cat->cat_to_iact[i]);
printf ("\n");
-
- printf ("Number of interactions %d\n", cat->n_iap);
+ printf ("Number of interactions %zu\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 ("\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)
- {
- 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 ("}");
- }
- printf (")\n");
- }
+ {
+ 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\" (number of categories: %d); ", ds_cstr (&str), iap->n_cats);
+ ds_destroy (&str);
+ printf ("Base index (df/categories): %d/%d\n", iap->base_df, iap->base_cats);
+
+ printf ("\t(");
+ for (v = 0; v < hmap_count (&iap->ivmap); ++v)
+ {
+ int vv;
+ const struct interaction_value *iv = iap->ivs[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);
+ struct variable_node *vn = iap->varnodes[vv];
+ const int width = var_get_width (var);
+ unsigned int valhash = value_hash (val, width, 0);
+ struct value_node *valn = lookup_value (&vn->valmap, val, valhash, width);
+
+ assert (vn->var == var);
+
+ printf ("%.*g(%d)", DBL_DIG + 1, val->f, valn->index);
+ if (vv < iact->n_vars - 1)
+ printf (", ");
+ }
+ printf ("}");
+ }
+ printf (")\n");
+ }
}
}
void
categoricals_destroy (struct categoricals *cat)
{
- struct variable_node *vn = NULL;
- int i;
- if (NULL == cat)
+ if (!cat)
return;
- for (i = 0; i < cat->n_iap; ++i)
+
+ for (int i = 0; i < cat->n_iap; ++i)
{
- struct interaction_value *iv = NULL;
- /* Interate over each interaction value, and unref any cases that we reffed */
+ /* Unref any cases that we reffed. */
+ struct interaction_value *iv;
HMAP_FOR_EACH (iv, struct interaction_value, node, &cat->iap[i].ivmap)
- {
- if (cat->payload && cat->payload->destroy)
- cat->payload->destroy (cat->aux1, iv->user_data);
- case_unref (iv->ccase);
- }
+ {
+ 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);
}
+ free (cat->iap);
+
+ /* Interate over each variable and delete its value map.
- /* Interate over each variable and delete its value map */
+ The values themselves are part of the pool. */
+ struct variable_node *vn;
HMAP_FOR_EACH (vn, struct variable_node, node, &cat->varmap)
- {
- hmap_destroy (&vn->valmap);
- }
+ hmap_destroy (&vn->valmap);
hmap_destroy (&cat->varmap);
free (cat);
}
-
-
static struct interaction_value *
-lookup_case (const struct hmap *map, const struct interaction *iact, const struct ccase *c)
+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, 0);
-
+ struct interaction_value *iv;
HMAP_FOR_EACH_WITH_HASH (iv, struct interaction_value, node, hash, map)
- {
- if (interaction_case_equal (iact, c, iv->ccase))
- break;
-
- fprintf (stderr, "Warning: Hash table collision\n");
- }
-
- return iv;
+ if (interaction_case_equal (iact, c, iv->ccase))
+ return iv;
+ return NULL;
}
-bool
+/* Returns true iff CAT is sane, that is, if it is complete and has at least
+ one value. */
+bool
categoricals_sane (const struct categoricals *cat)
{
return cat->sane;
}
+/* Creates and returns a new categoricals object whose variables come from the
+ N_INTER interactions objects in the array starting at INTER. (The INTER
+ objects must outlive the categoricals object because it uses them
+ internally.)
+
+ FCTR_EXCL determines which cases are listwise ignored by
+ categoricals_update(). */
struct categoricals *
-categoricals_create (struct interaction *const*inter, size_t n_inter,
- const struct variable *wv, enum mv_class dep_excl, enum mv_class fctr_excl)
+categoricals_create (struct interaction *const *inter, size_t n_inter,
+ const struct variable *wv, enum mv_class fctr_excl)
{
- size_t i;
- struct categoricals *cat = xmalloc (sizeof *cat);
-
+ struct categoricals *cat = XZALLOC (struct categoricals);
+ cat->iap = pool_calloc (cat->pool, n_inter, sizeof *cat->iap);
cat->n_iap = n_inter;
cat->wv = wv;
- cat->n_cats_total = 0;
- cat->n_vars = 0;
- cat->reverse_variable_map_short = NULL;
- cat->reverse_variable_map_long = NULL;
cat->pool = pool_create ();
- 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);
hmap_init (&cat->varmap);
- for (i = 0 ; i < cat->n_iap; ++i)
+ for (size_t i = 0; i < cat->n_iap; ++i)
{
- int v;
- hmap_init (&cat->iap[i].ivmap);
- cat->iap[i].iact = inter[i];
- cat->iap[i].cc = 0.0;
- for (v = 0; v < inter[i]->n_vars; ++v)
- {
- const struct variable *var = inter[i]->vars[v];
- unsigned int hash = hash_pointer (var, 0);
- struct variable_node *vn = lookup_variable (&cat->varmap, var, hash);
- if (vn == NULL)
- {
- 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);
- }
- }
+ struct interact_params *iap = &cat->iap[i];
+ hmap_init (&iap->ivmap);
+ iap->iact = inter[i];
+ iap->cc = 0.0;
+ iap->varnodes = pool_nmalloc (cat->pool, iap->iact->n_vars,
+ sizeof *iap->varnodes);
+ for (size_t v = 0; v < inter[i]->n_vars; ++v)
+ {
+ const struct variable *var = inter[i]->vars[v];
+ struct variable_node *vn = lookup_variable (&cat->varmap, var);
+ if (!vn)
+ {
+ vn = pool_malloc (cat->pool, sizeof *vn);
+ vn->var = var;
+ vn->values = NULL;
+ hmap_init (&vn->valmap);
+ hmap_insert (&cat->varmap, &vn->node, hash_pointer (var, 0));
+ }
+ iap->varnodes[v] = vn;
+ }
}
return cat;
}
-
-
void
categoricals_update (struct categoricals *cat, const struct ccase *c)
{
- int i;
- struct variable_node *vn = NULL;
- const double weight = cat->wv ? case_data (c, cat->wv)->f : 1.0;
+ if (!cat)
+ return;
+ assert (!cat->df_to_iact);
+ assert (!cat->cat_to_iact);
- assert (NULL == cat->reverse_variable_map_short);
- assert (NULL == cat->reverse_variable_map_long);
+ double weight;
+ weight = cat->wv ? case_num (c, cat->wv) : 1.0;
+ weight = var_force_valid_weight (cat->wv, weight, NULL);
- /* Interate over each variable, and add the value of that variable
- to the appropriate map, if it's not already present. */
+ /* Update the frequency table for each variable. */
+ struct variable_node *vn;
HMAP_FOR_EACH (vn, struct variable_node, node, &cat->varmap)
{
const int width = var_get_width (vn->var);
struct value_node *valn = lookup_value (&vn->valmap, val, hash, width);
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;
-
- size_t hash;
- struct interaction_value *node;
-
- if ( interaction_case_is_missing (iact, c, cat->fctr_excl))
- continue;
-
- hash = interaction_case_hash (iact, c, 0);
- node = lookup_case (&cat->iap[i].ivmap, iact, c);
-
- if ( NULL == node)
- {
- node = pool_malloc (cat->pool, sizeof *node);
-
- node->ccase = case_ref (c);
- node->cc = weight;
-
- hmap_insert (&cat->iap[i].ivmap, &node->node, hash);
+ {
+ valn = pool_malloc (cat->pool, sizeof *valn);
+ valn->index = -1;
+ value_init_pool (cat->pool, &valn->val, width);
+ value_copy (&valn->val, val, width);
+ hmap_insert (&vn->valmap, &valn->node, hash);
+ }
+ }
- if (cat->payload)
- {
- node->user_data = cat->payload->create (cat->aux1, cat->aux2);
- }
- }
+ /* Update the frequency table for full interactions. */
+ for (int i = 0; i < cat->n_iap; ++i)
+ {
+ struct interact_params *iap = &cat->iap[i];
+ const struct interaction *iact = iap->iact;
+ if (interaction_case_is_missing (iact, c, cat->fctr_excl))
+ continue;
+
+ unsigned int hash = interaction_case_hash (iact, c, 0);
+ struct interaction_value *node = lookup_case (&iap->ivmap, iact, c);
+ if (!node)
+ {
+ node = pool_malloc (cat->pool, sizeof *node);
+ node->ccase = case_ref (c);
+ node->cc = weight;
+
+ hmap_insert (&iap->ivmap, &node->node, hash);
+
+ if (cat->payload)
+ node->user_data = cat->payload->create (cat->aux1, cat->aux2);
+ }
else
- {
- node->cc += weight;
- }
- cat->iap[i].cc += weight;
+ node->cc += weight;
+ iap->cc += weight;
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);
- }
-
+ cat->payload->update (cat->aux1, cat->aux2, node->user_data, c,
+ weight);
}
}
-/* Return the number of categories (distinct values) for interction N */
+/* Return the number of categories (distinct values) for interaction IDX in
+ CAT. */
size_t
-categoricals_n_count (const struct categoricals *cat, size_t n)
+categoricals_n_count (const struct categoricals *cat, size_t idx)
{
- return hmap_count (&cat->iap[n].ivmap);
+ return hmap_count (&cat->iap[idx].ivmap);
}
-
+/* Return the total number of categories across all interactions in CAT. */
size_t
-categoricals_df (const struct categoricals *cat, size_t n)
+categoricals_n_total (const struct categoricals *cat)
{
- const struct interact_params *iap = &cat->iap[n];
- return iap->df_prod[iap->iact->n_vars - 1];
+ return categoricals_is_complete (cat) ? cat->n_cats_total : 0;
}
-
-/* Return the total number of categories */
+/* Returns the number of degrees of freedom for interaction IDX within CAT. */
size_t
-categoricals_n_total (const struct categoricals *cat)
+categoricals_df (const struct categoricals *cat, size_t idx)
{
- if (!categoricals_is_complete (cat))
- return 0;
-
- return cat->n_cats_total;
+ const struct interact_params *iap = &cat->iap[idx];
+ return iap->df_prod;
}
+/* Returns the total degrees of freedom for CAT. */
size_t
categoricals_df_total (const struct categoricals *cat)
{
- return cat->df_sum;
+ return cat ? cat->df_sum : 0;
}
+/* Returns true iff categoricals_done() has been called for CAT. */
bool
categoricals_is_complete (const struct categoricals *cat)
{
- return (NULL != cat->reverse_variable_map_short);
+ return cat->df_to_iact != NULL;
}
-
-/* This function must be called *before* any call to categoricals_get_*_by subscript and
- *after* all calls to categoricals_update */
+/* This function must be called (once) before any call to the *_by_subscript or
+ *_by_category functions, but AFTER any calls to categoricals_update. If this
+ function returns false, then no calls to _by_subscript or *_by_category are
+ allowed. */
void
categoricals_done (const struct categoricals *cat_)
{
- /* Implementation Note: Whilst this function is O(n) in cat->n_cats_total, in most
- uses it will be more efficient that using a tree based structure, since it
- is called only once, and means that subsequent lookups will be O(1).
-
- 1 call of O(n) + 10^9 calls of O(1) is better than 10^9 calls of O(log n).
- */
struct categoricals *cat = CONST_CAST (struct categoricals *, cat_);
- int v;
- int i;
- int idx_short = 0;
- int idx_long = 0;
- cat->df_sum = 0;
- cat->n_cats_total = 0;
+ if (!cat || categoricals_is_complete (cat))
+ return;
- /* Calculate the degrees of freedom, and the number of categories */
- for (i = 0 ; i < cat->n_iap; ++i)
+ /* Assign 'index' to each variables' value_nodes, counting up from 0 in
+ ascending order by value. */
+ struct variable_node *vn;
+ HMAP_FOR_EACH (vn, struct variable_node, node, &cat->varmap)
{
- int df = 1;
- const struct interaction *iact = cat->iap[i].iact;
-
- 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));
-
- if (hmap_count (&vn->valmap) == 0)
- {
- cat->sane = false;
- return;
- }
+ size_t n_vals = hmap_count (&vn->valmap);
+ if (!n_vals)
+ {
+ cat->sane = false;
+ return;
+ }
+
+ struct value_node **nodes = XCALLOC (n_vals, struct value_node *);
+ int x = 0;
+ struct value_node *valnd;
+ HMAP_FOR_EACH (valnd, struct value_node, node, &vn->valmap)
+ nodes[x++] = valnd;
+ sort (nodes, n_vals, sizeof *nodes, compare_value_node_3way, vn);
+ for (x = 0; x < n_vals; ++x)
+ nodes[x]->index = x;
+ free (nodes);
+ }
- cat->iap[i].df_prod[v] = df * (hmap_count (&vn->valmap) - 1);
- df = cat->iap[i].df_prod[v];
+ /* Calculate the degrees of freedom, and the number of categories. */
+ cat->df_sum = 0;
+ cat->n_cats_total = 0;
+ for (int i = 0 ; i < cat->n_iap; ++i)
+ {
+ struct interact_params *iap = &cat->iap[i];
+ const struct interaction *iact = iap->iact;
- cat->iap[i].n_cats *= hmap_count (&vn->valmap);
- }
+ iap->df_prod = 1;
+ iap->n_cats = 1;
+ for (int v = 0 ; v < iact->n_vars; ++v)
+ {
+ size_t n_vals = hmap_count (&iap->varnodes[v]->valmap);
- assert (v == iact->n_vars);
- if (v > 0)
- cat->df_sum += cat->iap[i].df_prod [v - 1];
+ iap->df_prod *= n_vals - 1;
+ iap->n_cats *= n_vals;
+ }
- cat->n_cats_total += cat->iap[i].n_cats;
+ if (iact->n_vars > 0)
+ cat->df_sum += iap->df_prod;
+ cat->n_cats_total += iap->n_cats;
}
- cat->reverse_variable_map_short = pool_calloc (cat->pool,
- cat->df_sum,
- sizeof *cat->reverse_variable_map_short);
+ cat->df_to_iact = pool_calloc (cat->pool, cat->df_sum,
+ sizeof *cat->df_to_iact);
- cat->reverse_variable_map_long = pool_calloc (cat->pool,
- cat->n_cats_total,
- sizeof *cat->reverse_variable_map_long);
+ cat->cat_to_iact = pool_calloc (cat->pool, cat->n_cats_total,
+ sizeof *cat->cat_to_iact);
- for (i = 0 ; i < cat->n_iap; ++i)
+ int idx_df = 0;
+ int idx_cat = 0;
+ for (int i = 0; i < cat->n_iap; ++i)
{
- struct interaction_value *ivn = NULL;
- int x = 0;
- int ii;
struct interact_params *iap = &cat->iap[i];
- iap->base_subscript_short = idx_short;
- iap->base_subscript_long = idx_long;
-
- iap->reverse_interaction_value_map = pool_calloc (cat->pool, iap->n_cats,
- sizeof *iap->reverse_interaction_value_map);
+ iap->base_df = idx_df;
+ iap->base_cats = idx_cat;
+ /* For some purposes (eg CONTRASTS in ONEWAY) the values need to be
+ sorted. */
+ iap->ivs = pool_nmalloc (cat->pool, hmap_count (&iap->ivmap),
+ sizeof *iap->ivs);
+ int x = 0;
+ struct interaction_value *ivn;
HMAP_FOR_EACH (ivn, struct interaction_value, node, &iap->ivmap)
- {
- iap->reverse_interaction_value_map[x++] = ivn;
-
- }
-
- assert (x <= iap->n_cats);
+ iap->ivs[x++] = ivn;
+ sort (iap->ivs, x, sizeof *iap->ivs,
+ compare_interaction_value_3way, iap);
- /* For some purposes (eg CONTRASTS in ONEWAY) the values need to be sorted */
- sort (iap->reverse_interaction_value_map, x, sizeof (*iap->reverse_interaction_value_map),
- compare_interaction_value_3way, iap);
+ /* Populate the variable maps. */
+ if (iap->iact->n_vars)
+ for (int j = 0; j < iap->df_prod; ++j)
+ cat->df_to_iact[idx_df++] = i;
- /* 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. */
- 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;
+ for (int j = 0; j < iap->n_cats; ++j)
+ cat->cat_to_iact[idx_cat++] = 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)
+ for (int 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->calculate)
- cat->payload->calculate (cat->aux1, cat->aux2, iv->user_data);
- }
+ const int df = iact->n_vars ? iap->df_prod : 0;
+
+ iap->enc_sum = xcalloc (df, sizeof *iap->enc_sum);
+
+ for (int y = 0; y < hmap_count (&iap->ivmap); ++y)
+ {
+ struct interaction_value *iv = iap->ivs[y];
+ for (int x = iap->base_df;
+ x < iap->base_df + df; ++x)
+ {
+ const double bin = categoricals_get_effects_code_for_case (
+ cat, x, iv->ccase);
+ iap->enc_sum[x - iap->base_df] += bin * iv->cc;
+ }
+ if (cat->payload && cat->payload->calculate)
+ cat->payload->calculate (cat->aux1, cat->aux2, iv->user_data);
+ }
}
cat->sane = true;
}
+union value *
+categoricals_get_var_values (const struct categoricals *cat,
+ const struct variable *var, size_t *np)
+{
+ struct variable_node *vn = lookup_variable (&cat->varmap, var);
+ *np = hmap_count (&vn->valmap);
+ if (!vn->values)
+ {
+ vn->values = pool_nalloc (cat->pool, *np, sizeof *vn->values);
-static int
-reverse_variable_lookup_short (const struct categoricals *cat, int subscript)
+ struct value_node *valnd;
+ HMAP_FOR_EACH (valnd, struct value_node, node, &vn->valmap)
+ vn->values[valnd->index] = valnd->val;
+ }
+ return vn->values;
+}
+
+static struct interact_params *
+df_to_iap (const struct categoricals *cat, int subscript)
{
- assert (cat->reverse_variable_map_short);
assert (subscript >= 0);
assert (subscript < cat->df_sum);
- return cat->reverse_variable_map_short[subscript];
+ return &cat->iap[cat->df_to_iact[subscript]];
}
-static int
-reverse_variable_lookup_long (const struct categoricals *cat, int subscript)
+static struct interact_params *
+cat_index_to_iap (const struct categoricals *cat, int cat_index)
{
- assert (cat->reverse_variable_map_long);
- assert (subscript >= 0);
- assert (subscript < cat->n_cats_total);
+ assert (cat_index >= 0);
+ assert (cat_index < cat->n_cats_total);
- return cat->reverse_variable_map_long[subscript];
+ return &cat->iap[cat->cat_to_iact[cat_index]];
}
-
-/* Return the interaction corresponding to SUBSCRIPT */
+/* Return the interaction corresponding to SUBSCRIPT. */
const struct interaction *
-categoricals_get_interaction_by_subscript (const struct categoricals *cat, int subscript)
+categoricals_get_interaction_by_subscript (const struct categoricals *cat,
+ int subscript)
{
- int index = reverse_variable_lookup_short (cat, subscript);
-
- return cat->iap[index].iact;
+ return df_to_iap (cat, subscript)->iact;
}
double
-categoricals_get_weight_by_subscript (const struct categoricals *cat, int subscript)
+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];
-
- return vp->cc;
+ return df_to_iap (cat, subscript)->cc;
}
double
-categoricals_get_sum_by_subscript (const struct categoricals *cat, int subscript)
+categoricals_get_sum_by_subscript (const struct categoricals *cat,
+ int subscript)
{
- int vindex = reverse_variable_lookup_short (cat, subscript);
- const struct interact_params *vp = &cat->iap[vindex];
-
- return vp->enc_sum[subscript - vp->base_subscript_short];
+ const struct interact_params *iap = df_to_iap (cat, subscript);
+ return iap->enc_sum[subscript - iap->base_df];
}
+
/* Returns unity if the value in case C at SUBSCRIPT is equal to the category
for that subscript */
-double
+static double
categoricals_get_code_for_case (const struct categoricals *cat, int subscript,
- const struct ccase *c)
+ const struct ccase *c,
+ bool effects_coding)
{
- const struct interaction *iact = categoricals_get_interaction_by_subscript (cat, subscript);
+ const struct interaction *iact
+ = categoricals_get_interaction_by_subscript (cat, subscript);
- const int i = reverse_variable_lookup_short (cat, subscript);
+ const struct interact_params *iap = df_to_iap (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];
-
- double dfp = 1.0;
- for (v = 0; v < iact->n_vars; ++v)
+ int dfp = 1;
+ for (int 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;
+ const struct value_node *valn
+ = lookup_value (&iap->varnodes[v]->valmap, val, hash, width);
- /* 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];
+ const int df = hmap_count (&iap->varnodes[v]->valmap) - 1;
+ const int dfpn = dfp * df;
-#if EFFECTS_CODING
- if ( valn->index == df )
- bin = -1.0;
- else
-#endif
- if ( valn->index != index )
- bin = 0;
-
- result *= bin;
+ if (effects_coding && valn->index == df)
+ result = -result;
+ else
+ {
+ /* Translate subscript into an index for the individual variable. */
+ const int index = ((subscript - iap->base_df) % dfpn) / dfp;
+ if (valn->index != index)
+ return 0.0;
+ }
+ dfp = dfpn;
}
return result;
}
-size_t
-categoricals_get_n_variables (const struct categoricals *cat)
+/* Returns unity if the value in case C at SUBSCRIPT is equal to the category
+ for that subscript. */
+double
+categoricals_get_dummy_code_for_case (const struct categoricals *cat,
+ int subscript, const struct ccase *c)
{
- printf ("%s\n", __FUNCTION__);
- return cat->n_vars;
+ return categoricals_get_code_for_case (cat, subscript, c, false);
}
+/* Returns unity if the value in case C at SUBSCRIPT is equal to the category
+ for that subscript.
+ Else if it is the last category, return -1.
+ Otherwise return 0.
+ */
+double
+categoricals_get_effects_code_for_case (const struct categoricals *cat,
+ int subscript, const struct ccase *c)
+{
+ return categoricals_get_code_for_case (cat, subscript, c, true);
+}
-/* Return a case containing the set of values corresponding to
+/* 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)
+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;
+ const struct interact_params *iap = &cat->iap[iact];
+ return n < hmap_count (&iap->ivmap) ? iap->ivs[n]->ccase : NULL;
}
-/* Return a the user data corresponding to the Nth Category of the IACTth interaction. */
+/* 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)
+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;
+ const struct interact_params *iap = &cat->iap[iact];
+ return n < hmap_count (&iap->ivmap) ? iap->ivs[n]->user_data : NULL;
}
+int
+categoricals_get_value_index_by_category_real (const struct categoricals *cat,
+ int iact_idx, int cat_idx,
+ int var_idx)
+{
+ const struct interact_params *iap = &cat->iap[iact_idx];
+ const struct interaction_value *ivn = iap->ivs[cat_idx];
+ const struct variable *var = iap->iact->vars[var_idx];
+ const struct variable_node *vn = iap->varnodes[var_idx];
+ const union value *val = case_data (ivn->ccase, var);
+ int width = var_get_width (var);
+ unsigned int hash = value_hash (val, width, 0);
+ return lookup_value (&vn->valmap, val, hash, width)->index;
+}
-
-/* Return a case containing the set of values corresponding to SUBSCRIPT */
+/* Return a case containing the set of values corresponding to CAT_INDEX. */
const struct ccase *
-categoricals_get_case_by_category (const struct categoricals *cat, int subscript)
+categoricals_get_case_by_category (const struct categoricals *cat,
+ int cat_index)
{
- 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 interact_params *iap = cat_index_to_iap (cat, cat_index);
+ const struct interaction_value *vn = iap->ivs[cat_index - iap->base_cats];
return vn->ccase;
}
void *
-categoricals_get_user_data_by_category (const struct categoricals *cat, int subscript)
+categoricals_get_user_data_by_category (const struct categoricals *cat,
+ int cat_index)
{
- int vindex = reverse_variable_lookup_long (cat, subscript);
- const struct interact_params *vp = &cat->iap[vindex];
-
- const struct interaction_value *iv = vp->reverse_interaction_value_map [subscript - vp->base_subscript_long];
+ const struct interact_params *iap = cat_index_to_iap (cat, cat_index);
+ const struct interaction_value *iv = iap->ivs[cat_index - iap->base_cats];
return iv->user_data;
}
-
-
\f
-
void
categoricals_set_payload (struct categoricals *cat, const struct payload *p,
- const void *aux1, void *aux2)
+ const void *aux1, void *aux2)
{
cat->payload = p;
cat->aux1 = aux1;