1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 2009 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <http://www.gnu.org/licenses/>. */
21 #include "categoricals.h"
23 #include <gl/xalloc.h>
24 #include <data/variable.h>
25 #include <data/case.h>
26 #include <data/value.h>
27 #include <libpspp/hmap.h>
28 #include <libpspp/pool.h>
30 #include <libpspp/str.h>
34 struct hmap_node node; /* Node in hash map. */
35 union value value; /* The value being labeled. */
36 double cc; /* The total of the weights of cases with this value */
37 int subscript; /* A zero based integer, unique within the variable.
38 Can be used as an index into an array */
44 /* A map indexed by a union values */
47 const struct variable *var;
51 /* The number of distinct values of this variable */
54 /* A map of values indexed by subscript */
55 const struct value_node **reverse_value_map;
57 /* Total of the weights of this variable */
64 /* The weight variable */
65 const struct variable *wv;
68 /* An array of var_params */
69 struct var_params *vp;
71 /* The size of VP. (ie, the number of variables involved.) */
74 /* The number of categorical variables which contain entries.
75 In the absence of missing values, this will be equal to N_VP */
78 /* A map to enable the lookup of variables indexed by subscript */
79 int *reverse_variable_map;
85 /* Missing values to be excluded */
86 enum mv_class exclude;
91 categoricals_destroy ( struct categoricals *cat)
94 for (i = 0 ; i < cat->n_vp; ++i)
95 hmap_destroy (&cat->vp[i].map);
97 pool_destroy (cat->pool);
103 categoricals_dump (const struct categoricals *cat)
107 for (v = 0 ; v < cat->n_vp; ++v)
109 const struct var_params *vp = &cat->vp[v];
110 const struct hmap *m = &vp->map;
111 struct hmap_node *node ;
114 printf ("\n%s (%d) CC=%g n_cats=%d:\n",
115 var_get_name (vp->var), vp->base_subscript, vp->cc, vp->n_cats);
117 printf ("Reverse map\n");
118 for (x = 0 ; x < vp->n_cats; ++x)
121 const struct value_node *vn = vp->reverse_value_map[x];
123 var_append_value_name (vp->var, &vn->value, &s);
124 printf ("Value for %d is %s\n", x, ds_cstr(&s));
128 printf ("\nForward map\n");
129 for (node = hmap_first (m); node; node = hmap_next (m, node))
133 const struct value_node *vn = HMAP_DATA (node, struct value_node, node);
134 var_append_value_name (vp->var, &vn->value, &s);
135 printf ("Value: %s; Index %d; CC %g\n",
137 vn->subscript, vn->cc);
142 assert (cat->n_vars <= cat->n_vp);
145 printf ("Number of categorical variables: %d\n", cat->n_vp);
146 printf ("Number of non-empty categorical variables: %d\n", cat->n_vars);
147 printf ("Total number of categories: %d\n", cat->n_cats_total);
152 static struct value_node *
153 lookup_value (const struct hmap *map, const struct variable *var, const union value *val)
155 struct value_node *foo;
156 unsigned int width = var_get_width (var);
157 size_t hash = value_hash (val, width, 0);
159 HMAP_FOR_EACH_WITH_HASH (foo, struct value_node, node, hash, map)
161 if (value_equal (val, &foo->value, width))
164 fprintf (stderr, "Warning: Hash table collision\n");
172 struct categoricals *
173 categoricals_create (const struct variable **v, size_t n_vars,
174 const struct variable *wv, enum mv_class exclude)
177 struct categoricals *cat = xmalloc (sizeof *cat);
181 cat->n_cats_total = 0;
183 cat->reverse_variable_map = NULL;
184 cat->pool = pool_create ();
185 cat->exclude = exclude;
187 cat->vp = pool_calloc (cat->pool, cat->n_vp, sizeof *cat->vp);
189 for (i = 0 ; i < cat->n_vp; ++i)
191 hmap_init (&cat->vp[i].map);
192 cat->vp[i].var = v[i];
201 categoricals_update (struct categoricals *cat, const struct ccase *c)
205 const double weight = cat->wv ? case_data (c, cat->wv)->f : 1.0;
207 assert (NULL == cat->reverse_variable_map);
209 for (i = 0 ; i < cat->n_vp; ++i)
211 const struct variable *var = cat->vp[i].var;
212 unsigned int width = var_get_width (var);
213 const union value *val = case_data (c, var);
215 struct value_node *node ;
217 if ( var_is_value_missing (var, val, cat->exclude))
220 hash = value_hash (val, width, 0);
221 node = lookup_value (&cat->vp[i].map, var, val);
225 node = pool_malloc (cat->pool, sizeof *node);
227 value_init (&node->value, width);
228 value_copy (&node->value, val, width);
231 hmap_insert (&cat->vp[i].map, &node->node, hash);
234 if ( 0 == cat->vp[i].n_cats)
237 node->subscript = cat->vp[i].n_cats++ ;
241 cat->vp[i].cc += weight;
245 /* Return the number of categories (distinct values) for variable N */
247 categoricals_n_count (const struct categoricals *cat, size_t n)
249 return hmap_count (&cat->vp[n].map);
253 /* Return the index for value VAL in the Nth variable */
255 categoricals_index (const struct categoricals *cat, size_t n, const union value *val)
257 struct value_node *vn = lookup_value (&cat->vp[n].map, cat->vp[n].var, val);
262 return vn->subscript;
266 /* Return the total number of categories */
268 categoricals_total (const struct categoricals *cat)
270 return cat->n_cats_total;
274 /* This function must be called *before* any call to categoricals_get_*_by subscript an
275 *after* all calls to categoricals_update */
277 categoricals_done (struct categoricals *cat)
279 /* Implementation Note: Whilst this function is O(n) in cat->n_cats_total, in most
280 uses it will be more efficient that using a tree based structure, since it
281 is called only once, and means that subsequent lookups will be O(1).
283 1 call of O(n) + 10^9 calls of O(1) is better than 10^9 calls of O(log n).
287 cat->reverse_variable_map = pool_calloc (cat->pool, cat->n_cats_total, sizeof *cat->reverse_variable_map);
289 for (v = 0 ; v < cat->n_vp; ++v)
292 struct var_params *vp = &cat->vp[v];
293 int n_cats_total = categoricals_n_count (cat, v);
294 struct hmap_node *node ;
296 vp->reverse_value_map = pool_calloc (cat->pool, n_cats_total, sizeof *vp->reverse_value_map);
298 vp->base_subscript = idx;
300 for (node = hmap_first (&vp->map); node; node = hmap_next (&vp->map, node))
302 const struct value_node *vn = HMAP_DATA (node, struct value_node, node);
303 vp->reverse_value_map[vn->subscript] = vn;
306 for (i = 0; i < vp->n_cats; ++i)
307 cat->reverse_variable_map[idx++] = v;
310 assert (cat->n_vars <= cat->n_vp);
315 /* Return the categorical variable corresponding to SUBSCRIPT */
316 const struct variable *
317 categoricals_get_variable_by_subscript (const struct categoricals *cat, int subscript)
321 assert (cat->reverse_variable_map);
323 index = cat->reverse_variable_map[subscript];
325 return cat->vp[index].var;
329 /* Return the value corresponding to SUBSCRIPT */
331 categoricals_get_value_by_subscript (const struct categoricals *cat, int subscript)
333 int vindex = cat->reverse_variable_map[subscript];
334 const struct var_params *vp = &cat->vp[vindex];
335 const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript];
342 categoricals_get_weight_by_subscript (const struct categoricals *cat, int subscript)
344 int vindex = cat->reverse_variable_map[subscript];
345 const struct var_params *vp = &cat->vp[vindex];
351 categoricals_get_sum_by_subscript (const struct categoricals *cat, int subscript)
353 int vindex = cat->reverse_variable_map[subscript];
354 const struct var_params *vp = &cat->vp[vindex];
356 const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript];
361 /* Returns unity if the value in case C at SUBSCRIPT is equal to the category
362 for that subscript */
364 categoricals_get_binary_by_subscript (const struct categoricals *cat, int subscript,
365 const struct ccase *c)
367 const struct variable *var = categoricals_get_variable_by_subscript (cat, subscript);
368 int width = var_get_width (var);
370 const union value *val = case_data (c, var);
372 return value_equal (val, categoricals_get_value_by_subscript (cat, subscript), width);
377 categoricals_get_n_variables (const struct categoricals *cat)