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)
96 for (i = 0 ; i < cat->n_vp; ++i)
97 hmap_destroy (&cat->vp[i].map);
99 pool_destroy (cat->pool);
106 categoricals_dump (const struct categoricals *cat)
110 for (v = 0 ; v < cat->n_vp; ++v)
112 const struct var_params *vp = &cat->vp[v];
113 const struct hmap *m = &vp->map;
114 struct hmap_node *node ;
117 printf ("\n%s (%d) CC=%g n_cats=%d:\n",
118 var_get_name (vp->var), vp->base_subscript, vp->cc, vp->n_cats);
120 printf ("Reverse map\n");
121 for (x = 0 ; x < vp->n_cats; ++x)
124 const struct value_node *vn = vp->reverse_value_map[x];
126 var_append_value_name (vp->var, &vn->value, &s);
127 printf ("Value for %d is %s\n", x, ds_cstr(&s));
131 printf ("\nForward map\n");
132 for (node = hmap_first (m); node; node = hmap_next (m, node))
136 const struct value_node *vn = HMAP_DATA (node, struct value_node, node);
137 var_append_value_name (vp->var, &vn->value, &s);
138 printf ("Value: %s; Index %d; CC %g\n",
140 vn->subscript, vn->cc);
145 assert (cat->n_vars <= cat->n_vp);
148 printf ("Number of categorical variables: %d\n", cat->n_vp);
149 printf ("Number of non-empty categorical variables: %d\n", cat->n_vars);
150 printf ("Total number of categories: %d\n", cat->n_cats_total);
152 printf ("\nReverse variable map:\n");
154 for (v = 0 ; v < cat->n_cats_total - cat->n_vars; ++v)
155 printf ("%d ", cat->reverse_variable_map[v]);
161 static struct value_node *
162 lookup_value (const struct hmap *map, const struct variable *var, const union value *val)
164 struct value_node *foo;
165 unsigned int width = var_get_width (var);
166 size_t hash = value_hash (val, width, 0);
168 HMAP_FOR_EACH_WITH_HASH (foo, struct value_node, node, hash, map)
170 if (value_equal (val, &foo->value, width))
173 fprintf (stderr, "Warning: Hash table collision\n");
181 struct categoricals *
182 categoricals_create (const struct variable **v, size_t n_vars,
183 const struct variable *wv, enum mv_class exclude)
186 struct categoricals *cat = xmalloc (sizeof *cat);
190 cat->n_cats_total = 0;
192 cat->reverse_variable_map = NULL;
193 cat->pool = pool_create ();
194 cat->exclude = exclude;
196 cat->vp = pool_calloc (cat->pool, cat->n_vp, sizeof *cat->vp);
198 for (i = 0 ; i < cat->n_vp; ++i)
200 hmap_init (&cat->vp[i].map);
201 cat->vp[i].var = v[i];
210 categoricals_update (struct categoricals *cat, const struct ccase *c)
214 const double weight = cat->wv ? case_data (c, cat->wv)->f : 1.0;
216 assert (NULL == cat->reverse_variable_map);
218 for (i = 0 ; i < cat->n_vp; ++i)
220 const struct variable *var = cat->vp[i].var;
221 unsigned int width = var_get_width (var);
222 const union value *val = case_data (c, var);
224 struct value_node *node ;
226 if ( var_is_value_missing (var, val, cat->exclude))
229 hash = value_hash (val, width, 0);
230 node = lookup_value (&cat->vp[i].map, var, val);
234 node = pool_malloc (cat->pool, sizeof *node);
236 value_init (&node->value, width);
237 value_copy (&node->value, val, width);
240 hmap_insert (&cat->vp[i].map, &node->node, hash);
243 if ( 0 == cat->vp[i].n_cats)
246 node->subscript = cat->vp[i].n_cats++ ;
250 cat->vp[i].cc += weight;
254 /* Return the number of categories (distinct values) for variable N */
256 categoricals_n_count (const struct categoricals *cat, size_t n)
258 return hmap_count (&cat->vp[n].map);
262 /* Return the total number of categories */
264 categoricals_total (const struct categoricals *cat)
266 return cat->n_cats_total;
270 /* This function must be called *before* any call to categoricals_get_*_by subscript an
271 *after* all calls to categoricals_update */
273 categoricals_done (struct categoricals *cat)
275 /* Implementation Note: Whilst this function is O(n) in cat->n_cats_total, in most
276 uses it will be more efficient that using a tree based structure, since it
277 is called only once, and means that subsequent lookups will be O(1).
279 1 call of O(n) + 10^9 calls of O(1) is better than 10^9 calls of O(log n).
283 cat->reverse_variable_map = pool_calloc (cat->pool,
284 cat->n_cats_total - cat->n_vars,
285 sizeof *cat->reverse_variable_map);
287 for (v = 0 ; v < cat->n_vp; ++v)
290 struct var_params *vp = &cat->vp[v];
291 int n_cats_total = categoricals_n_count (cat, v);
292 struct hmap_node *node ;
294 vp->reverse_value_map = pool_calloc (cat->pool, n_cats_total, sizeof *vp->reverse_value_map);
296 vp->base_subscript = idx;
298 for (node = hmap_first (&vp->map); node; node = hmap_next (&vp->map, node))
300 const struct value_node *vn = HMAP_DATA (node, struct value_node, node);
301 vp->reverse_value_map[vn->subscript] = vn;
304 /* Populate the reverse variable map.
305 This implementation considers the first value of each categorical variable
306 as the basis. Therefore, this loop starts from 1 instead of 0 */
307 for (i = 1; i < vp->n_cats; ++i)
308 cat->reverse_variable_map[idx++] = v;
311 assert (cat->n_vars <= cat->n_vp);
316 reverse_variable_lookup (const struct categoricals *cat, int subscript)
318 assert (cat->reverse_variable_map);
319 assert (subscript >= 0);
320 assert (subscript < cat->n_cats_total - cat->n_vars);
322 return cat->reverse_variable_map[subscript];
326 /* Return the categorical variable corresponding to SUBSCRIPT */
327 const struct variable *
328 categoricals_get_variable_by_subscript (const struct categoricals *cat, int subscript)
330 int index = reverse_variable_lookup (cat, subscript);
332 return cat->vp[index].var;
335 /* Return the value corresponding to SUBSCRIPT */
337 categoricals_get_value_by_subscript (const struct categoricals *cat, int subscript)
339 int vindex = reverse_variable_lookup (cat, subscript);
340 const struct var_params *vp = &cat->vp[vindex];
341 const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript];
348 categoricals_get_weight_by_subscript (const struct categoricals *cat, int subscript)
350 int vindex = reverse_variable_lookup (cat, subscript);
351 const struct var_params *vp = &cat->vp[vindex];
357 categoricals_get_sum_by_subscript (const struct categoricals *cat, int subscript)
359 int vindex = reverse_variable_lookup (cat, subscript);
360 const struct var_params *vp = &cat->vp[vindex];
362 const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript];
367 /* Returns unity if the value in case C at SUBSCRIPT is equal to the category
368 for that subscript */
370 categoricals_get_binary_by_subscript (const struct categoricals *cat, int subscript,
371 const struct ccase *c)
373 const struct variable *var = categoricals_get_variable_by_subscript (cat, subscript);
374 int width = var_get_width (var);
376 const union value *val = case_data (c, var);
378 return value_equal (val, categoricals_get_value_by_subscript (cat, subscript), width);
383 categoricals_get_n_variables (const struct categoricals *cat)