1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 2009, 2010, 2011 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/>. */
19 #include "math/categoricals.h"
20 #include "math/interaction.h"
24 #include "data/case.h"
25 #include "data/value.h"
26 #include "data/variable.h"
27 #include "libpspp/array.h"
28 #include "libpspp/hmap.h"
29 #include "libpspp/pool.h"
30 #include "libpspp/str.h"
32 #include "gl/xalloc.h"
36 struct hmap_node node; /* Node in hash map. */
38 double cc; /* The total of the weights of cases with this value */
40 void *user_data; /* A pointer to data which the caller can store stuff */
42 int subscript; /* A zero based integer, unique within the variable.
43 Can be used as an index into an array */
48 /* A map indexed by a union values */
51 const struct interaction *iact;
53 int base_subscript_short;
54 int base_subscript_long;
56 /* The number of distinct values of this variable */
59 /* A map of values indexed by subscript */
60 const struct value_node **reverse_value_map;
62 /* Total of the weights of this variable */
67 /* Comparison function to sort the reverse_value_map in ascending order */
69 compare_value_node (const void *vn1_, const void *vn2_, const void *aux)
71 const struct value_node * const *vn1 = vn1_;
72 const struct value_node * const *vn2 = vn2_;
73 const struct var_params *vp = aux;
75 return interaction_case_cmp_3way (vp->iact, (*vn1)->ccase, (*vn2)->ccase);
81 /* The weight variable */
82 const struct variable *wv;
84 /* An array of var_params */
85 struct var_params *vp;
87 /* The size of VP. (ie, the number of variables involved.) */
90 /* The number of categorical variables which contain entries.
91 In the absence of missing values, this will be equal to N_VP */
94 /* A map to enable the lookup of variables indexed by subscript.
95 This map considers only the N - 1 of the N variables.
97 int *reverse_variable_map_short;
99 /* Like the above, but uses all N variables */
100 int *reverse_variable_map_long;
106 /* Missing values to be excluded */
107 enum mv_class exclude;
109 /* Function to be called on each update */
112 /* Function specified by the caller to create user_data */
113 user_data_create_func *user_data_create;
115 /* Auxilliary data to be passed to update and user_data_create_func*/
122 categoricals_destroy ( struct categoricals *cat)
127 for (i = 0 ; i < cat->n_vp; ++i)
129 struct hmap *map = &cat->vp[i].map;
130 struct value_node *nn;
132 HMAP_FOR_EACH (nn, struct value_node, node, map)
134 case_unref (nn->ccase);
140 pool_destroy (cat->pool);
148 categoricals_dump (const struct categoricals *cat)
152 for (v = 0 ; v < cat->n_vp; ++v)
154 const struct var_params *vp = &cat->vp[v];
155 const struct hmap *m = &vp->map;
156 struct hmap_node *node ;
159 printf ("\n%s (%d) CC=%g n_cats=%d:\n",
160 var_get_name (vp->var), vp->base_subscript_long, vp->cc, vp->n_cats);
162 printf ("Reverse map\n");
163 for (x = 0 ; x < vp->n_cats; ++x)
166 const struct value_node *vn = vp->reverse_value_map[x];
168 var_append_value_name (vp->var, &vn->value, &s);
169 printf ("Value for %d is %s\n", x, ds_cstr(&s));
173 printf ("\nForward map\n");
174 for (node = hmap_first (m); node; node = hmap_next (m, node))
177 const struct value_node *vn = HMAP_DATA (node, struct value_node, node);
179 var_append_value_name (vp->var, &vn->value, &s);
180 printf ("Value: %s; Index %d; CC %g\n",
182 vn->subscript, vn->cc);
187 assert (cat->n_vars <= cat->n_vp);
190 printf ("Number of categorical variables: %d\n", cat->n_vp);
191 printf ("Number of non-empty categorical variables: %d\n", cat->n_vars);
192 printf ("Total number of categories: %d\n", cat->n_cats_total);
194 printf ("\nReverse variable map (short):\n");
195 for (v = 0 ; v < cat->n_cats_total - cat->n_vars; ++v)
196 printf ("%d ", cat->reverse_variable_map_short[v]);
198 printf ("\nReverse variable map (long):\n");
199 for (v = 0 ; v < cat->n_cats_total; ++v)
200 printf ("%d ", cat->reverse_variable_map_long[v]);
207 static struct value_node *
208 lookup_case (const struct hmap *map, const struct interaction *iact, const struct ccase *c)
210 struct value_node *nn;
211 size_t hash = interaction_case_hash (iact, c);
213 HMAP_FOR_EACH_WITH_HASH (nn, struct value_node, node, hash, map)
215 if (interaction_case_equal (iact, c, nn->ccase))
218 fprintf (stderr, "Warning: Hash table collision\n");
225 struct categoricals *
226 categoricals_create (const struct interaction **inter, size_t n_inter,
227 const struct variable *wv, enum mv_class exclude,
228 user_data_create_func *udf,
229 update_func *update, void *aux1, void *aux2
233 struct categoricals *cat = xmalloc (sizeof *cat);
237 cat->n_cats_total = 0;
239 cat->reverse_variable_map_short = NULL;
240 cat->reverse_variable_map_long = NULL;
241 cat->pool = pool_create ();
242 cat->exclude = exclude;
243 cat->update = update;
244 cat->user_data_create = udf;
250 cat->vp = pool_calloc (cat->pool, cat->n_vp, sizeof *cat->vp);
252 for (i = 0 ; i < cat->n_vp; ++i)
254 hmap_init (&cat->vp[i].map);
255 cat->vp[i].iact = inter[i];
264 categoricals_update (struct categoricals *cat, const struct ccase *c)
268 const double weight = cat->wv ? case_data (c, cat->wv)->f : 1.0;
270 assert (NULL == cat->reverse_variable_map_short);
271 assert (NULL == cat->reverse_variable_map_long);
273 for (i = 0 ; i < cat->n_vp; ++i)
275 const struct interaction *iact = cat->vp[i].iact;
277 struct value_node *node ;
279 if ( interaction_case_is_missing (iact, c, cat->exclude))
282 hash = interaction_case_hash (iact, c);
283 node = lookup_case (&cat->vp[i].map, iact, c);
287 node = pool_malloc (cat->pool, sizeof *node);
289 node->ccase = case_ref (c);
292 hmap_insert (&cat->vp[i].map, &node->node, hash);
295 if ( 0 == cat->vp[i].n_cats)
298 node->subscript = cat->vp[i].n_cats++ ;
300 if (cat->user_data_create)
301 node->user_data = cat->user_data_create (cat->aux1, cat->aux2);
305 cat->vp[i].cc += weight;
308 cat->update (node->user_data, cat->exclude, cat->wv, NULL, c, cat->aux1, cat->aux2);
312 /* Return the number of categories (distinct values) for variable N */
314 categoricals_n_count (const struct categoricals *cat, size_t n)
316 return hmap_count (&cat->vp[n].map);
320 /* Return the total number of categories */
322 categoricals_total (const struct categoricals *cat)
324 return cat->n_cats_total;
328 /* This function must be called *before* any call to categoricals_get_*_by subscript and
329 *after* all calls to categoricals_update */
331 categoricals_done (const struct categoricals *cat_)
333 /* Implementation Note: Whilst this function is O(n) in cat->n_cats_total, in most
334 uses it will be more efficient that using a tree based structure, since it
335 is called only once, and means that subsequent lookups will be O(1).
337 1 call of O(n) + 10^9 calls of O(1) is better than 10^9 calls of O(log n).
339 struct categoricals *cat = CONST_CAST (struct categoricals *, cat_);
343 cat->reverse_variable_map_short = pool_calloc (cat->pool,
344 cat->n_cats_total - cat->n_vars,
345 sizeof *cat->reverse_variable_map_short);
347 cat->reverse_variable_map_long = pool_calloc (cat->pool,
349 sizeof *cat->reverse_variable_map_long);
351 for (v = 0 ; v < cat->n_vp; ++v)
354 struct var_params *vp = &cat->vp[v];
355 int n_cats_total = categoricals_n_count (cat, v);
356 struct hmap_node *node ;
358 vp->reverse_value_map = pool_calloc (cat->pool, n_cats_total, sizeof *vp->reverse_value_map);
360 vp->base_subscript_short = idx_short;
361 vp->base_subscript_long = idx_long;
363 for (node = hmap_first (&vp->map); node; node = hmap_next (&vp->map, node))
365 const struct value_node *vn = HMAP_DATA (node, struct value_node, node);
366 vp->reverse_value_map[vn->subscript] = vn;
369 /* For some purposes (eg CONTRASTS in ONEWAY) the values need to be sorted */
370 sort (vp->reverse_value_map, vp->n_cats, sizeof (const struct value_node *),
371 compare_value_node, vp);
373 /* Populate the reverse variable maps. */
374 for (i = 0; i < vp->n_cats - 1; ++i)
375 cat->reverse_variable_map_short[idx_short++] = v;
377 for (i = 0; i < vp->n_cats; ++i)
378 cat->reverse_variable_map_long[idx_long++] = v;
381 assert (cat->n_vars <= cat->n_vp);
386 reverse_variable_lookup_short (const struct categoricals *cat, int subscript)
388 assert (cat->reverse_variable_map_short);
389 assert (subscript >= 0);
390 assert (subscript < cat->n_cats_total - cat->n_vars);
392 return cat->reverse_variable_map_short[subscript];
396 reverse_variable_lookup_long (const struct categoricals *cat, int subscript)
398 assert (cat->reverse_variable_map_long);
399 assert (subscript >= 0);
400 assert (subscript < cat->n_cats_total);
402 return cat->reverse_variable_map_long[subscript];
407 /* Return the categorical variable corresponding to SUBSCRIPT */
408 const struct variable *
409 categoricals_get_variable_by_subscript (const struct categoricals *cat, int subscript)
411 int index = reverse_variable_lookup_short (cat, subscript);
413 return cat->vp[index].iact->vars[0];
416 /* Return the interaction corresponding to SUBSCRIPT */
417 const struct interaction *
418 categoricals_get_interaction_by_subscript (const struct categoricals *cat, int subscript)
420 int index = reverse_variable_lookup_short (cat, subscript);
422 return cat->vp[index].iact;
426 /* Return the case corresponding to SUBSCRIPT */
427 static const struct ccase *
428 categoricals_get_case_by_subscript (const struct categoricals *cat, int subscript)
430 int vindex = reverse_variable_lookup_short (cat, subscript);
431 const struct var_params *vp = &cat->vp[vindex];
432 const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript_short];
439 categoricals_get_weight_by_subscript (const struct categoricals *cat, int subscript)
441 int vindex = reverse_variable_lookup_short (cat, subscript);
442 const struct var_params *vp = &cat->vp[vindex];
448 categoricals_get_sum_by_subscript (const struct categoricals *cat, int subscript)
450 int vindex = reverse_variable_lookup_short (cat, subscript);
451 const struct var_params *vp = &cat->vp[vindex];
453 const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript_short];
458 /* Returns unity if the value in case C at SUBSCRIPT is equal to the category
459 for that subscript */
461 categoricals_get_binary_by_subscript (const struct categoricals *cat, int subscript,
462 const struct ccase *c)
464 const struct interaction *iact = categoricals_get_interaction_by_subscript (cat, subscript);
466 const struct ccase *c2 = categoricals_get_case_by_subscript (cat, subscript);
468 return interaction_case_equal (iact, c, c2);
473 categoricals_get_n_variables (const struct categoricals *cat)
479 /* Return a case containing the set of values corresponding to SUBSCRIPT */
481 categoricals_get_case_by_category (const struct categoricals *cat, int subscript)
483 int vindex = reverse_variable_lookup_long (cat, subscript);
484 const struct var_params *vp = &cat->vp[vindex];
485 const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript_long];
492 categoricals_get_user_data_by_category (const struct categoricals *cat, int subscript)
494 int vindex = reverse_variable_lookup_long (cat, subscript);
495 const struct var_params *vp = &cat->vp[vindex];
497 const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript_long];
498 return vn->user_data;