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 variable *var;
52 const struct interaction *iact;
54 int base_subscript_short;
55 int base_subscript_long;
57 /* The number of distinct values of this variable */
60 /* A map of values indexed by subscript */
61 const struct value_node **reverse_value_map;
63 /* Total of the weights of this variable */
68 /* Comparison function to sort the reverse_value_map in ascending order */
70 compare_value_node (const void *vn1_, const void *vn2_, const void *aux)
72 const struct value_node * const *vn1 = vn1_;
73 const struct value_node * const *vn2 = vn2_;
74 const struct var_params *vp = aux;
76 return interaction_case_cmp_3way (vp->iact, (*vn1)->ccase, (*vn2)->ccase);
82 /* The weight variable */
83 const struct variable *wv;
86 /* An array of var_params */
87 struct var_params *vp;
89 /* The size of VP. (ie, the number of variables involved.) */
92 /* The number of categorical variables which contain entries.
93 In the absence of missing values, this will be equal to N_VP */
96 /* A map to enable the lookup of variables indexed by subscript.
97 This map considers only the N - 1 of the N variables.
99 int *reverse_variable_map_short;
101 /* Like the above, but uses all N variables */
102 int *reverse_variable_map_long;
108 /* Missing values to be excluded */
109 enum mv_class exclude;
111 /* Function to be called on each update */
114 /* Function specified by the caller to create user_data */
115 user_data_create_func *user_data_create;
117 /* Auxilliary data to be passed to update and user_data_create_func*/
124 categoricals_destroy ( struct categoricals *cat)
129 for (i = 0 ; i < cat->n_vp; ++i)
130 hmap_destroy (&cat->vp[i].map);
132 pool_destroy (cat->pool);
140 categoricals_dump (const struct categoricals *cat)
144 for (v = 0 ; v < cat->n_vp; ++v)
146 const struct var_params *vp = &cat->vp[v];
147 const struct hmap *m = &vp->map;
148 struct hmap_node *node ;
151 printf ("\n%s (%d) CC=%g n_cats=%d:\n",
152 var_get_name (vp->var), vp->base_subscript_long, vp->cc, vp->n_cats);
154 printf ("Reverse map\n");
155 for (x = 0 ; x < vp->n_cats; ++x)
158 const struct value_node *vn = vp->reverse_value_map[x];
160 var_append_value_name (vp->var, &vn->value, &s);
161 printf ("Value for %d is %s\n", x, ds_cstr(&s));
165 printf ("\nForward map\n");
166 for (node = hmap_first (m); node; node = hmap_next (m, node))
169 const struct value_node *vn = HMAP_DATA (node, struct value_node, node);
171 var_append_value_name (vp->var, &vn->value, &s);
172 printf ("Value: %s; Index %d; CC %g\n",
174 vn->subscript, vn->cc);
179 assert (cat->n_vars <= cat->n_vp);
182 printf ("Number of categorical variables: %d\n", cat->n_vp);
183 printf ("Number of non-empty categorical variables: %d\n", cat->n_vars);
184 printf ("Total number of categories: %d\n", cat->n_cats_total);
186 printf ("\nReverse variable map (short):\n");
187 for (v = 0 ; v < cat->n_cats_total - cat->n_vars; ++v)
188 printf ("%d ", cat->reverse_variable_map_short[v]);
190 printf ("\nReverse variable map (long):\n");
191 for (v = 0 ; v < cat->n_cats_total; ++v)
192 printf ("%d ", cat->reverse_variable_map_long[v]);
199 static struct value_node *
200 lookup_case (const struct hmap *map, const struct interaction *iact, const struct ccase *c)
202 struct value_node *nn;
203 size_t hash = interaction_case_hash (iact, c);
205 HMAP_FOR_EACH_WITH_HASH (nn, struct value_node, node, hash, map)
207 if (interaction_case_equal (iact, c, nn->ccase))
210 fprintf (stderr, "Warning: Hash table collision\n");
217 struct categoricals *
218 categoricals_create (const struct interaction **inter, size_t n_inter,
219 const struct variable *wv, enum mv_class exclude,
220 user_data_create_func *udf,
221 update_func *update, void *aux1, void *aux2
225 struct categoricals *cat = xmalloc (sizeof *cat);
229 cat->n_cats_total = 0;
231 cat->reverse_variable_map_short = NULL;
232 cat->reverse_variable_map_long = NULL;
233 cat->pool = pool_create ();
234 cat->exclude = exclude;
235 cat->update = update;
236 cat->user_data_create = udf;
242 cat->vp = pool_calloc (cat->pool, cat->n_vp, sizeof *cat->vp);
244 for (i = 0 ; i < cat->n_vp; ++i)
246 hmap_init (&cat->vp[i].map);
247 cat->vp[i].var = inter[i]->vars[0];
248 cat->vp[i].iact = inter[i];
257 categoricals_update (struct categoricals *cat, const struct ccase *c)
261 const double weight = cat->wv ? case_data (c, cat->wv)->f : 1.0;
263 assert (NULL == cat->reverse_variable_map_short);
264 assert (NULL == cat->reverse_variable_map_long);
266 for (i = 0 ; i < cat->n_vp; ++i)
268 const struct interaction *iact = cat->vp[i].iact;
269 const struct variable *var = cat->vp[i].var;
271 const union value *val = case_data (c, var);
273 struct value_node *node ;
276 if ( var_is_value_missing (var, val, cat->exclude))
280 hash = interaction_case_hash (iact, c);
281 node = lookup_case (&cat->vp[i].map, iact, c);
285 int width = var_get_width (var);
286 node = pool_malloc (cat->pool, sizeof *node);
288 node->ccase = case_ref (c);
291 hmap_insert (&cat->vp[i].map, &node->node, hash);
294 if ( 0 == cat->vp[i].n_cats)
297 node->subscript = cat->vp[i].n_cats++ ;
299 if (cat->user_data_create)
300 node->user_data = cat->user_data_create (cat->aux1, cat->aux2);
304 cat->vp[i].cc += weight;
307 cat->update (node->user_data, cat->exclude, cat->wv, var, c, cat->aux1, cat->aux2);
311 /* Return the number of categories (distinct values) for variable N */
313 categoricals_n_count (const struct categoricals *cat, size_t n)
315 return hmap_count (&cat->vp[n].map);
319 /* Return the total number of categories */
321 categoricals_total (const struct categoricals *cat)
323 return cat->n_cats_total;
327 /* This function must be called *before* any call to categoricals_get_*_by subscript and
328 *after* all calls to categoricals_update */
330 categoricals_done (const struct categoricals *cat_)
332 /* Implementation Note: Whilst this function is O(n) in cat->n_cats_total, in most
333 uses it will be more efficient that using a tree based structure, since it
334 is called only once, and means that subsequent lookups will be O(1).
336 1 call of O(n) + 10^9 calls of O(1) is better than 10^9 calls of O(log n).
338 struct categoricals *cat = CONST_CAST (struct categoricals *, cat_);
342 cat->reverse_variable_map_short = pool_calloc (cat->pool,
343 cat->n_cats_total - cat->n_vars,
344 sizeof *cat->reverse_variable_map_short);
346 cat->reverse_variable_map_long = pool_calloc (cat->pool,
348 sizeof *cat->reverse_variable_map_long);
350 for (v = 0 ; v < cat->n_vp; ++v)
353 struct var_params *vp = &cat->vp[v];
354 int n_cats_total = categoricals_n_count (cat, v);
355 struct hmap_node *node ;
357 vp->reverse_value_map = pool_calloc (cat->pool, n_cats_total, sizeof *vp->reverse_value_map);
359 vp->base_subscript_short = idx_short;
360 vp->base_subscript_long = idx_long;
362 for (node = hmap_first (&vp->map); node; node = hmap_next (&vp->map, node))
364 const struct value_node *vn = HMAP_DATA (node, struct value_node, node);
365 vp->reverse_value_map[vn->subscript] = vn;
368 /* For some purposes (eg CONTRASTS in ONEWAY) the values need to be sorted */
369 sort (vp->reverse_value_map, vp->n_cats, sizeof (const struct value_node *),
370 compare_value_node, vp);
372 /* Populate the reverse variable maps. */
373 for (i = 0; i < vp->n_cats - 1; ++i)
374 cat->reverse_variable_map_short[idx_short++] = v;
376 for (i = 0; i < vp->n_cats; ++i)
377 cat->reverse_variable_map_long[idx_long++] = v;
380 assert (cat->n_vars <= cat->n_vp);
385 reverse_variable_lookup_short (const struct categoricals *cat, int subscript)
387 assert (cat->reverse_variable_map_short);
388 assert (subscript >= 0);
389 assert (subscript < cat->n_cats_total - cat->n_vars);
391 return cat->reverse_variable_map_short[subscript];
395 reverse_variable_lookup_long (const struct categoricals *cat, int subscript)
397 assert (cat->reverse_variable_map_long);
398 assert (subscript >= 0);
399 assert (subscript < cat->n_cats_total);
401 return cat->reverse_variable_map_long[subscript];
406 /* Return the categorical variable corresponding to SUBSCRIPT */
407 const struct variable *
408 categoricals_get_variable_by_subscript (const struct categoricals *cat, int subscript)
410 int index = reverse_variable_lookup_short (cat, subscript);
412 return cat->vp[index].var;
415 /* Return the interaction corresponding to SUBSCRIPT */
416 static const struct interaction *
417 categoricals_get_interaction_by_subscript (const struct categoricals *cat, int subscript)
419 int index = reverse_variable_lookup_short (cat, subscript);
421 return cat->vp[index].iact;
425 /* Return the case corresponding to SUBSCRIPT */
426 static const struct ccase *
427 categoricals_get_case_by_subscript (const struct categoricals *cat, int subscript)
429 int vindex = reverse_variable_lookup_short (cat, subscript);
430 const struct var_params *vp = &cat->vp[vindex];
431 const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript_short];
438 categoricals_get_weight_by_subscript (const struct categoricals *cat, int subscript)
440 int vindex = reverse_variable_lookup_short (cat, subscript);
441 const struct var_params *vp = &cat->vp[vindex];
447 categoricals_get_sum_by_subscript (const struct categoricals *cat, int subscript)
449 int vindex = reverse_variable_lookup_short (cat, subscript);
450 const struct var_params *vp = &cat->vp[vindex];
452 const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript_short];
457 /* Returns unity if the value in case C at SUBSCRIPT is equal to the category
458 for that subscript */
460 categoricals_get_binary_by_subscript (const struct categoricals *cat, int subscript,
461 const struct ccase *c)
463 const struct interaction *iact = categoricals_get_interaction_by_subscript (cat, subscript);
465 const struct ccase *c2 = categoricals_get_case_by_subscript (cat, subscript);
467 return interaction_case_equal (iact, c, c2);
472 categoricals_get_n_variables (const struct categoricals *cat)
478 /* Return a case containing the set of values corresponding to SUBSCRIPT */
480 categoricals_get_case_by_category (const struct categoricals *cat, int subscript)
482 int vindex = reverse_variable_lookup_long (cat, subscript);
483 const struct var_params *vp = &cat->vp[vindex];
484 const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript_long];
491 categoricals_get_user_data_by_category (const struct categoricals *cat, int subscript)
493 int vindex = reverse_variable_lookup_long (cat, subscript);
494 const struct var_params *vp = &cat->vp[vindex];
496 const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript_long];
497 return vn->user_data;