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"
31 #include "libpspp/hash-functions.h"
33 #include "gl/xalloc.h"
37 struct hmap_node node; /* Node in hash map. */
39 union value val; /* The value */
43 struct interaction_value
45 struct hmap_node node; /* Node in hash map. */
47 struct ccase *ccase; /* A case (probably the first in the dataset) which matches this value */
49 /* Total of the weights of cases matching this interaction */
52 void *user_data; /* A pointer to data which the caller can store stuff */
55 static struct value_node *
56 lookup_value (const struct hmap *map, const union value *val, unsigned int hash, int width)
58 struct value_node *vn = NULL;
59 HMAP_FOR_EACH_WITH_HASH (vn, struct value_node, node, hash, map)
61 if (value_equal (&vn->val, val, width))
70 struct hmap_node node; /* Node in hash map. */
71 const struct variable *var; /* The variable */
73 struct hmap valmap; /* A map of value nodes */
78 dump_interaction (const struct interaction *iact)
80 struct string str = DS_EMPTY_INITIALIZER;
81 interaction_to_string (iact, &str);
82 printf ("Interaction: %s\n", ds_cstr (&str));
88 static struct variable_node *
89 lookup_variable (const struct hmap *map, const struct variable *var, unsigned int hash)
91 struct variable_node *vn = NULL;
92 HMAP_FOR_EACH_WITH_HASH (vn, struct variable_node, node, hash, map)
97 fprintf (stderr, "Warning: Hash table collision\n");
104 struct interact_params
106 /* A map indexed by a interaction_value */
109 const struct interaction *iact;
111 int base_subscript_short;
112 int base_subscript_long;
114 /* The number of distinct values of this interaction */
117 /* The degrees of freedom for this interaction */
120 /* A map of interaction_values indexed by subscript */
121 struct interaction_value **reverse_interaction_value_map;
127 /* Comparison function to sort the reverse_value_map in ascending order */
129 compare_interaction_value_3way (const void *vn1_, const void *vn2_, const void *aux)
131 const struct interaction_value *const *vn1p = vn1_;
132 const struct interaction_value *const *vn2p = vn2_;
134 const struct interact_params *iap = aux;
136 return interaction_case_cmp_3way (iap->iact, (*vn1p)->ccase, (*vn2p)->ccase);
141 /* The weight variable */
142 const struct variable *wv;
144 /* An array of interact_params */
145 struct interact_params *iap;
147 /* Map whose members are the union of the variables which comprise IAP */
150 /* The size of IAP. (ie, the number of interactions involved.) */
153 /* The number of categorical variables which contain entries.
154 In the absence of missing values, this will be equal to N_IAP */
159 /* A map to enable the lookup of variables indexed by subscript.
160 This map considers only the N - 1 of the N variables.
162 int *reverse_variable_map_short;
164 /* Like the above, but uses all N variables */
165 int *reverse_variable_map_long;
171 /* Missing values to be excluded */
172 enum mv_class exclude;
174 /* Function to be called on each update */
177 /* Function specified by the caller to create user_data */
178 user_data_create_func *user_data_create;
180 /* Auxilliary data to be passed to update and user_data_create_func*/
187 categoricals_dump (const struct categoricals *cat)
191 printf ("Reverse Variable Map (short):\n");
192 for (i = 0; i < cat->df_sum; ++i)
194 printf (" %d", cat->reverse_variable_map_short[i]);
198 printf ("Reverse Variable Map (long):\n");
199 for (i = 0; i < cat->n_cats_total; ++i)
201 printf (" %d", cat->reverse_variable_map_long[i]);
206 printf ("Number of interactions %d\n", cat->n_iap);
207 for (i = 0 ; i < cat->n_iap; ++i)
211 const struct interact_params *iap = &cat->iap[i];
212 const struct interaction *iact = iap->iact;
214 ds_init_empty (&str);
215 interaction_to_string (iact, &str);
217 printf ("\nInteraction: %s (n: %d; df: %d ); ", ds_cstr (&str), iap->n_cats, iap->df);
219 printf ("Base subscript: %d\n", iap->base_subscript_short);
222 for (v = 0; v < hmap_count (&iap->ivmap); ++v)
225 const struct interaction_value *iv = iap->reverse_interaction_value_map[v];
227 if (v > 0) printf (" ");
229 for (vv = 0; vv < iact->n_vars; ++vv)
231 const struct variable *var = iact->vars[vv];
232 const union value *val = case_data (iv->ccase, var);
234 printf ("%g", val->f);
235 if (vv < iact->n_vars - 1)
246 categoricals_destroy (struct categoricals *cat)
248 struct variable_node *vn = NULL;
252 for (i = 0; i < cat->n_iap; ++i)
254 struct interaction_value *iv = NULL;
255 /* Interate over each interaction value, and unref any cases that we reffed */
256 HMAP_FOR_EACH (iv, struct interaction_value, node, &cat->iap[i].ivmap)
258 case_unref (iv->ccase);
260 hmap_destroy (&cat->iap[i].ivmap);
263 /* Interate over each variable and delete its value map */
264 HMAP_FOR_EACH (vn, struct variable_node, node, &cat->varmap)
266 hmap_destroy (&vn->valmap);
269 hmap_destroy (&cat->varmap);
271 pool_destroy (cat->pool);
278 static struct interaction_value *
279 lookup_case (const struct hmap *map, const struct interaction *iact, const struct ccase *c)
281 struct interaction_value *iv = NULL;
282 size_t hash = interaction_case_hash (iact, c, 0);
284 HMAP_FOR_EACH_WITH_HASH (iv, struct interaction_value, node, hash, map)
286 if (interaction_case_equal (iact, c, iv->ccase))
289 fprintf (stderr, "Warning: Hash table collision\n");
296 struct categoricals *
297 categoricals_create (struct interaction *const*inter, size_t n_inter,
298 const struct variable *wv, enum mv_class exclude,
299 user_data_create_func *udf,
300 update_func *update, void *aux1, void *aux2
304 struct categoricals *cat = xmalloc (sizeof *cat);
306 cat->n_iap = n_inter;
308 cat->n_cats_total = 0;
310 cat->reverse_variable_map_short = NULL;
311 cat->reverse_variable_map_long = NULL;
312 cat->pool = pool_create ();
313 cat->exclude = exclude;
314 cat->update = update;
315 cat->user_data_create = udf;
320 cat->iap = pool_calloc (cat->pool, cat->n_iap, sizeof *cat->iap);
322 hmap_init (&cat->varmap);
323 for (i = 0 ; i < cat->n_iap; ++i)
326 hmap_init (&cat->iap[i].ivmap);
327 cat->iap[i].iact = inter[i];
328 cat->iap[i].cc = 0.0;
329 for (v = 0; v < inter[i]->n_vars; ++v)
331 const struct variable *var = inter[i]->vars[v];
332 unsigned int hash = hash_pointer (var, 0);
333 struct variable_node *vn = lookup_variable (&cat->varmap, var, hash);
336 vn = pool_malloc (cat->pool, sizeof *vn);
338 hmap_init (&vn->valmap);
340 hmap_insert (&cat->varmap, &vn->node, hash);
351 categoricals_update (struct categoricals *cat, const struct ccase *c)
354 struct variable_node *vn = NULL;
355 const double weight = cat->wv ? case_data (c, cat->wv)->f : 1.0;
357 assert (NULL == cat->reverse_variable_map_short);
358 assert (NULL == cat->reverse_variable_map_long);
360 /* Interate over each variable, and add the value of that variable
361 to the appropriate map, if it's not already present. */
362 HMAP_FOR_EACH (vn, struct variable_node, node, &cat->varmap)
364 const int width = var_get_width (vn->var);
365 const union value *val = case_data (c, vn->var);
366 unsigned int hash = value_hash (val, width, 0);
368 struct value_node *valn = lookup_value (&vn->valmap, val, hash, width);
371 valn = pool_malloc (cat->pool, sizeof *valn);
372 value_init (&valn->val, width);
373 value_copy (&valn->val, val, width);
374 hmap_insert (&vn->valmap, &valn->node, hash);
378 for (i = 0 ; i < cat->n_iap; ++i)
380 const struct interaction *iact = cat->iap[i].iact;
382 // if ( interaction_case_is_missing (iact, c, cat->exclude))
385 size_t hash = interaction_case_hash (iact, c, 0);
386 struct interaction_value *node = lookup_case (&cat->iap[i].ivmap, iact, c);
390 node = pool_malloc (cat->pool, sizeof *node);
392 node->ccase = case_ref (c);
395 hmap_insert (&cat->iap[i].ivmap, &node->node, hash);
397 if (cat->user_data_create)
398 node->user_data = cat->user_data_create (cat->aux1, cat->aux2);
404 cat->iap[i].cc += weight;
407 cat->update (node->user_data, cat->exclude, cat->wv, NULL, c, cat->aux1, cat->aux2);
411 /* Return the number of categories (distinct values) for interction N */
413 categoricals_n_count (const struct categoricals *cat, size_t n)
415 return hmap_count (&cat->iap[n].ivmap);
420 categoricals_df (const struct categoricals *cat, size_t n)
422 return cat->iap[n].df;
426 /* Return the total number of categories */
428 categoricals_n_total (const struct categoricals *cat)
430 assert (cat->reverse_variable_map_long);
432 return cat->n_cats_total;
436 categoricals_df_total (const struct categoricals *cat)
441 /* This function must be called *before* any call to categoricals_get_*_by subscript and
442 *after* all calls to categoricals_update */
444 categoricals_done (const struct categoricals *cat_)
446 /* Implementation Note: Whilst this function is O(n) in cat->n_cats_total, in most
447 uses it will be more efficient that using a tree based structure, since it
448 is called only once, and means that subsequent lookups will be O(1).
450 1 call of O(n) + 10^9 calls of O(1) is better than 10^9 calls of O(log n).
452 struct categoricals *cat = CONST_CAST (struct categoricals *, cat_);
458 cat->n_cats_total = 0;
460 /* Calculate the degrees of freedom, and the number of categories */
461 for (i = 0 ; i < cat->n_iap; ++i)
463 const struct interaction *iact = cat->iap[i].iact;
466 cat->iap[i].n_cats = 1;
468 for (v = 0 ; v < iact->n_vars; ++v)
470 const struct variable *var = iact->vars[v];
472 struct variable_node *vn = lookup_variable (&cat->varmap, var, hash_pointer (var, 0));
474 cat->iap[i].df *= hmap_count (&vn->valmap) - 1;
475 cat->iap[i].n_cats *= hmap_count (&vn->valmap);
478 cat->df_sum += cat->iap[i].df;
479 cat->n_cats_total += cat->iap[i].n_cats;
483 cat->reverse_variable_map_short = pool_calloc (cat->pool,
485 sizeof *cat->reverse_variable_map_short);
487 cat->reverse_variable_map_long = pool_calloc (cat->pool,
489 sizeof *cat->reverse_variable_map_long);
491 for (i = 0 ; i < cat->n_iap; ++i)
493 struct interaction_value *ivn = NULL;
496 struct interact_params *iap = &cat->iap[i];
498 iap->base_subscript_short = idx_short;
499 iap->base_subscript_long = idx_long;
501 iap->reverse_interaction_value_map = pool_calloc (cat->pool, iap->n_cats,
502 sizeof *iap->reverse_interaction_value_map);
504 HMAP_FOR_EACH (ivn, struct interaction_value, node, &iap->ivmap)
506 iap->reverse_interaction_value_map[x++] = ivn;
509 assert (x <= iap->n_cats);
511 /* For some purposes (eg CONTRASTS in ONEWAY) the values need to be sorted */
512 sort (iap->reverse_interaction_value_map, x, sizeof (*iap->reverse_interaction_value_map),
513 compare_interaction_value_3way, iap);
515 /* Fill the remaining values with null */
516 for (ii = x ; ii < iap->n_cats; ++ii)
517 iap->reverse_interaction_value_map[ii] = NULL;
519 /* Populate the reverse variable maps. */
520 for (ii = 0; ii < iap->df; ++ii)
521 cat->reverse_variable_map_short[idx_short++] = i;
523 for (ii = 0; ii < iap->n_cats; ++ii)
524 cat->reverse_variable_map_long[idx_long++] = i;
527 assert (cat->n_vars <= cat->n_iap);
529 // categoricals_dump (cat);
534 reverse_variable_lookup_short (const struct categoricals *cat, int subscript)
536 assert (cat->reverse_variable_map_short);
537 assert (subscript >= 0);
538 assert (subscript < cat->df_sum);
540 return cat->reverse_variable_map_short[subscript];
544 reverse_variable_lookup_long (const struct categoricals *cat, int subscript)
546 assert (cat->reverse_variable_map_long);
547 assert (subscript >= 0);
548 assert (subscript < cat->n_cats_total);
550 return cat->reverse_variable_map_long[subscript];
554 /* Return the interaction corresponding to SUBSCRIPT */
555 const struct interaction *
556 categoricals_get_interaction_by_subscript (const struct categoricals *cat, int subscript)
558 int index = reverse_variable_lookup_short (cat, subscript);
560 return cat->iap[index].iact;
564 /* Return the case corresponding to SUBSCRIPT */
565 static const struct ccase *
566 categoricals_get_case_by_subscript (const struct categoricals *cat, int subscript)
568 int vindex = reverse_variable_lookup_short (cat, subscript);
569 const struct interact_params *vp = &cat->iap[vindex];
570 const struct interaction_value *vn = vp->reverse_interaction_value_map [subscript - vp->base_subscript_short];
580 categoricals_get_weight_by_subscript (const struct categoricals *cat, int subscript)
582 int vindex = reverse_variable_lookup_short (cat, subscript);
583 const struct interact_params *vp = &cat->iap[vindex];
589 categoricals_get_sum_by_subscript (const struct categoricals *cat, int subscript)
591 int vindex = reverse_variable_lookup_short (cat, subscript);
592 const struct interact_params *vp = &cat->iap[vindex];
594 const struct interaction_value *iv = vp->reverse_interaction_value_map [subscript - vp->base_subscript_short];
602 /* Returns unity if the value in case C at SUBSCRIPT is equal to the category
603 for that subscript */
605 categoricals_get_binary_by_subscript (const struct categoricals *cat, int subscript,
606 const struct ccase *c)
608 const struct interaction *iact = categoricals_get_interaction_by_subscript (cat, subscript);
610 const struct ccase *c2 = categoricals_get_case_by_subscript (cat, subscript);
615 return interaction_case_equal (iact, c, c2);
620 categoricals_get_n_variables (const struct categoricals *cat)
622 printf ("%s\n", __FUNCTION__);
626 /* Return a case containing the set of values corresponding to SUBSCRIPT */
628 categoricals_get_case_by_category (const struct categoricals *cat, int subscript)
630 int vindex = reverse_variable_lookup_long (cat, subscript);
631 const struct interact_params *vp = &cat->iap[vindex];
632 const struct interaction_value *vn = vp->reverse_interaction_value_map [subscript - vp->base_subscript_long];
638 categoricals_get_user_data_by_category (const struct categoricals *cat, int subscript)
640 int vindex = reverse_variable_lookup_long (cat, subscript);
641 const struct interact_params *vp = &cat->iap[vindex];
643 const struct interaction_value *iv = vp->reverse_interaction_value_map [subscript - vp->base_subscript_long];
644 return iv->user_data;