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 */
77 dump_interaction (const struct interaction *iact)
79 struct string str = DS_EMPTY_INITIALIZER;
80 interaction_to_string (iact, &str);
81 printf ("Interaction: %s\n", ds_cstr (&str));
86 static struct variable_node *
87 lookup_variable (const struct hmap *map, const struct variable *var, unsigned int hash)
89 struct variable_node *vn = NULL;
90 HMAP_FOR_EACH_WITH_HASH (vn, struct variable_node, node, hash, map)
95 fprintf (stderr, "Warning: Hash table collision\n");
102 struct interact_params
104 /* A map indexed by a interaction_value */
107 const struct interaction *iact;
109 int base_subscript_short;
110 int base_subscript_long;
112 /* The number of distinct values of this interaction */
115 /* The degrees of freedom for this interaction */
118 /* A map of interaction_values indexed by subscript */
119 struct interaction_value **reverse_interaction_value_map;
125 /* Comparison function to sort the reverse_value_map in ascending order */
127 compare_interaction_value_3way (const void *vn1_, const void *vn2_, const void *aux)
129 const struct interaction_value *const *vn1p = vn1_;
130 const struct interaction_value *const *vn2p = vn2_;
132 const struct interact_params *iap = aux;
134 return interaction_case_cmp_3way (iap->iact, (*vn1p)->ccase, (*vn2p)->ccase);
139 /* The weight variable */
140 const struct variable *wv;
142 /* An array of interact_params */
143 struct interact_params *iap;
145 /* Map whose members are the union of the variables which comprise IAP */
148 /* The size of IAP. (ie, the number of interactions involved.) */
151 /* The number of categorical variables which contain entries.
152 In the absence of missing values, this will be equal to N_IAP */
157 /* A map to enable the lookup of variables indexed by subscript.
158 This map considers only the N - 1 of the N variables.
160 int *reverse_variable_map_short;
162 /* Like the above, but uses all N variables */
163 int *reverse_variable_map_long;
169 /* Missing values to be excluded */
170 enum mv_class exclude;
172 /* Function to be called on each update */
175 /* Function specified by the caller to create user_data */
176 user_data_create_func *user_data_create;
178 /* Auxilliary data to be passed to update and user_data_create_func*/
185 categoricals_dump (const struct categoricals *cat)
189 printf ("Reverse Variable Map (short):\n");
190 for (i = 0; i < cat->df_sum; ++i)
192 printf (" %d", cat->reverse_variable_map_short[i]);
196 printf ("Reverse Variable Map (long):\n");
197 for (i = 0; i < cat->n_cats_total; ++i)
199 printf (" %d", cat->reverse_variable_map_long[i]);
204 printf ("Number of interactions %d\n", cat->n_iap);
205 for (i = 0 ; i < cat->n_iap; ++i)
209 const struct interact_params *iap = &cat->iap[i];
210 const struct interaction *iact = iap->iact;
212 ds_init_empty (&str);
213 interaction_to_string (iact, &str);
215 printf ("\nInteraction: %s (n: %d; df: %d ); ", ds_cstr (&str), iap->n_cats, iap->df);
217 printf ("Base subscript: %d\n", iap->base_subscript_short);
220 for (v = 0; v < hmap_count (&iap->ivmap); ++v)
223 const struct interaction_value *iv = iap->reverse_interaction_value_map[v];
225 if (v > 0) printf (" ");
227 for (vv = 0; vv < iact->n_vars; ++vv)
229 const struct variable *var = iact->vars[vv];
230 const union value *val = case_data (iv->ccase, var);
232 printf ("%g", val->f);
233 if (vv < iact->n_vars - 1)
244 categoricals_destroy (struct categoricals *cat)
251 static struct interaction_value *
252 lookup_case (const struct hmap *map, const struct interaction *iact, const struct ccase *c)
254 struct interaction_value *iv = NULL;
255 size_t hash = interaction_case_hash (iact, c);
257 HMAP_FOR_EACH_WITH_HASH (iv, struct interaction_value, node, hash, map)
259 if (interaction_case_equal (iact, c, iv->ccase))
262 fprintf (stderr, "Warning: Hash table collision\n");
269 struct categoricals *
270 categoricals_create (struct interaction *const*inter, size_t n_inter,
271 const struct variable *wv, enum mv_class exclude,
272 user_data_create_func *udf,
273 update_func *update, void *aux1, void *aux2
277 struct categoricals *cat = xmalloc (sizeof *cat);
279 cat->n_iap = n_inter;
281 cat->n_cats_total = 0;
283 cat->reverse_variable_map_short = NULL;
284 cat->reverse_variable_map_long = NULL;
285 cat->pool = pool_create ();
286 cat->exclude = exclude;
287 cat->update = update;
288 cat->user_data_create = udf;
293 cat->iap = pool_calloc (cat->pool, cat->n_iap, sizeof *cat->iap);
295 hmap_init (&cat->varmap);
296 for (i = 0 ; i < cat->n_iap; ++i)
299 hmap_init (&cat->iap[i].ivmap);
300 cat->iap[i].iact = inter[i];
301 cat->iap[i].cc = 0.0;
302 for (v = 0; v < inter[i]->n_vars; ++v)
304 const struct variable *var = inter[i]->vars[v];
305 unsigned int hash = hash_pointer (var, 0);
306 struct variable_node *vn = lookup_variable (&cat->varmap, var, hash);
309 vn = pool_malloc (cat->pool, sizeof *vn);
311 hmap_init (&vn->valmap);
313 hmap_insert (&cat->varmap, &vn->node, hash);
324 categoricals_update (struct categoricals *cat, const struct ccase *c)
327 struct variable_node *vn = NULL;
328 const double weight = cat->wv ? case_data (c, cat->wv)->f : 1.0;
330 assert (NULL == cat->reverse_variable_map_short);
331 assert (NULL == cat->reverse_variable_map_long);
333 /* Interate over each variable, and add the value of that variable
334 to the appropriate map, if it's not already present. */
335 HMAP_FOR_EACH (vn, struct variable_node, node, &cat->varmap)
337 const int width = var_get_width (vn->var);
338 const union value *val = case_data (c, vn->var);
339 unsigned int hash = value_hash (val, width, 0);
341 struct value_node *valn = lookup_value (&vn->valmap, val, hash, width);
344 valn = pool_malloc (cat->pool, sizeof *valn);
345 value_init (&valn->val, width);
346 value_copy (&valn->val, val, width);
347 hmap_insert (&vn->valmap, &valn->node, hash);
352 for (i = 0 ; i < cat->n_iap; ++i)
354 const struct interaction *iact = cat->iap[i].iact;
356 // if ( interaction_case_is_missing (iact, c, cat->exclude))
359 size_t hash = interaction_case_hash (iact, c);
360 struct interaction_value *node = lookup_case (&cat->iap[i].ivmap, iact, c);
364 node = pool_malloc (cat->pool, sizeof *node);
366 node->ccase = case_ref (c);
369 hmap_insert (&cat->iap[i].ivmap, &node->node, hash);
371 if (cat->user_data_create)
372 node->user_data = cat->user_data_create (cat->aux1, cat->aux2);
378 cat->iap[i].cc += weight;
381 cat->update (node->user_data, cat->exclude, cat->wv, NULL, c, cat->aux1, cat->aux2);
385 /* Return the number of categories (distinct values) for interction N */
387 categoricals_n_count (const struct categoricals *cat, size_t n)
389 return hmap_count (&cat->iap[n].ivmap);
394 categoricals_df (const struct categoricals *cat, size_t n)
396 return cat->iap[n].df;
400 /* Return the total number of categories */
402 categoricals_n_total (const struct categoricals *cat)
404 assert (cat->reverse_variable_map_long);
406 return cat->n_cats_total;
410 categoricals_df_total (const struct categoricals *cat)
415 /* This function must be called *before* any call to categoricals_get_*_by subscript and
416 *after* all calls to categoricals_update */
418 categoricals_done (const struct categoricals *cat_)
420 /* Implementation Note: Whilst this function is O(n) in cat->n_cats_total, in most
421 uses it will be more efficient that using a tree based structure, since it
422 is called only once, and means that subsequent lookups will be O(1).
424 1 call of O(n) + 10^9 calls of O(1) is better than 10^9 calls of O(log n).
426 struct categoricals *cat = CONST_CAST (struct categoricals *, cat_);
432 cat->n_cats_total = 0;
434 /* Calculate the degrees of freedom, and the number of categories */
435 for (i = 0 ; i < cat->n_iap; ++i)
437 const struct interaction *iact = cat->iap[i].iact;
440 cat->iap[i].n_cats = 1;
442 for (v = 0 ; v < iact->n_vars; ++v)
444 const struct variable *var = iact->vars[v];
446 struct variable_node *vn = lookup_variable (&cat->varmap, var, hash_pointer (var, 0));
448 cat->iap[i].df *= hmap_count (&vn->valmap) - 1;
449 cat->iap[i].n_cats *= hmap_count (&vn->valmap);
452 cat->df_sum += cat->iap[i].df;
453 cat->n_cats_total += cat->iap[i].n_cats;
457 cat->reverse_variable_map_short = pool_calloc (cat->pool,
459 sizeof *cat->reverse_variable_map_short);
461 cat->reverse_variable_map_long = pool_calloc (cat->pool,
463 sizeof *cat->reverse_variable_map_long);
465 for (i = 0 ; i < cat->n_iap; ++i)
467 struct interaction_value *ivn = NULL;
470 struct interact_params *iap = &cat->iap[i];
472 iap->base_subscript_short = idx_short;
473 iap->base_subscript_long = idx_long;
475 iap->reverse_interaction_value_map = pool_calloc (cat->pool, iap->n_cats,
476 sizeof *iap->reverse_interaction_value_map);
478 HMAP_FOR_EACH (ivn, struct interaction_value, node, &iap->ivmap)
480 iap->reverse_interaction_value_map[x++] = ivn;
483 assert (x <= iap->n_cats);
485 /* For some purposes (eg CONTRASTS in ONEWAY) the values need to be sorted */
486 sort (iap->reverse_interaction_value_map, x, sizeof (*iap->reverse_interaction_value_map),
487 compare_interaction_value_3way, iap);
489 /* Populate the reverse variable maps. */
490 for (ii = 0; ii < iap->df; ++ii)
491 cat->reverse_variable_map_short[idx_short++] = i;
493 for (ii = 0; ii < iap->n_cats; ++ii)
494 cat->reverse_variable_map_long[idx_long++] = i;
497 assert (cat->n_vars <= cat->n_iap);
499 // categoricals_dump (cat);
504 reverse_variable_lookup_short (const struct categoricals *cat, int subscript)
506 assert (cat->reverse_variable_map_short);
507 assert (subscript >= 0);
508 assert (subscript < cat->df_sum);
510 return cat->reverse_variable_map_short[subscript];
514 reverse_variable_lookup_long (const struct categoricals *cat, int subscript)
516 assert (cat->reverse_variable_map_long);
517 assert (subscript >= 0);
518 assert (subscript < cat->n_cats_total);
520 return cat->reverse_variable_map_long[subscript];
524 /* Return the interaction corresponding to SUBSCRIPT */
525 const struct interaction *
526 categoricals_get_interaction_by_subscript (const struct categoricals *cat, int subscript)
528 int index = reverse_variable_lookup_short (cat, subscript);
530 return cat->iap[index].iact;
534 /* Return the case corresponding to SUBSCRIPT */
535 static const struct ccase *
536 categoricals_get_case_by_subscript (const struct categoricals *cat, int subscript)
538 int vindex = reverse_variable_lookup_short (cat, subscript);
539 const struct interact_params *vp = &cat->iap[vindex];
540 const struct interaction_value *vn = vp->reverse_interaction_value_map [subscript - vp->base_subscript_short];
546 categoricals_get_weight_by_subscript (const struct categoricals *cat, int subscript)
548 int vindex = reverse_variable_lookup_short (cat, subscript);
549 const struct interact_params *vp = &cat->iap[vindex];
555 categoricals_get_sum_by_subscript (const struct categoricals *cat, int subscript)
557 int vindex = reverse_variable_lookup_short (cat, subscript);
558 const struct interact_params *vp = &cat->iap[vindex];
560 const struct interaction_value *iv = vp->reverse_interaction_value_map [subscript - vp->base_subscript_short];
565 /* Returns unity if the value in case C at SUBSCRIPT is equal to the category
566 for that subscript */
568 categoricals_get_binary_by_subscript (const struct categoricals *cat, int subscript,
569 const struct ccase *c)
571 const struct interaction *iact = categoricals_get_interaction_by_subscript (cat, subscript);
573 const struct ccase *c2 = categoricals_get_case_by_subscript (cat, subscript);
575 return interaction_case_equal (iact, c, c2);
580 categoricals_get_n_variables (const struct categoricals *cat)
582 printf ("%s\n", __FUNCTION__);
586 /* Return a case containing the set of values corresponding to SUBSCRIPT */
588 categoricals_get_case_by_category (const struct categoricals *cat, int subscript)
590 int vindex = reverse_variable_lookup_long (cat, subscript);
591 const struct interact_params *vp = &cat->iap[vindex];
592 const struct interaction_value *vn = vp->reverse_interaction_value_map [subscript - vp->base_subscript_long];
598 categoricals_get_user_data_by_category (const struct categoricals *cat, int subscript)
600 int vindex = reverse_variable_lookup_long (cat, subscript);
601 const struct interact_params *vp = &cat->iap[vindex];
603 const struct interaction_value *iv = vp->reverse_interaction_value_map [subscript - vp->base_subscript_long];
604 return iv->user_data;