X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Fmath%2Fcategoricals.c;h=215f4d1fbc89ca5d406fae4a196bb4797e6cdac0;hb=81579d9e9f994fb2908f50af41c3eb033d216e58;hp=b1f0ce84d9a397bdca0c71062dab79988d9eef0f;hpb=11b8d81d6f403209b15bdf5a58ed2b98cd4ee3bb;p=pspp-builds.git diff --git a/src/math/categoricals.c b/src/math/categoricals.c index b1f0ce84..215f4d1f 100644 --- a/src/math/categoricals.c +++ b/src/math/categoricals.c @@ -1,5 +1,5 @@ /* PSPP - a program for statistical analysis. - Copyright (C) 2009 Free Software Foundation, Inc. + Copyright (C) 2009, 2010, 2011 Free Software Foundation, Inc. This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by @@ -16,29 +16,32 @@ #include -#include +#include "math/categoricals.h" -#include "categoricals.h" +#include -#include -#include -#include -#include -#include -#include +#include "data/case.h" +#include "data/value.h" +#include "data/variable.h" +#include "libpspp/array.h" +#include "libpspp/hmap.h" +#include "libpspp/pool.h" +#include "libpspp/str.h" -#include +#include "gl/xalloc.h" struct value_node { struct hmap_node node; /* Node in hash map. */ union value value; /* The value being labeled. */ double cc; /* The total of the weights of cases with this value */ + + void *user_data; /* A pointer to data which the caller can store stuff */ + int subscript; /* A zero based integer, unique within the variable. Can be used as an index into an array */ }; - struct var_params { /* A map indexed by a union values */ @@ -59,15 +62,34 @@ struct var_params }; +/* Comparison function to sort the reverse_value_map in ascending order */ +static int +compare_value_node (const void *vn1_, const void *vn2_, const void *aux) +{ + const struct value_node * const *vn1 = vn1_; + const struct value_node * const *vn2 = vn2_; + const struct var_params *vp = aux; + + return value_compare_3way (&(*vn1)->value, &(*vn2)->value, var_get_width (vp->var)); +} + + struct categoricals { + /* The weight variable */ const struct variable *wv; - size_t n_vars; /* An array of var_params */ struct var_params *vp; + /* The size of VP. (ie, the number of variables involved.) */ + size_t n_vp; + + /* The number of categorical variables which contain entries. + In the absence of missing values, this will be equal to N_VP */ + size_t n_vars; + /* A map to enable the lookup of variables indexed by subscript */ int *reverse_variable_map; @@ -77,6 +99,16 @@ struct categoricals /* Missing values to be excluded */ enum mv_class exclude; + + /* Function to be called on each update */ + update_func *update; + + /* Function specified by the caller to create user_data */ + user_data_create_func *user_data_create; + + /* Auxilliary data to be passed to update and user_data_create_func*/ + void *aux1; + void *aux2; }; @@ -84,27 +116,32 @@ void categoricals_destroy ( struct categoricals *cat) { int i; - for (i = 0 ; i < cat->n_vars; ++i) - hmap_destroy (&cat->vp[i].map); - - pool_destroy (cat->pool); - free (cat); + if (cat != NULL) + { + for (i = 0 ; i < cat->n_vp; ++i) + hmap_destroy (&cat->vp[i].map); + + pool_destroy (cat->pool); + free (cat); + } } +#if 0 void categoricals_dump (const struct categoricals *cat) { int v; - for (v = 0 ; v < cat->n_vars; ++v) + for (v = 0 ; v < cat->n_vp; ++v) { const struct var_params *vp = &cat->vp[v]; const struct hmap *m = &vp->map; struct hmap_node *node ; int x; - printf ("\n%s (%d) CC=%g n_cats=%d:\n", var_get_name (vp->var), vp->base_subscript, vp->cc, vp->n_cats); + printf ("\n%s (%d) CC=%g n_cats=%d:\n", + var_get_name (vp->var), vp->base_subscript, vp->cc, vp->n_cats); printf ("Reverse map\n"); for (x = 0 ; x < vp->n_cats; ++x) @@ -121,8 +158,8 @@ categoricals_dump (const struct categoricals *cat) for (node = hmap_first (m); node; node = hmap_next (m, node)) { struct string s; - ds_init_empty (&s); const struct value_node *vn = HMAP_DATA (node, struct value_node, node); + ds_init_empty (&s); var_append_value_name (vp->var, &vn->value, &s); printf ("Value: %s; Index %d; CC %g\n", ds_cstr (&s), @@ -130,7 +167,21 @@ categoricals_dump (const struct categoricals *cat) ds_destroy (&s); } } + + assert (cat->n_vars <= cat->n_vp); + + printf ("\n"); + printf ("Number of categorical variables: %d\n", cat->n_vp); + printf ("Number of non-empty categorical variables: %d\n", cat->n_vars); + printf ("Total number of categories: %d\n", cat->n_cats_total); + + printf ("\nReverse variable map:\n"); + + for (v = 0 ; v < cat->n_cats_total; ++v) + printf ("%d ", cat->reverse_variable_map[v]); + printf ("\n"); } +#endif @@ -153,24 +204,33 @@ lookup_value (const struct hmap *map, const struct variable *var, const union va } - struct categoricals * -categoricals_create (const struct variable **v, size_t n_vars, - const struct variable *wv, enum mv_class exclude) +categoricals_create (const struct variable *const *v, size_t n_vars, + const struct variable *wv, enum mv_class exclude, + user_data_create_func *udf, + update_func *update, void *aux1, void *aux2 + ) { size_t i; struct categoricals *cat = xmalloc (sizeof *cat); - cat->n_vars = n_vars; + cat->n_vp = n_vars; cat->wv = wv; cat->n_cats_total = 0; + cat->n_vars = 0; cat->reverse_variable_map = NULL; cat->pool = pool_create (); cat->exclude = exclude; + cat->update = update; + cat->user_data_create = udf; + + cat->aux1 = aux1; + cat->aux2 = aux2; - cat->vp = pool_calloc (cat->pool, n_vars, sizeof *cat->vp); - for (i = 0 ; i < cat->n_vars; ++i) + cat->vp = pool_calloc (cat->pool, cat->n_vp, sizeof *cat->vp); + + for (i = 0 ; i < cat->n_vp; ++i) { hmap_init (&cat->vp[i].map); cat->vp[i].var = v[i]; @@ -190,7 +250,7 @@ categoricals_update (struct categoricals *cat, const struct ccase *c) assert (NULL == cat->reverse_variable_map); - for (i = 0 ; i < cat->n_vars; ++i) + for (i = 0 ; i < cat->n_vp; ++i) { const struct variable *var = cat->vp[i].var; unsigned int width = var_get_width (var); @@ -214,11 +274,21 @@ categoricals_update (struct categoricals *cat, const struct ccase *c) hmap_insert (&cat->vp[i].map, &node->node, hash); cat->n_cats_total++; + + if ( 0 == cat->vp[i].n_cats) + cat->n_vars++; + node->subscript = cat->vp[i].n_cats++ ; + + if (cat->user_data_create) + node->user_data = cat->user_data_create (cat->aux1, cat->aux2); } node->cc += weight; cat->vp[i].cc += weight; + + if (cat->update) + cat->update (node->user_data, cat->exclude, cat->wv, var, c, cat->aux1, cat->aux2); } } @@ -230,19 +300,6 @@ categoricals_n_count (const struct categoricals *cat, size_t n) } -/* Return the index for value VAL in the Nth variable */ -int -categoricals_index (const struct categoricals *cat, size_t n, const union value *val) -{ - struct value_node *vn = lookup_value (&cat->vp[n].map, cat->vp[n].var, val); - - if ( vn == NULL) - return -1; - - return vn->subscript; -} - - /* Return the total number of categories */ size_t categoricals_total (const struct categoricals *cat) @@ -254,7 +311,7 @@ categoricals_total (const struct categoricals *cat) /* This function must be called *before* any call to categoricals_get_*_by subscript an *after* all calls to categoricals_update */ void -categoricals_done (struct categoricals *cat) +categoricals_done (const struct categoricals *cat_) { /* Implementation Note: Whilst this function is O(n) in cat->n_cats_total, in most uses it will be more efficient that using a tree based structure, since it @@ -262,11 +319,14 @@ categoricals_done (struct categoricals *cat) 1 call of O(n) + 10^9 calls of O(1) is better than 10^9 calls of O(log n). */ + struct categoricals *cat = CONST_CAST (struct categoricals *, cat_); int v; int idx = 0; - cat->reverse_variable_map = pool_calloc (cat->pool, cat->n_cats_total, sizeof *cat->reverse_variable_map); + cat->reverse_variable_map = pool_calloc (cat->pool, + cat->n_cats_total, + sizeof *cat->reverse_variable_map); - for (v = 0 ; v < cat->n_vars; ++v) + for (v = 0 ; v < cat->n_vp; ++v) { int i; struct var_params *vp = &cat->vp[v]; @@ -283,32 +343,46 @@ categoricals_done (struct categoricals *cat) vp->reverse_value_map[vn->subscript] = vn; } + /* For some purposes (eg CONTRASTS in ONEWAY) the values need to be sorted */ + sort (vp->reverse_value_map, vp->n_cats, sizeof (const struct value_node *), + compare_value_node, vp); + + /* Populate the reverse variable map. + */ for (i = 0; i < vp->n_cats; ++i) cat->reverse_variable_map[idx++] = v; } + + assert (cat->n_vars <= cat->n_vp); + } +static int +reverse_variable_lookup (const struct categoricals *cat, int subscript) +{ + assert (cat->reverse_variable_map); + assert (subscript >= 0); + assert (subscript < cat->n_cats_total); + + return cat->reverse_variable_map[subscript]; +} + /* Return the categorical variable corresponding to SUBSCRIPT */ const struct variable * categoricals_get_variable_by_subscript (const struct categoricals *cat, int subscript) { - int index; - - assert (cat->reverse_variable_map); - - index = cat->reverse_variable_map[subscript]; + int index = reverse_variable_lookup (cat, subscript); return cat->vp[index].var; } - /* Return the value corresponding to SUBSCRIPT */ const union value * categoricals_get_value_by_subscript (const struct categoricals *cat, int subscript) { - int vindex = cat->reverse_variable_map[subscript]; + int vindex = reverse_variable_lookup (cat, subscript); const struct var_params *vp = &cat->vp[vindex]; const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript]; @@ -319,7 +393,7 @@ categoricals_get_value_by_subscript (const struct categoricals *cat, int subscri double categoricals_get_weight_by_subscript (const struct categoricals *cat, int subscript) { - int vindex = cat->reverse_variable_map[subscript]; + int vindex = reverse_variable_lookup (cat, subscript); const struct var_params *vp = &cat->vp[vindex]; return vp->cc; @@ -328,7 +402,7 @@ categoricals_get_weight_by_subscript (const struct categoricals *cat, int subscr double categoricals_get_sum_by_subscript (const struct categoricals *cat, int subscript) { - int vindex = cat->reverse_variable_map[subscript]; + int vindex = reverse_variable_lookup (cat, subscript); const struct var_params *vp = &cat->vp[vindex]; const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript]; @@ -349,3 +423,22 @@ categoricals_get_binary_by_subscript (const struct categoricals *cat, int subscr return value_equal (val, categoricals_get_value_by_subscript (cat, subscript), width); } + + +size_t +categoricals_get_n_variables (const struct categoricals *cat) +{ + return cat->n_vars; +} + + + +void * +categoricals_get_user_data_by_subscript (const struct categoricals *cat, int subscript) +{ + int vindex = reverse_variable_lookup (cat, subscript); + const struct var_params *vp = &cat->vp[vindex]; + + const struct value_node *vn = vp->reverse_value_map [subscript - vp->base_subscript]; + return vn->user_data; +}