X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Fmath%2Fcategoricals.c;h=e1bbe1f99b141994365d32de6f888a2dee2619e4;hb=40c169c1409b0f16e6ee8f0ceaa0ce6937c90a5c;hp=a2cf9dd7854c130afee0e6de6cda28b506e9d486;hpb=3b45b998691d080fcb64cb8783633cf7b14a54ff;p=pspp diff --git a/src/math/categoricals.c b/src/math/categoricals.c index a2cf9dd785..e1bbe1f99b 100644 --- a/src/math/categoricals.c +++ b/src/math/categoricals.c @@ -1,5 +1,5 @@ /* PSPP - a program for statistical analysis. - Copyright (C) 2009, 2010, 2011 Free Software Foundation, Inc. + Copyright (C) 2009, 2010, 2011, 2012, 2014 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 @@ -19,6 +19,7 @@ #include "math/categoricals.h" #include "math/interaction.h" +#include #include #include "data/case.h" @@ -32,7 +33,7 @@ #include "gl/xalloc.h" -#define EFFECTS_CODING 1 +#define CATEGORICALS_DEBUG 0 struct value_node { @@ -43,13 +44,15 @@ struct value_node int index; /* A zero based unique index for this value */ }; + struct interaction_value { struct hmap_node node; /* Node in hash map */ - struct ccase *ccase; /* A case (probably the first in the dataset) which matches this value */ + struct ccase *ccase; /* A case (probably the first in the dataset) which matches + this value */ - double cc; /* Total of the weights of cases matching this interaction */ + double cc; /* Total of the weights of cases matching this interaction */ void *user_data; /* A pointer to data which the caller can store stuff */ }; @@ -63,7 +66,7 @@ lookup_value (const struct hmap *map, const union value *val, unsigned int hash, if (value_equal (&vn->val, val, width)) break; } - + return vn; } @@ -76,16 +79,20 @@ struct variable_node int n_vals; /* Number of values for this variable */ }; -#if 0 -static void -dump_interaction (const struct interaction *iact) + +/* Comparison function to sort value_nodes in ascending order */ +static int +compare_value_node_3way (const void *vn1_, const void *vn2_, const void *aux) { - struct string str = DS_EMPTY_INITIALIZER; - interaction_to_string (iact, &str); - printf ("Interaction: %s\n", ds_cstr (&str)); - ds_destroy (&str); + const struct value_node *const *vn1p = vn1_; + const struct value_node *const *vn2p = vn2_; + + const struct variable_node *vn = aux; + + + return value_compare_3way (&(*vn1p)->val, &(*vn2p)->val, var_get_width (vn->var)); } -#endif + static struct variable_node * @@ -96,17 +103,17 @@ lookup_variable (const struct hmap *map, const struct variable *var, unsigned in { if (vn->var == var) break; - - fprintf (stderr, "Warning: Hash table collision\n"); + + fprintf (stderr, "%s:%d Warning: Hash table collision\n", __FILE__, __LINE__); } - + return vn; } struct interact_params { - /* A map indexed by a interaction_value */ + /* A map of cases indexed by a interaction_value */ struct hmap ivmap; const struct interaction *iact; @@ -118,9 +125,9 @@ struct interact_params int n_cats; /* An array of integers df_n * df_{n-1} * df_{n-2} ... - These are the products of the degrees of freedom for the current - variable and all preceeding variables */ - int *df_prod; + These are the products of the degrees of freedom for the current + variable and all preceding variables */ + int *df_prod; double *enc_sum; @@ -165,7 +172,7 @@ struct categoricals /* A map to enable the lookup of variables indexed by subscript. This map considers only the N - 1 of the N variables. - */ + */ int *reverse_variable_map_short; /* Like the above, but uses all N variables */ @@ -175,79 +182,112 @@ struct categoricals struct pool *pool; - /* Missing values to be excluded */ - enum mv_class exclude; + /* Missing values in the dependent varirable to be excluded */ + enum mv_class dep_excl; - /* Function to be called on each update */ - update_func *update; + /* Missing values in the factor variables to be excluded */ + enum mv_class fctr_excl; - /* 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; + const void *aux1; void *aux2; + + bool sane; + + const struct payload *payload; }; -#if 0 -static void -categoricals_dump (const struct categoricals *cat) + +bool +categoricals_isbalanced (const struct categoricals *cat) { int i; - printf ("Reverse Variable Map (short):\n"); - for (i = 0; i < cat->df_sum; ++i) + for (i = 0 ; i < cat->n_iap; ++i) { - printf (" %d", cat->reverse_variable_map_short[i]); - } - printf ("\n"); + int v; + const struct interact_params *iap = &cat->iap[i]; - printf ("Reverse Variable Map (long):\n"); - for (i = 0; i < cat->n_cats_total; ++i) - { - printf (" %d", cat->reverse_variable_map_long[i]); + double oval = -1.0; + for (v = 0; v < hmap_count (&iap->ivmap); ++v) + { + const struct interaction_value *iv = iap->reverse_interaction_value_map[v]; + if (oval == -1.0) + oval = iv->cc; + if (oval != iv->cc) + return false; + } } - printf ("\n"); + return true; +} - printf ("Number of interactions %d\n", cat->n_iap); - for (i = 0 ; i < cat->n_iap; ++i) +static void +categoricals_dump (const struct categoricals *cat) +{ + if (CATEGORICALS_DEBUG) { - int v; - struct string str; - const struct interact_params *iap = &cat->iap[i]; - const struct interaction *iact = iap->iact; + int i; - ds_init_empty (&str); - interaction_to_string (iact, &str); + printf ("Reverse Variable Map (short):\n"); + for (i = 0; i < cat->df_sum; ++i) + { + printf (" %d", cat->reverse_variable_map_short[i]); + } + printf ("\n"); - printf ("\nInteraction: %s (n: %d; df: %d ); ", ds_cstr (&str), iap->n_cats, iap->df); - ds_destroy (&str); - printf ("Base subscript: %d\n", iap->base_subscript_short); + printf ("Reverse Variable Map (long):\n"); + for (i = 0; i < cat->n_cats_total; ++i) + { + printf (" %d", cat->reverse_variable_map_long[i]); + } + printf ("\n"); - printf ("\t("); - for (v = 0; v < hmap_count (&iap->ivmap); ++v) + printf ("Number of interactions %zu\n", cat->n_iap); + for (i = 0 ; i < cat->n_iap; ++i) { - int vv; - const struct interaction_value *iv = iap->reverse_interaction_value_map[v]; - - if (v > 0) printf (" "); - printf ("{"); - for (vv = 0; vv < iact->n_vars; ++vv) + int v; + struct string str; + const struct interact_params *iap = &cat->iap[i]; + const struct interaction *iact = iap->iact; + + ds_init_empty (&str); + interaction_to_string (iact, &str); + + printf ("\nInteraction: \"%s\" (number of categories: %d); ", ds_cstr (&str), iap->n_cats); + ds_destroy (&str); + printf ("Base index (short/long): %d/%d\n", iap->base_subscript_short, iap->base_subscript_long); + + printf ("\t("); + for (v = 0; v < hmap_count (&iap->ivmap); ++v) { - const struct variable *var = iact->vars[vv]; - const union value *val = case_data (iv->ccase, var); - - printf ("%g", val->f); - if (vv < iact->n_vars - 1) - printf (", "); + int vv; + const struct interaction_value *iv = iap->reverse_interaction_value_map[v]; + + if (v > 0) printf (" "); + printf ("{"); + for (vv = 0; vv < iact->n_vars; ++vv) + { + const struct variable *var = iact->vars[vv]; + const union value *val = case_data (iv->ccase, var); + unsigned int varhash = hash_pointer (var, 0); + struct variable_node *vn = lookup_variable (&cat->varmap, var, varhash); + + const int width = var_get_width (var); + unsigned int valhash = value_hash (val, width, 0); + struct value_node *valn = lookup_value (&vn->valmap, val, valhash, width); + + assert (vn->var == var); + + printf ("%.*g(%d)", DBL_DIG + 1, val->f, valn->index); + if (vv < iact->n_vars - 1) + printf (", "); + } + printf ("}"); } - printf ("}"); + printf (")\n"); } - printf (")\n"); } } -#endif void categoricals_destroy (struct categoricals *cat) @@ -256,12 +296,15 @@ categoricals_destroy (struct categoricals *cat) int i; if (NULL == cat) return; + for (i = 0; i < cat->n_iap; ++i) { struct interaction_value *iv = NULL; /* Interate over each interaction value, and unref any cases that we reffed */ HMAP_FOR_EACH (iv, struct interaction_value, node, &cat->iap[i].ivmap) { + if (cat->payload && cat->payload->destroy) + cat->payload->destroy (cat->aux1, cat->aux2, iv->user_data); case_unref (iv->ccase); } @@ -302,17 +345,19 @@ lookup_case (const struct hmap *map, const struct interaction *iact, const struc return iv; } +bool +categoricals_sane (const struct categoricals *cat) +{ + return cat->sane; +} struct categoricals * categoricals_create (struct interaction *const*inter, size_t n_inter, - const struct variable *wv, enum mv_class exclude, - user_data_create_func *udf, - update_func *update, void *aux1, void *aux2 - ) + const struct variable *wv, enum mv_class dep_excl, enum mv_class fctr_excl) { size_t i; struct categoricals *cat = xmalloc (sizeof *cat); - + cat->n_iap = n_inter; cat->wv = wv; cat->n_cats_total = 0; @@ -320,12 +365,11 @@ categoricals_create (struct interaction *const*inter, size_t n_inter, cat->reverse_variable_map_short = NULL; cat->reverse_variable_map_long = NULL; cat->pool = pool_create (); - cat->exclude = exclude; - cat->update = update; - cat->user_data_create = udf; - - cat->aux1 = aux1; - cat->aux2 = aux2; + cat->dep_excl = dep_excl; + cat->fctr_excl = fctr_excl; + cat->payload = NULL; + cat->aux2 = NULL; + cat->sane = false; cat->iap = pool_calloc (cat->pool, cat->n_iap, sizeof *cat->iap); @@ -363,7 +407,13 @@ categoricals_update (struct categoricals *cat, const struct ccase *c) { int i; struct variable_node *vn = NULL; - const double weight = cat->wv ? case_data (c, cat->wv)->f : 1.0; + double weight; + + if (NULL == cat) + return; + + weight = cat->wv ? case_data (c, cat->wv)->f : 1.0; + weight = var_force_valid_weight (cat->wv, weight, NULL); assert (NULL == cat->reverse_variable_map_short); assert (NULL == cat->reverse_variable_map_long); @@ -380,34 +430,39 @@ categoricals_update (struct categoricals *cat, const struct ccase *c) if (valn == NULL) { valn = pool_malloc (cat->pool, sizeof *valn); - valn->index = vn->n_vals++; + valn->index = -1; + vn->n_vals++; value_init (&valn->val, width); value_copy (&valn->val, val, width); hmap_insert (&vn->valmap, &valn->node, hash); } - } - + } + for (i = 0 ; i < cat->n_iap; ++i) { const struct interaction *iact = cat->iap[i].iact; - // if ( interaction_case_is_missing (iact, c, cat->exclude)) - // continue; + size_t hash; + struct interaction_value *node; + + if ( interaction_case_is_missing (iact, c, cat->fctr_excl)) + continue; - size_t hash = interaction_case_hash (iact, c, 0); - struct interaction_value *node = lookup_case (&cat->iap[i].ivmap, iact, c); + hash = interaction_case_hash (iact, c, 0); + node = lookup_case (&cat->iap[i].ivmap, iact, c); if ( NULL == node) { node = pool_malloc (cat->pool, sizeof *node); - node->ccase = case_ref (c); node->cc = weight; hmap_insert (&cat->iap[i].ivmap, &node->node, hash); - if (cat->user_data_create) - node->user_data = cat->user_data_create (cat->aux1, cat->aux2); + if (cat->payload) + { + node->user_data = cat->payload->create (cat->aux1, cat->aux2); + } } else { @@ -415,8 +470,10 @@ categoricals_update (struct categoricals *cat, const struct ccase *c) } cat->iap[i].cc += weight; - if (cat->update) - cat->update (node->user_data, cat->exclude, cat->wv, NULL, c, cat->aux1, cat->aux2); + if (cat->payload) + { + cat->payload->update (cat->aux1, cat->aux2, node->user_data, c, weight); + } } } @@ -440,7 +497,8 @@ categoricals_df (const struct categoricals *cat, size_t n) size_t categoricals_n_total (const struct categoricals *cat) { - assert (cat->reverse_variable_map_long); + if (!categoricals_is_complete (cat)) + return 0; return cat->n_cats_total; } @@ -448,9 +506,19 @@ categoricals_n_total (const struct categoricals *cat) size_t categoricals_df_total (const struct categoricals *cat) { + if (NULL == cat) + return 0; + return cat->df_sum; } +bool +categoricals_is_complete (const struct categoricals *cat) +{ + return (NULL != cat->reverse_variable_map_short); +} + + /* This function must be called *before* any call to categoricals_get_*_by subscript and *after* all calls to categoricals_update */ void @@ -467,6 +535,10 @@ categoricals_done (const struct categoricals *cat_) int i; int idx_short = 0; int idx_long = 0; + + if (NULL == cat) + return; + cat->df_sum = 0; cat->n_cats_total = 0; @@ -476,23 +548,57 @@ categoricals_done (const struct categoricals *cat_) int df = 1; const struct interaction *iact = cat->iap[i].iact; - cat->iap[i].df_prod = xcalloc (iact->n_vars, sizeof (int)); + cat->iap[i].df_prod = iact->n_vars ? xcalloc (iact->n_vars, sizeof (int)) : NULL; cat->iap[i].n_cats = 1; - + for (v = 0 ; v < iact->n_vars; ++v) { + int x; const struct variable *var = iact->vars[v]; struct variable_node *vn = lookup_variable (&cat->varmap, var, hash_pointer (var, 0)); - cat->iap[i].df_prod[v] = df * (hmap_count (&vn->valmap) - 1); - df = cat->iap[i].df_prod[v]; + struct value_node *valnd = NULL; + struct value_node **array ; + + assert (vn->n_vals == hmap_count (&vn->valmap)); + + if (vn->n_vals == 0) + { + cat->sane = false; + return; + } + + /* Sort the VALMAP here */ + array = xcalloc (sizeof *array, vn->n_vals); + x = 0; + HMAP_FOR_EACH (valnd, struct value_node, node, &vn->valmap) + { + /* Note: This loop is probably superfluous, it could be done in the + update stage (at the expense of a realloc) */ + array[x++] = valnd; + } + + sort (array, vn->n_vals, sizeof (*array), + compare_value_node_3way, vn); - cat->iap[i].n_cats *= hmap_count (&vn->valmap); + for (x = 0; x < vn->n_vals; ++x) + { + struct value_node *vvv = array[x]; + vvv->index = x; + } + free (array); + + cat->iap[i].df_prod[v] = df * (vn->n_vals - 1); + df = cat->iap[i].df_prod[v]; + + cat->iap[i].n_cats *= vn->n_vals; } - cat->df_sum += cat->iap[i].df_prod [v - 1]; + if (v > 0) + cat->df_sum += cat->iap[i].df_prod [v - 1]; + cat->n_cats_total += cat->iap[i].n_cats; } @@ -534,8 +640,11 @@ categoricals_done (const struct categoricals *cat_) iap->reverse_interaction_value_map[ii] = NULL; /* Populate the reverse variable maps. */ - for (ii = 0; ii < iap->df_prod [iap->iact->n_vars - 1]; ++ii) - cat->reverse_variable_map_short[idx_short++] = i; + if (iap->df_prod) + { + for (ii = 0; ii < iap->df_prod [iap->iact->n_vars - 1]; ++ii) + cat->reverse_variable_map_short[idx_short++] = i; + } for (ii = 0; ii < iap->n_cats; ++ii) cat->reverse_variable_map_long[idx_long++] = i; @@ -543,7 +652,7 @@ categoricals_done (const struct categoricals *cat_) assert (cat->n_vars <= cat->n_iap); - // categoricals_dump (cat); + categoricals_dump (cat); /* Tally up the sums for all the encodings */ for (i = 0 ; i < cat->n_iap; ++i) @@ -552,7 +661,7 @@ categoricals_done (const struct categoricals *cat_) struct interact_params *iap = &cat->iap[i]; const struct interaction *iact = iap->iact; - const int df = iap->df_prod [iact->n_vars - 1]; + const int df = iap->df_prod ? iap->df_prod [iact->n_vars - 1] : 0; iap->enc_sum = xcalloc (df, sizeof (*(iap->enc_sum))); @@ -561,11 +670,15 @@ categoricals_done (const struct categoricals *cat_) struct interaction_value *iv = iap->reverse_interaction_value_map[y]; for (x = iap->base_subscript_short; x < iap->base_subscript_short + df ;++x) { - const double bin = categoricals_get_binary_by_subscript (cat, x, iv->ccase); \ + const double bin = categoricals_get_effects_code_for_case (cat, x, iv->ccase); iap->enc_sum [x - iap->base_subscript_short] += bin * iv->cc; } + if (cat->payload && cat->payload->calculate) + cat->payload->calculate (cat->aux1, cat->aux2, iv->user_data); } } + + cat->sane = true; } @@ -599,21 +712,6 @@ categoricals_get_interaction_by_subscript (const struct categoricals *cat, int s return cat->iap[index].iact; } -/* Return the case corresponding to SUBSCRIPT */ -static const struct ccase * -categoricals_get_case_by_subscript (const struct categoricals *cat, int subscript) -{ - int vindex = reverse_variable_lookup_short (cat, subscript); - const struct interact_params *vp = &cat->iap[vindex]; - const struct interaction_value *vn = vp->reverse_interaction_value_map [subscript - vp->base_subscript_short]; - - if ( vn == NULL) - return NULL; - - return vn->ccase; -} - - double categoricals_get_weight_by_subscript (const struct categoricals *cat, int subscript) { @@ -632,12 +730,13 @@ categoricals_get_sum_by_subscript (const struct categoricals *cat, int subscript return vp->enc_sum[subscript - vp->base_subscript_short]; } + /* Returns unity if the value in case C at SUBSCRIPT is equal to the category for that subscript */ -double -categoricals_get_binary_by_subscript (const struct categoricals *cat, - int subscript, - const struct ccase *c) +static double +categoricals_get_code_for_case (const struct categoricals *cat, int subscript, + const struct ccase *c, + bool effects_coding) { const struct interaction *iact = categoricals_get_interaction_by_subscript (cat, subscript); @@ -652,39 +751,58 @@ categoricals_get_binary_by_subscript (const struct categoricals *cat, double dfp = 1.0; for (v = 0; v < iact->n_vars; ++v) - { - const struct variable *var = iact->vars[v]; + { + const struct variable *var = iact->vars[v]; - const union value *val = case_data (c, var); - const int width = var_get_width (var); - const struct variable_node *vn = lookup_variable (&cat->varmap, var, hash_pointer (var, 0)); + const union value *val = case_data (c, var); + const int width = var_get_width (var); + const struct variable_node *vn = lookup_variable (&cat->varmap, var, hash_pointer (var, 0)); - const unsigned int hash = value_hash (val, width, 0); - const struct value_node *valn = lookup_value (&vn->valmap, val, hash, width); + const unsigned int hash = value_hash (val, width, 0); + const struct value_node *valn = lookup_value (&vn->valmap, val, hash, width); - double bin = 1.0; + double bin = 1.0; - const double df = iap->df_prod[v] / dfp; + const double df = iap->df_prod[v] / dfp; - /* Translate the subscript into an index for the individual variable */ - const int index = ((subscript - base_index) % iap->df_prod[v] ) / dfp; - dfp = iap->df_prod [v]; + /* Translate the subscript into an index for the individual variable */ + const int index = ((subscript - base_index) % iap->df_prod[v] ) / dfp; + dfp = iap->df_prod [v]; -#if EFFECTS_CODING - if ( valn->index == df ) - bin = -1.0; - else -#endif - if ( valn->index != index ) - bin = 0; - - result *= bin; - } + if (effects_coding && valn->index == df ) + bin = -1.0; + else if ( valn->index != index ) + bin = 0; + + result *= bin; + } return result; } +/* Returns unity if the value in case C at SUBSCRIPT is equal to the category + for that subscript */ +double +categoricals_get_dummy_code_for_case (const struct categoricals *cat, int subscript, + const struct ccase *c) +{ + return categoricals_get_code_for_case (cat, subscript, c, false); +} + +/* Returns unity if the value in case C at SUBSCRIPT is equal to the category + for that subscript. + Else if it is the last category, return -1. + Otherwise return 0. + */ +double +categoricals_get_effects_code_for_case (const struct categoricals *cat, int subscript, + const struct ccase *c) +{ + return categoricals_get_code_for_case (cat, subscript, c, true); +} + + size_t categoricals_get_n_variables (const struct categoricals *cat) { @@ -692,6 +810,41 @@ categoricals_get_n_variables (const struct categoricals *cat) return cat->n_vars; } + +/* Return a case containing the set of values corresponding to + the Nth Category of the IACTth interaction */ +const struct ccase * +categoricals_get_case_by_category_real (const struct categoricals *cat, int iact, int n) +{ + const struct interaction_value *vn; + + const struct interact_params *vp = &cat->iap[iact]; + + if ( n >= hmap_count (&vp->ivmap)) + return NULL; + + vn = vp->reverse_interaction_value_map [n]; + + return vn->ccase; +} + +/* Return a the user data corresponding to the Nth Category of the IACTth interaction. */ +void * +categoricals_get_user_data_by_category_real (const struct categoricals *cat, int iact, int n) +{ + const struct interact_params *vp = &cat->iap[iact]; + const struct interaction_value *iv ; + + if ( n >= hmap_count (&vp->ivmap)) + return NULL; + + iv = vp->reverse_interaction_value_map [n]; + + return iv->user_data; +} + + + /* Return a case containing the set of values corresponding to SUBSCRIPT */ const struct ccase * categoricals_get_case_by_category (const struct categoricals *cat, int subscript) @@ -714,5 +867,13 @@ categoricals_get_user_data_by_category (const struct categoricals *cat, int subs } + - +void +categoricals_set_payload (struct categoricals *cat, const struct payload *p, + const void *aux1, void *aux2) +{ + cat->payload = p; + cat->aux1 = aux1; + cat->aux2 = aux2; +}