X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Fmath%2Fcategoricals.c;h=a2cf9dd7854c130afee0e6de6cda28b506e9d486;hb=3b45b998691d080fcb64cb8783633cf7b14a54ff;hp=840e2bbfbd31cf9e5cc70420d6e69f1d578d9729;hpb=3d10374169a4abfc58a1d3510ba3792286039240;p=pspp diff --git a/src/math/categoricals.c b/src/math/categoricals.c index 840e2bbfbd..a2cf9dd785 100644 --- a/src/math/categoricals.c +++ b/src/math/categoricals.c @@ -32,6 +32,8 @@ #include "gl/xalloc.h" +#define EFFECTS_CODING 1 + struct value_node { struct hmap_node node; /* Node in hash map. */ @@ -43,12 +45,11 @@ struct value_node struct interaction_value { - struct hmap_node node; /* Node in hash map. */ + 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 */ - /* Total of the weights of cases matching this interaction */ - double cc; + double cc; /* Total of the weights of cases matching this interaction */ void *user_data; /* A pointer to data which the caller can store stuff */ }; @@ -116,8 +117,12 @@ struct interact_params /* The number of distinct values of this interaction */ int n_cats; - /* The degrees of freedom for this interaction */ - int *df; + /* 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; + + double *enc_sum; /* A map of interaction_values indexed by subscript */ struct interaction_value **reverse_interaction_value_map; @@ -259,6 +264,9 @@ categoricals_destroy (struct categoricals *cat) { case_unref (iv->ccase); } + + free (cat->iap[i].enc_sum); + free (cat->iap[i].df_prod); hmap_destroy (&cat->iap[i].ivmap); } @@ -424,7 +432,7 @@ size_t categoricals_df (const struct categoricals *cat, size_t n) { const struct interact_params *iap = &cat->iap[n]; - return iap->df[iap->iact->n_vars - 1]; + return iap->df_prod[iap->iact->n_vars - 1]; } @@ -468,7 +476,7 @@ categoricals_done (const struct categoricals *cat_) int df = 1; const struct interaction *iact = cat->iap[i].iact; - cat->iap[i].df = xcalloc (iact->n_vars, sizeof (int)); + cat->iap[i].df_prod = xcalloc (iact->n_vars, sizeof (int)); cat->iap[i].n_cats = 1; @@ -478,13 +486,13 @@ categoricals_done (const struct categoricals *cat_) struct variable_node *vn = lookup_variable (&cat->varmap, var, hash_pointer (var, 0)); - cat->iap[i].df[v] = df * (hmap_count (&vn->valmap) - 1); - df = cat->iap[i].df[v]; + cat->iap[i].df_prod[v] = df * (hmap_count (&vn->valmap) - 1); + df = cat->iap[i].df_prod[v]; cat->iap[i].n_cats *= hmap_count (&vn->valmap); } - cat->df_sum += cat->iap[i].df [v - 1]; + cat->df_sum += cat->iap[i].df_prod [v - 1]; cat->n_cats_total += cat->iap[i].n_cats; } @@ -526,7 +534,7 @@ categoricals_done (const struct categoricals *cat_) iap->reverse_interaction_value_map[ii] = NULL; /* Populate the reverse variable maps. */ - for (ii = 0; ii < iap->df [iap->iact->n_vars - 1]; ++ii) + 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) @@ -536,6 +544,28 @@ categoricals_done (const struct categoricals *cat_) assert (cat->n_vars <= cat->n_iap); // categoricals_dump (cat); + + /* Tally up the sums for all the encodings */ + for (i = 0 ; i < cat->n_iap; ++i) + { + int x, y; + struct interact_params *iap = &cat->iap[i]; + const struct interaction *iact = iap->iact; + + const int df = iap->df_prod [iact->n_vars - 1]; + + iap->enc_sum = xcalloc (df, sizeof (*(iap->enc_sum))); + + for (y = 0; y < hmap_count (&iap->ivmap); ++y) + { + 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); \ + iap->enc_sum [x - iap->base_subscript_short] += bin * iv->cc; + } + } + } } @@ -599,12 +629,7 @@ categoricals_get_sum_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 *iv = vp->reverse_interaction_value_map [subscript - vp->base_subscript_short]; - - if (iv == NULL) - return 0; - - return iv->cc; + return vp->enc_sum[subscript - vp->base_subscript_short]; } /* Returns unity if the value in case C at SUBSCRIPT is equal to the category @@ -623,6 +648,9 @@ categoricals_get_binary_by_subscript (const struct categoricals *cat, int v; double result = 1.0; + const struct interact_params *iap = &cat->iap[i]; + + double dfp = 1.0; for (v = 0; v < iact->n_vars; ++v) { const struct variable *var = iact->vars[v]; @@ -634,14 +662,16 @@ categoricals_get_binary_by_subscript (const struct categoricals *cat, 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; + + const double df = iap->df_prod[v] / dfp; + /* Translate the subscript into an index for the individual variable */ - int index = (subscript - base_index) % cat->iap[i].df[v]; - if ( v > 0) - index /= cat->iap[i].df[v - 1]; + const int index = ((subscript - base_index) % iap->df_prod[v] ) / dfp; + dfp = iap->df_prod [v]; - double bin = 1.0; #if EFFECTS_CODING - if ( valn->index == 0) + if ( valn->index == df ) bin = -1.0; else #endif @@ -682,3 +712,7 @@ categoricals_get_user_data_by_category (const struct categoricals *cat, int subs const struct interaction_value *iv = vp->reverse_interaction_value_map [subscript - vp->base_subscript_long]; return iv->user_data; } + + + +