X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Fmath%2Fcovariance.c;h=aa7f41771c1ac9f3d55ff90d5a418010e5e032a8;hb=66946a0a0cab3a51e6fded95aef4e991baee319f;hp=7e9736291c4a32e81cad37b0a6fac7aa902f5b26;hpb=8f471c51c709556ebfe46eaeb716ad171feae972;p=pspp diff --git a/src/math/covariance.c b/src/math/covariance.c index 7e9736291c..aa7f41771c 100644 --- a/src/math/covariance.c +++ b/src/math/covariance.c @@ -68,7 +68,7 @@ struct covariance { /* The variables for which the covariance matrix is to be calculated. */ size_t n_vars; - const struct variable **vars; + const struct variable *const *vars; /* Categorical variables. */ struct categoricals *categoricals; @@ -131,11 +131,11 @@ covariance_moments (const struct covariance *cov, int m) /* Create a covariance struct. */ struct covariance * -covariance_1pass_create (size_t n_vars, const struct variable **vars, +covariance_1pass_create (size_t n_vars, const struct variable *const *vars, const struct variable *weight, enum mv_class exclude) { size_t i; - struct covariance *cov = xmalloc (sizeof *cov); + struct covariance *cov = xzalloc (sizeof *cov); cov->passes = 1; cov->state = 0; @@ -156,7 +156,9 @@ covariance_1pass_create (size_t n_vars, const struct variable **vars, cov->n_cm = (n_vars * (n_vars - 1) ) / 2; - cov->cm = xcalloc (sizeof *cov->cm, cov->n_cm); + if (cov->n_cm > 0) + cov->cm = xcalloc (sizeof *cov->cm, cov->n_cm); + cov->categoricals = NULL; return cov; } @@ -168,8 +170,8 @@ covariance_1pass_create (size_t n_vars, const struct variable **vars, until then. */ struct covariance * -covariance_2pass_create (size_t n_vars, const struct variable **vars, - size_t n_catvars, const struct variable **catvars, +covariance_2pass_create (size_t n_vars, const struct variable *const *vars, + struct categoricals *cats, const struct variable *wv, enum mv_class exclude) { size_t i; @@ -192,10 +194,10 @@ covariance_2pass_create (size_t n_vars, const struct variable **vars, cov->exclude = exclude; - cov->n_cm = - 1; + cov->n_cm = -1; cov->cm = NULL; - cov->categoricals = categoricals_create (catvars, n_catvars, wv); + cov->categoricals = cats; return cov; } @@ -265,7 +267,8 @@ get_val (const struct covariance *cov, int i, const struct ccase *c) return categoricals_get_binary_by_subscript (cov->categoricals, i - cov->n_vars, c); } -static void +#if 0 +void dump_matrix (const gsl_matrix *m) { size_t i, j; @@ -277,6 +280,7 @@ dump_matrix (const gsl_matrix *m) printf ("\n"); } } +#endif /* Call this function for every case in the data set */ void @@ -292,7 +296,8 @@ covariance_accumulate_pass1 (struct covariance *cov, const struct ccase *c) cov->state = 1; } - categoricals_update (cov->categoricals, c); + if (cov->categoricals) + categoricals_update (cov->categoricals, c); for (i = 0 ; i < cov->dim; ++i) { @@ -334,10 +339,16 @@ covariance_accumulate_pass2 (struct covariance *cov, const struct ccase *c) if (! cov->pass_two_first_case_seen) { + size_t m; assert (cov->state == 1); cov->state = 2; - cov->dim = cov->n_vars + categoricals_total (cov->categoricals); + cov->dim = cov->n_vars; + + if (cov->categoricals) + cov->dim += categoricals_total (cov->categoricals) + - categoricals_get_n_variables (cov->categoricals); + cov->n_cm = (cov->dim * (cov->dim - 1) ) / 2; cov->cm = xcalloc (sizeof *cov->cm, cov->n_cm); @@ -348,12 +359,42 @@ covariance_accumulate_pass2 (struct covariance *cov, const struct ccase *c) cov->moments[i] = resize_matrix (cov->moments[i], cov->dim); } - categoricals_done (cov->categoricals); + if (cov->categoricals) + categoricals_done (cov->categoricals); + + /* Populate the moments matrices with the categorical value elements */ + for (i = cov->n_vars; i < cov->dim; ++i) + { + for (j = 0 ; j < cov->dim ; ++j) /* FIXME: This is WRONG !!! */ + { + double w = categoricals_get_weight_by_subscript (cov->categoricals, i - cov->n_vars); + + gsl_matrix_set (cov->moments[MOMENT_NONE], i, j, w); + + w = categoricals_get_sum_by_subscript (cov->categoricals, i - cov->n_vars); + + gsl_matrix_set (cov->moments[MOMENT_MEAN], i, j, w); + } + } + + /* FIXME: This is WRONG!! It must be fixed to properly handle missing values. For + now it assumes there are none */ + for (m = 0 ; m < n_MOMENTS; ++m) + { + for (i = 0 ; i < cov->dim ; ++i) + { + double x = gsl_matrix_get (cov->moments[m], i, cov->n_vars -1); + for (j = cov->n_vars; j < cov->dim; ++j) + { + gsl_matrix_set (cov->moments[m], i, j, x); + } + } + } /* Divide the means by the number of samples */ - for (i = 0; i < cov->n_vars; ++i) + for (i = 0; i < cov->dim; ++i) { - for (j = 0; j < cov->n_vars; ++j) + for (j = 0; j < cov->dim; ++j) { double *x = gsl_matrix_ptr (cov->moments[MOMENT_MEAN], i, j); *x /= gsl_matrix_get (cov->moments[MOMENT_NONE], i, j); @@ -396,7 +437,6 @@ covariance_accumulate_pass2 (struct covariance *cov, const struct ccase *c) { cov->cm [idx] += ss; } - } } @@ -558,7 +598,6 @@ covariance_calculate_single_pass (struct covariance *cov) } - /* Return a pointer to gsl_matrix containing the pairwise covariances. The matrix remains owned by the COV object, and must not be freed. @@ -567,7 +606,8 @@ covariance_calculate_single_pass (struct covariance *cov) const gsl_matrix * covariance_calculate (struct covariance *cov) { - assert ( cov->state > 0 ); + if ( cov->state <= 0 ) + return NULL; switch (cov->passes) { @@ -582,7 +622,95 @@ covariance_calculate (struct covariance *cov) } } +/* + Covariance computed without dividing by the sample size. + */ +static const gsl_matrix * +covariance_calculate_double_pass_unnormalized (struct covariance *cov) +{ + size_t i, j; + for (i = 0 ; i < cov->dim; ++i) + { + for (j = 0 ; j < cov->dim; ++j) + { + int idx; + double *x = gsl_matrix_ptr (cov->moments[MOMENT_VARIANCE], i, j); + + idx = cm_idx (cov, i, j); + if ( idx >= 0) + { + x = &cov->cm [idx]; + } + } + } + + return cm_to_gsl (cov); +} + +static const gsl_matrix * +covariance_calculate_single_pass_unnormalized (struct covariance *cov) +{ + size_t i, j; + + for (i = 0 ; i < cov->dim; ++i) + { + for (j = 0 ; j < cov->dim; ++j) + { + double *x = gsl_matrix_ptr (cov->moments[MOMENT_VARIANCE], i, j); + *x -= pow2 (gsl_matrix_get (cov->moments[MOMENT_MEAN], i, j)) + / gsl_matrix_get (cov->moments[MOMENT_NONE], i, j); + } + } + for ( j = 0 ; j < cov->dim - 1; ++j) + { + for (i = j + 1 ; i < cov->dim; ++i) + { + double *x = &cov->cm [cm_idx (cov, i, j)]; + + *x -= + gsl_matrix_get (cov->moments[MOMENT_MEAN], i, j) + * + gsl_matrix_get (cov->moments[MOMENT_MEAN], j, i) + / gsl_matrix_get (cov->moments[MOMENT_NONE], i, j); + } + } + + return cm_to_gsl (cov); +} + +/* + Return a pointer to gsl_matrix containing the pairwise covariances. + The matrix remains owned by the COV object, and must not be freed. + Call this function only after all data have been accumulated. +*/ +const gsl_matrix * +covariance_calculate_unnormalized (struct covariance *cov) +{ + if ( cov->state <= 0 ) + return NULL; + + switch (cov->passes) + { + case 1: + return covariance_calculate_single_pass_unnormalized (cov); + break; + case 2: + return covariance_calculate_double_pass_unnormalized (cov); + break; + default: + NOT_REACHED (); + } +} + +/* Function to access the categoricals used by COV + The return value is owned by the COV +*/ +const struct categoricals * +covariance_get_categoricals (const struct covariance *cov) +{ + return cov->categoricals; +} /* Destroy the COV object */ @@ -590,7 +718,7 @@ void covariance_destroy (struct covariance *cov) { size_t i; - free (cov->vars); + categoricals_destroy (cov->categoricals); for (i = 0; i < n_MOMENTS; ++i)