X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Fmath%2Fcovariance.c;h=66b44c12c10859cd5a882db66e7c22407ce47d84;hb=12c7324445fc5fd00d0536172373ddd8b253d0e2;hp=f0161a54c529c59b187d840611e3cde134d45852;hpb=b4a1fe2442a0b6fc8716e92574d33726d7170fe8;p=pspp diff --git a/src/math/covariance.c b/src/math/covariance.c index f0161a54c5..66b44c12c1 100644 --- a/src/math/covariance.c +++ b/src/math/covariance.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 @@ -16,15 +16,19 @@ #include -#include -#include "covariance.h" -#include -#include "moments.h" +#include "math/covariance.h" + #include -#include -#include -#include -#include "categoricals.h" + +#include "data/case.h" +#include "data/variable.h" +#include "libpspp/assertion.h" +#include "libpspp/misc.h" +#include "math/categoricals.h" +#include "math/interaction.h" +#include "math/moments.h" + +#include "gl/xalloc.h" #define n_MOMENTS (MOMENT_VARIANCE + 1) @@ -58,7 +62,7 @@ resize_matrix (gsl_matrix *in, size_t new_size) gsl_matrix_set (out, i, j, x); } } - + gsl_matrix_free (in); return out; @@ -66,6 +70,9 @@ resize_matrix (gsl_matrix *in, size_t new_size) struct covariance { + /* True if the covariances are centerered. (ie Real covariances) */ + bool centered; + /* The variables for which the covariance matrix is to be calculated. */ size_t n_vars; const struct variable *const *vars; @@ -93,7 +100,7 @@ struct covariance double *cm; int n_cm; - /* 1 for single pass algorithm; + /* 1 for single pass algorithm; 2 for double pass algorithm */ short passes; @@ -101,14 +108,16 @@ struct covariance /* 0 : No pass has been made 1 : First pass has been started - 2 : Second pass has been - + 2 : Second pass has been + IE: How many passes have been (partially) made. */ short state; /* Flags indicating that the first case has been seen */ bool pass_one_first_case_seen; bool pass_two_first_case_seen; + + gsl_matrix *unnormalised; }; @@ -132,15 +141,17 @@ covariance_moments (const struct covariance *cov, int m) */ struct covariance * covariance_1pass_create (size_t n_vars, const struct variable *const *vars, - const struct variable *weight, enum mv_class exclude) + const struct variable *weight, enum mv_class exclude, + bool centered) { size_t i; struct covariance *cov = xzalloc (sizeof *cov); + cov->centered = centered; cov->passes = 1; cov->state = 0; cov->pass_one_first_case_seen = cov->pass_two_first_case_seen = false; - + cov->vars = vars; cov->wv = weight; @@ -148,7 +159,7 @@ covariance_1pass_create (size_t n_vars, const struct variable *const *vars, cov->dim = n_vars; cov->moments = xmalloc (sizeof *cov->moments * n_MOMENTS); - + for (i = 0; i < n_MOMENTS; ++i) cov->moments[i] = gsl_matrix_calloc (n_vars, n_vars); @@ -156,8 +167,8 @@ covariance_1pass_create (size_t n_vars, const struct variable *const *vars, cov->n_cm = (n_vars * (n_vars - 1) ) / 2; - if (cov->n_cm > 0) - cov->cm = xcalloc (sizeof *cov->cm, cov->n_cm); + + cov->cm = xcalloc (cov->n_cm, sizeof *cov->cm); cov->categoricals = NULL; return cov; @@ -171,16 +182,18 @@ covariance_1pass_create (size_t n_vars, const struct variable *const *vars, */ struct covariance * covariance_2pass_create (size_t n_vars, const struct variable *const *vars, - size_t n_catvars, const struct variable *const *catvars, - const struct variable *wv, enum mv_class exclude) + struct categoricals *cats, + const struct variable *wv, enum mv_class exclude, + bool centered) { size_t i; struct covariance *cov = xmalloc (sizeof *cov); + cov->centered = centered; cov->passes = 2; cov->state = 0; cov->pass_one_first_case_seen = cov->pass_two_first_case_seen = false; - + cov->vars = vars; cov->wv = wv; @@ -188,7 +201,7 @@ covariance_2pass_create (size_t n_vars, const struct variable *const *vars, cov->dim = n_vars; cov->moments = xmalloc (sizeof *cov->moments * n_MOMENTS); - + for (i = 0; i < n_MOMENTS; ++i) cov->moments[i] = gsl_matrix_calloc (n_vars, n_vars); @@ -197,12 +210,13 @@ covariance_2pass_create (size_t n_vars, const struct variable *const *vars, cov->n_cm = -1; cov->cm = NULL; - cov->categoricals = categoricals_create (catvars, n_catvars, wv, exclude); + cov->categoricals = cats; + cov->unnormalised = NULL; return cov; } -/* Return an integer, which can be used to index +/* Return an integer, which can be used to index into COV->cm, to obtain the I, J th element of the covariance matrix. If COV->cm does not contain that element, then a negative value @@ -214,7 +228,7 @@ cm_idx (const struct covariance *cov, int i, int j) int as; const int n2j = cov->dim - 2 - j; const int nj = cov->dim - 2 ; - + assert (i >= 0); assert (j < cov->dim); @@ -224,11 +238,11 @@ cm_idx (const struct covariance *cov, int i, int j) if (j >= cov->dim - 1) return -1; - if ( i <= j) + if ( i <= j) return -1 ; as = nj * (nj + 1) ; - as -= n2j * (n2j + 1) ; + as -= n2j * (n2j + 1) ; as /= 2; return i - 1 + as; @@ -236,15 +250,15 @@ cm_idx (const struct covariance *cov, int i, int j) /* - Returns true iff the variable corresponding to the Ith element of the covariance matrix + Returns true iff the variable corresponding to the Ith element of the covariance matrix has a missing value for case C */ static bool is_missing (const struct covariance *cov, int i, const struct ccase *c) { const struct variable *var = i < cov->n_vars ? - cov->vars[i] : - categoricals_get_variable_by_subscript (cov->categoricals, i - cov->n_vars); + cov->vars[i] : + categoricals_get_interaction_by_subscript (cov->categoricals, i - cov->n_vars)->vars[0]; const union value *val = case_data (c, var); @@ -264,7 +278,7 @@ get_val (const struct covariance *cov, int i, const struct ccase *c) return val->f; } - return categoricals_get_binary_by_subscript (cov->categoricals, i - cov->n_vars, c); + return categoricals_get_effects_code_for_case (cov->categoricals, i - cov->n_vars, c); } #if 0 @@ -296,7 +310,8 @@ 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) { @@ -342,11 +357,16 @@ covariance_accumulate_pass2 (struct covariance *cov, const struct ccase *c) assert (cov->state == 1); cov->state = 2; - cov->dim = cov->n_vars + - categoricals_total (cov->categoricals) - categoricals_get_n_variables (cov->categoricals); + if (cov->categoricals) + categoricals_done (cov->categoricals); + + cov->dim = cov->n_vars; + + if (cov->categoricals) + cov->dim += categoricals_df_total (cov->categoricals); cov->n_cm = (cov->dim * (cov->dim - 1) ) / 2; - cov->cm = xcalloc (sizeof *cov->cm, cov->n_cm); + cov->cm = xcalloc (cov->n_cm, sizeof *cov->cm); /* Grow the moment matrices so that they're large enough to accommodate the categorical elements */ @@ -355,8 +375,6 @@ covariance_accumulate_pass2 (struct covariance *cov, const struct ccase *c) cov->moments[i] = resize_matrix (cov->moments[i], cov->dim); } - categoricals_done (cov->categoricals); - /* Populate the moments matrices with the categorical value elements */ for (i = cov->n_vars; i < cov->dim; ++i) { @@ -420,9 +438,9 @@ covariance_accumulate_pass2 (struct covariance *cov, const struct ccase *c) *x += s; } - ss = + ss = (v1 - gsl_matrix_get (cov->moments[MOMENT_MEAN], i, j)) - * + * (v2 - gsl_matrix_get (cov->moments[MOMENT_MEAN], i, j)) * weight ; @@ -492,7 +510,7 @@ covariance_accumulate (struct covariance *cov, const struct ccase *c) } -/* +/* Allocate and return a gsl_matrix containing the covariances of the data. */ @@ -524,7 +542,7 @@ cm_to_gsl (struct covariance *cov) } -static const gsl_matrix * +static gsl_matrix * covariance_calculate_double_pass (struct covariance *cov) { size_t i, j; @@ -548,7 +566,7 @@ covariance_calculate_double_pass (struct covariance *cov) return cm_to_gsl (cov); } -static const gsl_matrix * +static gsl_matrix * covariance_calculate_single_pass (struct covariance *cov) { size_t i, j; @@ -573,19 +591,22 @@ covariance_calculate_single_pass (struct covariance *cov) } } - /* Centre the moments */ - for ( j = 0 ; j < cov->dim - 1; ++j) + if (cov->centered) { - for (i = j + 1 ; i < cov->dim; ++i) + /* Centre the moments */ + for ( j = 0 ; j < cov->dim - 1; ++j) { - double *x = &cov->cm [cm_idx (cov, i, j)]; - - *x /= gsl_matrix_get (cov->moments[0], i, 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); + *x /= gsl_matrix_get (cov->moments[0], i, j); + + *x -= + gsl_matrix_get (cov->moments[MOMENT_MEAN], i, j) + * + gsl_matrix_get (cov->moments[MOMENT_MEAN], j, i); + } } } @@ -593,12 +614,12 @@ 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. - Call this function only after all data have been accumulated. -*/ -const gsl_matrix * +/* Return a pointer to gsl_matrix containing the pairwise covariances. The + caller owns the returned matrix and must free it when it is no longer + needed. + + Call this function only after all data have been accumulated. */ +gsl_matrix * covariance_calculate (struct covariance *cov) { if ( cov->state <= 0 ) @@ -607,10 +628,10 @@ covariance_calculate (struct covariance *cov) switch (cov->passes) { case 1: - return covariance_calculate_single_pass (cov); + return covariance_calculate_single_pass (cov); break; case 2: - return covariance_calculate_double_pass (cov); + return covariance_calculate_double_pass (cov); break; default: NOT_REACHED (); @@ -620,82 +641,74 @@ covariance_calculate (struct covariance *cov) /* Covariance computed without dividing by the sample size. */ -static const gsl_matrix * +static 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 * +static gsl_matrix * covariance_calculate_single_pass_unnormalized (struct covariance *cov) { size_t i, j; - for (i = 0 ; i < cov->dim; ++i) + if (cov->centered) { - for (j = 0 ; j < cov->dim; ++j) + for (i = 0 ; i < cov->dim; ++i) { - 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; ++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) + + for ( j = 0 ; j < cov->dim - 1; ++j) { - 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); + 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. -*/ +/* Return a pointer to gsl_matrix containing the pairwise covariances. The + returned matrix is owned by the structure, 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; + if (cov->unnormalised != NULL) + return cov->unnormalised; + switch (cov->passes) { case 1: - return covariance_calculate_single_pass_unnormalized (cov); + cov->unnormalised = covariance_calculate_single_pass_unnormalized (cov); break; case 2: - return covariance_calculate_double_pass_unnormalized (cov); + cov->unnormalised = covariance_calculate_double_pass_unnormalized (cov); break; default: NOT_REACHED (); } + + return cov->unnormalised; } /* Function to access the categoricals used by COV @@ -719,7 +732,101 @@ covariance_destroy (struct covariance *cov) for (i = 0; i < n_MOMENTS; ++i) gsl_matrix_free (cov->moments[i]); + gsl_matrix_free (cov->unnormalised); free (cov->moments); free (cov->cm); free (cov); } + +size_t +covariance_dim (const struct covariance * cov) +{ + return (cov->dim); +} + + + +/* + Routines to assist debugging. + The following are not thoroughly tested and in certain respects + unreliable. They should only be + used for aids to development. Not as user accessible code. +*/ + +#include "libpspp/str.h" +#include "output/tab.h" +#include "data/format.h" + + +/* Create a table which can be populated with the encodings for + the covariance matrix COV */ +struct tab_table * +covariance_dump_enc_header (const struct covariance *cov, int length) +{ + struct tab_table *t = tab_create (cov->dim, length); + int n; + int i; + + tab_title (t, "Covariance Encoding"); + + tab_box (t, + TAL_2, TAL_2, 0, 0, + 0, 0, tab_nc (t) - 1, tab_nr (t) - 1); + + tab_hline (t, TAL_2, 0, tab_nc (t) - 1, 1); + + + for (i = 0 ; i < cov->n_vars; ++i) + { + tab_text (t, i, 0, TAT_TITLE, var_get_name (cov->vars[i])); + tab_vline (t, TAL_1, i + 1, 0, tab_nr (t) - 1); + } + + n = 0; + while (i < cov->dim) + { + struct string str; + int idx = i - cov->n_vars; + const struct interaction *iact = + categoricals_get_interaction_by_subscript (cov->categoricals, idx); + int df; + + ds_init_empty (&str); + interaction_to_string (iact, &str); + + df = categoricals_df (cov->categoricals, n); + + tab_joint_text (t, + i, 0, + i + df - 1, 0, + TAT_TITLE, ds_cstr (&str)); + + if (i + df < tab_nr (t) - 1) + tab_vline (t, TAL_1, i + df, 0, tab_nr (t) - 1); + + i += df; + n++; + ds_destroy (&str); + } + + return t; +} + + +/* + Append table T, which should have been returned by covariance_dump_enc_header + with an entry corresponding to case C for the covariance matrix COV + */ +void +covariance_dump_enc (const struct covariance *cov, const struct ccase *c, + struct tab_table *t) +{ + static int row = 0; + int i; + ++row; + for (i = 0 ; i < cov->dim; ++i) + { + double v = get_val (cov, i, c); + tab_double (t, i, row, 0, v, i < cov->n_vars ? NULL : &F_8_0, RC_OTHER); + } +}