--- /dev/null
- struct tab_table *t = tab_create (nc, nr, 0);
+ /* PSPP - a program for statistical analysis.
+ Copyright (C) 2009 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
+ the Free Software Foundation, either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>. */
+
+ #include <config.h>
+
+ #include <libpspp/assertion.h>
+ #include <math/covariance.h>
+ #include <math/design-matrix.h>
+ #include <gsl/gsl_matrix.h>
+ #include <data/casegrouper.h>
+ #include <data/casereader.h>
+ #include <data/dictionary.h>
+ #include <data/procedure.h>
+ #include <data/variable.h>
+ #include <language/command.h>
+ #include <language/dictionary/split-file.h>
+ #include <language/lexer/lexer.h>
+ #include <language/lexer/variable-parser.h>
+ #include <output/manager.h>
+ #include <output/table.h>
+ #include <libpspp/message.h>
+ #include <data/format.h>
+ #include <math/moments.h>
+
+ #include <math.h>
+ #include "xalloc.h"
+ #include "minmax.h"
+ #include <libpspp/misc.h>
+ #include <gsl/gsl_cdf.h>
+
+ #include "gettext.h"
+ #define _(msgid) gettext (msgid)
+ #define N_(msgid) msgid
+
+
+ static double
+ significance_of_correlation (double rho, double w)
+ {
+ double t = w - 2;
+ t /= 1 - MIN (1, pow2 (rho));
+ t = sqrt (t);
+ t *= rho;
+
+ if (t > 0)
+ return gsl_cdf_tdist_Q (t, w - 2);
+ else
+ return gsl_cdf_tdist_P (t, w - 2);
+ }
+
+
+ struct corr
+ {
+ size_t n_vars_total;
+ size_t n_vars1;
+
+ const struct variable **vars;
+ };
+
+
+ /* Handling of missing values. */
+ enum corr_missing_type
+ {
+ CORR_PAIRWISE, /* Handle missing values on a per-variable-pair basis. */
+ CORR_LISTWISE /* Discard entire case if any variable is missing. */
+ };
+
+ enum stats_opts
+ {
+ STATS_DESCRIPTIVES = 0x01,
+ STATS_XPROD = 0x02,
+ STATS_ALL = STATS_XPROD | STATS_DESCRIPTIVES
+ };
+
+ struct corr_opts
+ {
+ enum corr_missing_type missing_type;
+ enum mv_class exclude; /* Classes of missing values to exclude. */
+
+ bool sig; /* Flag significant values or not */
+ int tails; /* Report significance with how many tails ? */
+ enum stats_opts statistics;
+
+ const struct variable *wv; /* The weight variable (if any) */
+ };
+
+
+ static void
+ output_descriptives (const struct corr *corr, const gsl_matrix *means,
+ const gsl_matrix *vars, const gsl_matrix *ns)
+ {
+ const int nr = corr->n_vars_total + 1;
+ const int nc = 4;
+ int c, r;
+
+ const int heading_columns = 1;
+ const int heading_rows = 1;
+
- tab_dim (t, tab_natural_dimensions, NULL);
++ struct tab_table *t = tab_create (nc, nr);
+ tab_title (t, _("Descriptive Statistics"));
- t = tab_create (nc, nr, 0);
++ tab_dim (t, tab_natural_dimensions, NULL, NULL);
+
+ tab_headers (t, heading_columns, 0, heading_rows, 0);
+
+ /* Outline the box */
+ tab_box (t,
+ TAL_2, TAL_2,
+ -1, -1,
+ 0, 0,
+ nc - 1, nr - 1);
+
+ /* Vertical lines */
+ tab_box (t,
+ -1, -1,
+ -1, TAL_1,
+ heading_columns, 0,
+ nc - 1, nr - 1);
+
+ tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
+ tab_hline (t, TAL_1, 0, nc - 1, heading_rows);
+
+ tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
+ tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
+ tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
+
+ for (r = 0 ; r < corr->n_vars_total ; ++r)
+ {
+ const struct variable *v = corr->vars[r];
+ tab_text (t, 0, r + heading_rows, TAB_LEFT | TAT_TITLE, var_to_string (v));
+
+ for (c = 1 ; c < nc ; ++c)
+ {
+ double x ;
+ double n;
+ switch (c)
+ {
+ case 1:
+ x = gsl_matrix_get (means, r, 0);
+ break;
+ case 2:
+ x = gsl_matrix_get (vars, r, 0);
+
+ /* Here we want to display the non-biased estimator */
+ n = gsl_matrix_get (ns, r, 0);
+ x *= n / (n -1);
+
+ x = sqrt (x);
+ break;
+ case 3:
+ x = gsl_matrix_get (ns, r, 0);
+ break;
+ default:
+ NOT_REACHED ();
+ };
+
+ tab_double (t, c, r + heading_rows, 0, x, NULL);
+ }
+ }
+
+ tab_submit (t);
+ }
+
+ static void
+ output_correlation (const struct corr *corr, const struct corr_opts *opts,
+ const gsl_matrix *cm, const gsl_matrix *samples,
+ const gsl_matrix *cv)
+ {
+ int r, c;
+ struct tab_table *t;
+ int matrix_cols;
+ int nr = corr->n_vars1;
+ int nc = matrix_cols = corr->n_vars_total > corr->n_vars1 ?
+ corr->n_vars_total - corr->n_vars1 : corr->n_vars1;
+
+ const struct fmt_spec *wfmt = opts->wv ? var_get_print_format (opts->wv) : & F_8_0;
+
+ const int heading_columns = 2;
+ const int heading_rows = 1;
+
+ int rows_per_variable = opts->missing_type == CORR_LISTWISE ? 2 : 3;
+
+ if (opts->statistics & STATS_XPROD)
+ rows_per_variable += 2;
+
+ /* Two header columns */
+ nc += heading_columns;
+
+ /* Three data per variable */
+ nr *= rows_per_variable;
+
+ /* One header row */
+ nr += heading_rows;
+
- tab_dim (t, tab_natural_dimensions, NULL);
++ t = tab_create (nc, nr);
+ tab_title (t, _("Correlations"));
++ tab_dim (t, tab_natural_dimensions, NULL, NULL);
+
+ tab_headers (t, heading_columns, 0, heading_rows, 0);
+
+ /* Outline the box */
+ tab_box (t,
+ TAL_2, TAL_2,
+ -1, -1,
+ 0, 0,
+ nc - 1, nr - 1);
+
+ /* Vertical lines */
+ tab_box (t,
+ -1, -1,
+ -1, TAL_1,
+ heading_columns, 0,
+ nc - 1, nr - 1);
+
+ tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
+ tab_vline (t, TAL_1, 1, heading_rows, nr - 1);
+
+ for (r = 0 ; r < corr->n_vars1 ; ++r)
+ {
+ tab_text (t, 0, 1 + r * rows_per_variable, TAB_LEFT | TAT_TITLE,
+ var_to_string (corr->vars[r]));
+
+ tab_text (t, 1, 1 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Pearson Correlation"));
+ tab_text (t, 1, 2 + r * rows_per_variable, TAB_LEFT | TAT_TITLE,
+ (opts->tails == 2) ? _("Sig. (2-tailed)") : _("Sig. (1-tailed)"));
+
+ if (opts->statistics & STATS_XPROD)
+ {
+ tab_text (t, 1, 3 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Cross-products"));
+ tab_text (t, 1, 4 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Covariance"));
+ }
+
+ if ( opts->missing_type != CORR_LISTWISE )
+ tab_text (t, 1, rows_per_variable + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("N"));
+
+ tab_hline (t, TAL_1, 0, nc - 1, r * rows_per_variable + 1);
+ }
+
+ for (c = 0 ; c < matrix_cols ; ++c)
+ {
+ const struct variable *v = corr->n_vars_total > corr->n_vars1 ? corr->vars[corr->n_vars_total - corr->n_vars1 + c] : corr->vars[c];
+ tab_text (t, heading_columns + c, 0, TAB_LEFT | TAT_TITLE, var_to_string (v));
+ }
+
+ for (r = 0 ; r < corr->n_vars1 ; ++r)
+ {
+ const int row = r * rows_per_variable + heading_rows;
+ for (c = 0 ; c < matrix_cols ; ++c)
+ {
+ unsigned char flags = 0;
+ const int col_index = corr->n_vars_total - corr->n_vars1 + c;
+ double pearson = gsl_matrix_get (cm, r, col_index);
+ double w = gsl_matrix_get (samples, r, col_index);
+ double sig = opts->tails * significance_of_correlation (pearson, w);
+
+ if ( opts->missing_type != CORR_LISTWISE )
+ tab_double (t, c + heading_columns, row + rows_per_variable - 1, 0, w, wfmt);
+
+ if ( c != r)
+ tab_double (t, c + heading_columns, row + 1, 0, sig, NULL);
+
+ if ( opts->sig && c != r && sig < 0.05)
+ flags = TAB_EMPH;
+
+ tab_double (t, c + heading_columns, row, flags, pearson, NULL);
+
+ if (opts->statistics & STATS_XPROD)
+ {
+ double cov = gsl_matrix_get (cv, r, col_index);
+ const double xprod_dev = cov * w;
+ cov *= w / (w - 1.0);
+
+ tab_double (t, c + heading_columns, row + 2, 0, xprod_dev, NULL);
+ tab_double (t, c + heading_columns, row + 3, 0, cov, NULL);
+ }
+ }
+ }
+
+ tab_submit (t);
+ }
+
+
+ static gsl_matrix *
+ correlation_from_covariance (const gsl_matrix *cv, const gsl_matrix *v)
+ {
+ size_t i, j;
+ gsl_matrix *corr = gsl_matrix_calloc (cv->size1, cv->size2);
+
+ for (i = 0 ; i < cv->size1; ++i)
+ {
+ for (j = 0 ; j < cv->size2; ++j)
+ {
+ double rho = gsl_matrix_get (cv, i, j);
+
+ rho /= sqrt (gsl_matrix_get (v, i, j))
+ *
+ sqrt (gsl_matrix_get (v, j, i));
+
+ gsl_matrix_set (corr, i, j, rho);
+ }
+ }
+
+ return corr;
+ }
+
+
+
+
+ static void
+ run_corr (struct casereader *r, const struct corr_opts *opts, const struct corr *corr)
+ {
+ struct ccase *c;
+ const gsl_matrix *var_matrix, *samples_matrix, *mean_matrix;
+ const gsl_matrix *cov_matrix;
+ gsl_matrix *corr_matrix;
+ struct covariance *cov = covariance_create (corr->n_vars_total, corr->vars,
+ opts->wv, opts->exclude);
+
+ for ( ; (c = casereader_read (r) ); case_unref (c))
+ {
+ covariance_accumulate (cov, c);
+ }
+
+ cov_matrix = covariance_calculate (cov);
+
+ samples_matrix = covariance_moments (cov, MOMENT_NONE);
+ var_matrix = covariance_moments (cov, MOMENT_VARIANCE);
+ mean_matrix = covariance_moments (cov, MOMENT_MEAN);
+
+ corr_matrix = correlation_from_covariance (cov_matrix, var_matrix);
+
+ if ( opts->statistics & STATS_DESCRIPTIVES)
+ output_descriptives (corr, mean_matrix, var_matrix, samples_matrix);
+
+ output_correlation (corr, opts,
+ corr_matrix,
+ samples_matrix,
+ cov_matrix);
+
+ covariance_destroy (cov);
+ gsl_matrix_free (corr_matrix);
+ }
+
+ int
+ cmd_correlation (struct lexer *lexer, struct dataset *ds)
+ {
+ int i;
+ int n_all_vars = 0; /* Total number of variables involved in this command */
+ const struct variable **all_vars ;
+ const struct dictionary *dict = dataset_dict (ds);
+ bool ok = true;
+
+ struct casegrouper *grouper;
+ struct casereader *group;
+
+ struct corr *corr = NULL;
+ size_t n_corrs = 0;
+
+ struct corr_opts opts;
+ opts.missing_type = CORR_PAIRWISE;
+ opts.wv = dict_get_weight (dict);
+ opts.tails = 2;
+ opts.sig = false;
+ opts.exclude = MV_ANY;
+ opts.statistics = 0;
+
+ /* Parse CORRELATIONS. */
+ while (lex_token (lexer) != '.')
+ {
+ lex_match (lexer, '/');
+ if (lex_match_id (lexer, "MISSING"))
+ {
+ lex_match (lexer, '=');
+ while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
+ {
+ if (lex_match_id (lexer, "PAIRWISE"))
+ opts.missing_type = CORR_PAIRWISE;
+ else if (lex_match_id (lexer, "LISTWISE"))
+ opts.missing_type = CORR_LISTWISE;
+
+ else if (lex_match_id (lexer, "INCLUDE"))
+ opts.exclude = MV_SYSTEM;
+ else if (lex_match_id (lexer, "EXCLUDE"))
+ opts.exclude = MV_ANY;
+ else
+ {
+ lex_error (lexer, NULL);
+ goto error;
+ }
+ lex_match (lexer, ',');
+ }
+ }
+ else if (lex_match_id (lexer, "PRINT"))
+ {
+ lex_match (lexer, '=');
+ while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
+ {
+ if ( lex_match_id (lexer, "TWOTAIL"))
+ opts.tails = 2;
+ else if (lex_match_id (lexer, "ONETAIL"))
+ opts.tails = 1;
+ else if (lex_match_id (lexer, "SIG"))
+ opts.sig = false;
+ else if (lex_match_id (lexer, "NOSIG"))
+ opts.sig = true;
+ else
+ {
+ lex_error (lexer, NULL);
+ goto error;
+ }
+
+ lex_match (lexer, ',');
+ }
+ }
+ else if (lex_match_id (lexer, "STATISTICS"))
+ {
+ lex_match (lexer, '=');
+ while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
+ {
+ if ( lex_match_id (lexer, "DESCRIPTIVES"))
+ opts.statistics = STATS_DESCRIPTIVES;
+ else if (lex_match_id (lexer, "XPROD"))
+ opts.statistics = STATS_XPROD;
+ else if (lex_token (lexer) == T_ALL)
+ {
+ opts.statistics = STATS_ALL;
+ lex_get (lexer);
+ }
+ else
+ {
+ lex_error (lexer, NULL);
+ goto error;
+ }
+
+ lex_match (lexer, ',');
+ }
+ }
+ else
+ {
+ if (lex_match_id (lexer, "VARIABLES"))
+ {
+ lex_match (lexer, '=');
+ }
+
+ corr = xrealloc (corr, sizeof (*corr) * (n_corrs + 1));
+ corr[n_corrs].n_vars_total = corr[n_corrs].n_vars1 = 0;
+
+ if ( ! parse_variables_const (lexer, dict, &corr[n_corrs].vars,
+ &corr[n_corrs].n_vars_total,
+ PV_NUMERIC))
+ {
+ ok = false;
+ break;
+ }
+
+
+ corr[n_corrs].n_vars1 = corr[n_corrs].n_vars_total;
+
+ if ( lex_match (lexer, T_WITH))
+ {
+ if ( ! parse_variables_const (lexer, dict,
+ &corr[n_corrs].vars, &corr[n_corrs].n_vars_total,
+ PV_NUMERIC | PV_APPEND))
+ {
+ ok = false;
+ break;
+ }
+ }
+
+ n_all_vars += corr[n_corrs].n_vars_total;
+
+ n_corrs++;
+ }
+ }
+
+ if (n_corrs == 0)
+ {
+ msg (SE, _("No variables specified."));
+ goto error;
+ }
+
+
+ all_vars = xmalloc (sizeof (*all_vars) * n_all_vars);
+
+ {
+ /* FIXME: Using a hash here would make more sense */
+ const struct variable **vv = all_vars;
+
+ for (i = 0 ; i < n_corrs; ++i)
+ {
+ int v;
+ const struct corr *c = &corr[i];
+ for (v = 0 ; v < c->n_vars_total; ++v)
+ *vv++ = c->vars[v];
+ }
+ }
+
+ grouper = casegrouper_create_splits (proc_open (ds), dict);
+
+ while (casegrouper_get_next_group (grouper, &group))
+ {
+ for (i = 0 ; i < n_corrs; ++i)
+ {
+ /* FIXME: No need to iterate the data multiple times */
+ struct casereader *r = casereader_clone (group);
+
+ if ( opts.missing_type == CORR_LISTWISE)
+ r = casereader_create_filter_missing (r, all_vars, n_all_vars,
+ opts.exclude, NULL, NULL);
+
+
+ run_corr (r, &opts, &corr[i]);
+ casereader_destroy (r);
+ }
+ casereader_destroy (group);
+ }
+
+ ok = casegrouper_destroy (grouper);
+ ok = proc_commit (ds) && ok;
+
+ free (all_vars);
+
+
+ /* Done. */
+ free (corr);
+ return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
+
+ error:
+ free (corr);
+ return CMD_FAILURE;
+ }
projects / pspp / commitdiff