1 /* PSPP - linear regression.
2 Copyright (C) 2005 Free Software Foundation, Inc.
3 Written by Jason H Stover <jason@sakla.net>.
5 This program is free software; you can redistribute it and/or
6 modify it under the terms of the GNU General Public License as
7 published by the Free Software Foundation; either version 2 of the
8 License, or (at your option) any later version.
10 This program is distributed in the hope that it will be useful, but
11 WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with this program; if not, write to the Free Software
17 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
22 #include <gsl/gsl_cdf.h>
23 #include <gsl/gsl_vector.h>
24 #include <gsl/gsl_matrix.h>
29 #include "cat-routines.h"
31 #include "design-matrix.h"
32 #include "dictionary.h"
34 #include "file-handle.h"
37 #include <linreg/pspp_linreg.h>
38 #include "missing-values.h"
39 #include "regression_export.h"
41 #include "value-labels.h"
45 #define REG_LARGE_DATA 1000
50 "REGRESSION" (regression_):
75 static struct cmd_regression cmd;
78 Array holding the subscripts of the independent variables.
83 File where the model will be saved if the EXPORT subcommand
86 struct file_handle *model_file;
89 Return value for the procedure.
91 int pspp_reg_rc = CMD_SUCCESS;
93 static void run_regression (const struct casefile *, void *);
95 STATISTICS subcommand output functions.
97 static void reg_stats_r (pspp_linreg_cache *);
98 static void reg_stats_coeff (pspp_linreg_cache *);
99 static void reg_stats_anova (pspp_linreg_cache *);
100 static void reg_stats_outs (pspp_linreg_cache *);
101 static void reg_stats_zpp (pspp_linreg_cache *);
102 static void reg_stats_label (pspp_linreg_cache *);
103 static void reg_stats_sha (pspp_linreg_cache *);
104 static void reg_stats_ci (pspp_linreg_cache *);
105 static void reg_stats_f (pspp_linreg_cache *);
106 static void reg_stats_bcov (pspp_linreg_cache *);
107 static void reg_stats_ses (pspp_linreg_cache *);
108 static void reg_stats_xtx (pspp_linreg_cache *);
109 static void reg_stats_collin (pspp_linreg_cache *);
110 static void reg_stats_tol (pspp_linreg_cache *);
111 static void reg_stats_selection (pspp_linreg_cache *);
112 static void statistics_keyword_output (void (*)(pspp_linreg_cache *),
113 int, pspp_linreg_cache *);
116 reg_stats_r (pspp_linreg_cache * c)
126 rsq = c->ssm / c->sst;
127 adjrsq = 1.0 - (1.0 - rsq) * (c->n_obs - 1.0) / (c->n_obs - c->n_indeps);
128 std_error = sqrt ((c->n_indeps - 1.0) / (c->n_obs - 1.0));
129 t = tab_create (n_cols, n_rows, 0);
130 tab_dim (t, tab_natural_dimensions);
131 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
132 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
133 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
134 tab_vline (t, TAL_0, 1, 0, 0);
136 tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("R"));
137 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("R Square"));
138 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Adjusted R Square"));
139 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Error of the Estimate"));
140 tab_float (t, 1, 1, TAB_RIGHT, sqrt (rsq), 10, 2);
141 tab_float (t, 2, 1, TAB_RIGHT, rsq, 10, 2);
142 tab_float (t, 3, 1, TAB_RIGHT, adjrsq, 10, 2);
143 tab_float (t, 4, 1, TAB_RIGHT, std_error, 10, 2);
144 tab_title (t, 0, _("Model Summary"));
149 Table showing estimated regression coefficients.
152 reg_stats_coeff (pspp_linreg_cache * c)
167 n_rows = c->n_coeffs + 2;
169 t = tab_create (n_cols, n_rows, 0);
170 tab_headers (t, 2, 0, 1, 0);
171 tab_dim (t, tab_natural_dimensions);
172 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
173 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
174 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
175 tab_vline (t, TAL_0, 1, 0, 0);
177 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("B"));
178 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Error"));
179 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Beta"));
180 tab_text (t, 5, 0, TAB_CENTER | TAT_TITLE, _("t"));
181 tab_text (t, 6, 0, TAB_CENTER | TAT_TITLE, _("Significance"));
182 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("(Constant)"));
183 coeff = c->coeff[0].estimate;
184 tab_float (t, 2, 1, 0, coeff, 10, 2);
185 std_err = sqrt (gsl_matrix_get (c->cov, 0, 0));
186 tab_float (t, 3, 1, 0, std_err, 10, 2);
187 beta = coeff / c->depvar_std;
188 tab_float (t, 4, 1, 0, beta, 10, 2);
189 t_stat = coeff / std_err;
190 tab_float (t, 5, 1, 0, t_stat, 10, 2);
191 pval = 2 * gsl_cdf_tdist_Q (fabs (t_stat), 1.0);
192 tab_float (t, 6, 1, 0, pval, 10, 2);
193 for (j = 1; j <= c->n_indeps; j++)
196 label = var_to_string (c->coeff[j].v);
197 tab_text (t, 1, j + 1, TAB_CENTER, label);
199 Regression coefficients.
201 coeff = c->coeff[j].estimate;
202 tab_float (t, 2, j + 1, 0, coeff, 10, 2);
204 Standard error of the coefficients.
206 std_err = sqrt (gsl_matrix_get (c->cov, j, j));
207 tab_float (t, 3, j + 1, 0, std_err, 10, 2);
209 'Standardized' coefficient, i.e., regression coefficient
210 if all variables had unit variance.
212 beta = gsl_vector_get (c->indep_std, j);
213 beta *= coeff / c->depvar_std;
214 tab_float (t, 4, j + 1, 0, beta, 10, 2);
217 Test statistic for H0: coefficient is 0.
219 t_stat = coeff / std_err;
220 tab_float (t, 5, j + 1, 0, t_stat, 10, 2);
222 P values for the test statistic above.
224 pval = 2 * gsl_cdf_tdist_Q (fabs (t_stat), 1.0);
225 tab_float (t, 6, j + 1, 0, pval, 10, 2);
227 tab_title (t, 0, _("Coefficients"));
232 Display the ANOVA table.
235 reg_stats_anova (pspp_linreg_cache * c)
239 const double msm = c->ssm / c->dfm;
240 const double mse = c->sse / c->dfe;
241 const double F = msm / mse;
242 const double pval = gsl_cdf_fdist_Q (F, c->dfm, c->dfe);
247 t = tab_create (n_cols, n_rows, 0);
248 tab_headers (t, 2, 0, 1, 0);
249 tab_dim (t, tab_natural_dimensions);
251 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
253 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
254 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
255 tab_vline (t, TAL_0, 1, 0, 0);
257 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Sum of Squares"));
258 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("df"));
259 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Mean Square"));
260 tab_text (t, 5, 0, TAB_CENTER | TAT_TITLE, _("F"));
261 tab_text (t, 6, 0, TAB_CENTER | TAT_TITLE, _("Significance"));
263 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Regression"));
264 tab_text (t, 1, 2, TAB_LEFT | TAT_TITLE, _("Residual"));
265 tab_text (t, 1, 3, TAB_LEFT | TAT_TITLE, _("Total"));
267 /* Sums of Squares */
268 tab_float (t, 2, 1, 0, c->ssm, 10, 2);
269 tab_float (t, 2, 3, 0, c->sst, 10, 2);
270 tab_float (t, 2, 2, 0, c->sse, 10, 2);
273 /* Degrees of freedom */
274 tab_float (t, 3, 1, 0, c->dfm, 4, 0);
275 tab_float (t, 3, 2, 0, c->dfe, 4, 0);
276 tab_float (t, 3, 3, 0, c->dft, 4, 0);
280 tab_float (t, 4, 1, TAB_RIGHT, msm, 8, 3);
281 tab_float (t, 4, 2, TAB_RIGHT, mse, 8, 3);
283 tab_float (t, 5, 1, 0, F, 8, 3);
285 tab_float (t, 6, 1, 0, pval, 8, 3);
287 tab_title (t, 0, _("ANOVA"));
291 reg_stats_outs (pspp_linreg_cache * c)
296 reg_stats_zpp (pspp_linreg_cache * c)
301 reg_stats_label (pspp_linreg_cache * c)
306 reg_stats_sha (pspp_linreg_cache * c)
311 reg_stats_ci (pspp_linreg_cache * c)
316 reg_stats_f (pspp_linreg_cache * c)
321 reg_stats_bcov (pspp_linreg_cache * c)
334 n_cols = c->n_indeps + 1 + 2;
335 n_rows = 2 * (c->n_indeps + 1);
336 t = tab_create (n_cols, n_rows, 0);
337 tab_headers (t, 2, 0, 1, 0);
338 tab_dim (t, tab_natural_dimensions);
339 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
340 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
341 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
342 tab_vline (t, TAL_0, 1, 0, 0);
343 tab_text (t, 0, 0, TAB_CENTER | TAT_TITLE, _("Model"));
344 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Covariances"));
345 for (i = 1; i < c->n_indeps + 1; i++)
347 j = indep_vars[(i - 1)];
348 struct variable *v = cmd.v_variables[j];
349 label = var_to_string (v);
350 tab_text (t, 2, i, TAB_CENTER, label);
351 tab_text (t, i + 2, 0, TAB_CENTER, label);
352 for (k = 1; k < c->n_indeps + 1; k++)
354 col = (i <= k) ? k : i;
355 row = (i <= k) ? i : k;
356 tab_float (t, k + 2, i, TAB_CENTER,
357 gsl_matrix_get (c->cov, row, col), 8, 3);
360 tab_title (t, 0, _("Coefficient Correlations"));
364 reg_stats_ses (pspp_linreg_cache * c)
369 reg_stats_xtx (pspp_linreg_cache * c)
374 reg_stats_collin (pspp_linreg_cache * c)
379 reg_stats_tol (pspp_linreg_cache * c)
384 reg_stats_selection (pspp_linreg_cache * c)
390 statistics_keyword_output (void (*function) (pspp_linreg_cache *),
391 int keyword, pspp_linreg_cache * c)
400 subcommand_statistics (int *keywords, pspp_linreg_cache * c)
403 The order here must match the order in which the STATISTICS
404 keywords appear in the specification section above.
431 Set everything but F.
433 for (i = 0; i < f; i++)
440 for (i = 0; i < all; i++)
448 Default output: ANOVA table, parameter estimates,
449 and statistics for variables not entered into model,
452 if (keywords[defaults] | d)
460 statistics_keyword_output (reg_stats_r, keywords[r], c);
461 statistics_keyword_output (reg_stats_anova, keywords[anova], c);
462 statistics_keyword_output (reg_stats_coeff, keywords[coeff], c);
463 statistics_keyword_output (reg_stats_outs, keywords[outs], c);
464 statistics_keyword_output (reg_stats_zpp, keywords[zpp], c);
465 statistics_keyword_output (reg_stats_label, keywords[label], c);
466 statistics_keyword_output (reg_stats_sha, keywords[sha], c);
467 statistics_keyword_output (reg_stats_ci, keywords[ci], c);
468 statistics_keyword_output (reg_stats_f, keywords[f], c);
469 statistics_keyword_output (reg_stats_bcov, keywords[bcov], c);
470 statistics_keyword_output (reg_stats_ses, keywords[ses], c);
471 statistics_keyword_output (reg_stats_xtx, keywords[xtx], c);
472 statistics_keyword_output (reg_stats_collin, keywords[collin], c);
473 statistics_keyword_output (reg_stats_tol, keywords[tol], c);
474 statistics_keyword_output (reg_stats_selection, keywords[selection], c);
477 int reg_inserted (struct variable *v, struct variable **varlist, int n_vars)
481 for (i = 0; i < n_vars; i++)
483 if (v->index == varlist[i]->index)
491 reg_print_categorical_encoding (FILE *fp, pspp_linreg_cache *c)
496 struct variable **varlist;
497 struct pspp_linreg_coeff coeff;
500 fprintf (fp, "%s", reg_export_categorical_encode_1);
502 varlist = xnmalloc (c->n_indeps, sizeof (*varlist));
503 for (i = 1; i < c->n_indeps; i++) /* c->coeff[0] is the intercept. */
506 if (coeff.v->type == ALPHA)
508 if (!reg_inserted (coeff.v, varlist, n_vars))
510 fprintf (fp, "struct pspp_reg_categorical_variable %s;\n\t", coeff.v->name);
511 varlist[n_vars] = coeff.v;
516 fprintf (fp, "int n_vars = %d;\n\t", n_vars);
517 fprintf (fp, "struct pspp_reg_categorical_variable *varlist[%d] = {", n_vars);
518 for (i = 0; i < n_vars - 1; i++)
520 fprintf (fp, "&%s,\n\t\t", varlist[i]->name);
522 fprintf (fp, "&%s};\n\t", varlist[i]->name);
524 for (i = 0; i < n_vars; i++)
527 fprintf (fp, "%s.name = \"%s\";\n\t", varlist[i]->name, varlist[i]->name);
528 fprintf (fp, "%s.n_vals = %d;\n\t", varlist[i]->name, varlist[i]->obs_vals->n_categories);
530 for (j = 0; j < varlist[i]->obs_vals->n_categories; j++)
532 val = cat_subscript_to_value ( (const size_t) j, varlist[i]);
533 fprintf (fp, "%s.values[%d] = \"%s\";\n\t", varlist[i]->name, j, value_to_string (val, varlist[i]));
536 fprintf (fp, "%s", reg_export_categorical_encode_2);
540 reg_print_depvars (FILE *fp, pspp_linreg_cache *c)
543 struct pspp_linreg_coeff coeff;
545 fprintf (fp, "char *model_depvars[%d] = {", c->n_indeps);
546 for (i = 1; i < c->n_indeps; i++)
549 fprintf (fp, "\"%s\",\n\t\t", coeff.v->name);
552 fprintf (fp, "\"%s\"};\n\t", coeff.v->name);
555 reg_print_getvar (FILE *fp, pspp_linreg_cache *c)
557 fprintf (fp, "static int\npspp_reg_getvar (char *v_name)\n{\n\t");
558 fprintf (fp, "int i;\n\tint n_vars = %d;\n\t",c->n_indeps);
559 reg_print_depvars (fp, c);
560 fprintf (fp, "for (i = 0; i < n_vars; i++)\n\t{\n\t\t");
561 fprintf (fp, "if (strcmp (v_name, model_depvars[i]) == 0)\n\t\t{\n\t\t\t");
562 fprintf (fp, "return i;\n\t\t}\n\t}\n}\n");
565 subcommand_export (int export, pspp_linreg_cache *c)
570 int n_quantiles = 100;
573 struct pspp_linreg_coeff coeff;
578 assert (model_file != NULL);
580 fp = fopen (handle_get_filename (model_file), "w");
581 fprintf (fp, "%s", reg_preamble);
582 fprintf (fp, "#include <string.h>\n#include <math.h>\n\n");
583 reg_print_getvar (fp, c);
584 reg_print_categorical_encoding (fp, c);
585 fprintf (fp, "%s", reg_export_t_quantiles_1);
586 increment = 0.5 / (double) increment;
587 for (i = 0; i < n_quantiles - 1; i++)
589 tmp = 0.5 + 0.005 * (double) i;
590 fprintf (fp, "%.15e,\n\t\t", gsl_cdf_tdist_Pinv (tmp, c->n_obs - c->n_indeps));
592 fprintf (fp, "%.15e};\n\t", gsl_cdf_tdist_Pinv (.9995, c->n_obs - c->n_indeps));
593 fprintf (fp, "%s", reg_export_t_quantiles_2);
594 fprintf (fp, "%s", reg_mean_cmt);
595 fprintf (fp, "double\npspp_reg_estimate (const double *var_vals,");
596 fprintf (fp, "const char *var_names[])\n{\n\t");
597 fprintf (fp, "double model_coeffs[%d] = {", c->n_indeps);
598 for (i = 1; i < c->n_indeps; i++)
601 fprintf (fp, "%.15e,\n\t\t", coeff.estimate);
604 fprintf (fp, "%.15e};\n\t", coeff.estimate);
606 fprintf (fp, "double estimate = %.15e;\n\t", coeff.estimate);
607 fprintf (fp, "int i;\n\tint j;\n\n\t");
608 fprintf (fp, "for (i = 0; i < %d; i++)\n\t", c->n_indeps);
609 fprintf (fp, "%s", reg_getvar);
610 fprintf (fp, "const double cov[%d][%d] = {\n\t", c->n_coeffs, c->n_coeffs);
611 for (i = 0; i < c->cov->size1 - 1; i++)
614 for (j = 0; j < c->cov->size2 - 1; j++)
616 fprintf (fp, "%.15e, ", gsl_matrix_get (c->cov, i, j));
618 fprintf (fp, "%.15e},\n\t", gsl_matrix_get (c->cov, i, j));
621 for (j = 0; j < c->cov->size2 - 1; j++)
623 fprintf (fp, "%.15e, ", gsl_matrix_get (c->cov, c->cov->size1 - 1, j));
625 fprintf (fp, "%.15e}\n\t", gsl_matrix_get (c->cov, c->cov->size1 - 1, c->cov->size2 - 1));
626 fprintf (fp, "};\n\tint n_vars = %d;\n\tint i;\n\tint j;\n\t", c->n_indeps);
627 fprintf (fp, "double unshuffled_vals[%d];\n\t",c->n_indeps);
628 fprintf (fp, "%s", reg_variance);
629 fprintf (fp, "%s", reg_export_confidence_interval);
630 tmp = c->mse * c->mse;
631 fprintf (fp, "%s %.15e", reg_export_prediction_interval_1, tmp);
632 fprintf (fp, "%s %.15e", reg_export_prediction_interval_2, tmp);
633 fprintf (fp, "%s", reg_export_prediction_interval_3);
635 fp = fopen ("pspp_model_reg.h", "w");
636 fprintf (fp, "%s", reg_header);
641 regression_custom_export (struct cmd_regression *cmd)
643 /* 0 on failure, 1 on success, 2 on failure that should result in syntax error */
644 if (!lex_force_match ('('))
651 model_file = fh_parse ();
652 if (model_file == NULL)
656 if (!lex_force_match (')'))
663 cmd_regression (void)
665 if (!parse_regression (&cmd))
669 multipass_procedure_with_splits (run_regression, &cmd);
675 Is variable k one of the dependent variables?
681 for (j = 0; j < cmd.n_dependent; j++)
684 compare_var_names returns 0 if the variable
687 if (!compare_var_names (cmd.v_dependent[j], cmd.v_variables[k], NULL))
694 run_regression (const struct casefile *cf, void *cmd_ UNUSED)
703 Keep track of the missing cases.
705 int *is_missing_case;
706 const union value *val;
707 struct casereader *r;
708 struct casereader *r2;
711 struct variable **indep_vars;
712 struct design_matrix *X;
714 pspp_linreg_cache *lcache;
715 pspp_linreg_opts lopts;
717 n_data = casefile_get_case_cnt (cf);
719 is_missing_case = xnmalloc (n_data, sizeof (*is_missing_case));
720 for (i = 0; i < n_data; i++)
721 is_missing_case[i] = 0;
723 n_indep = cmd.n_variables - cmd.n_dependent;
724 indep_vars = xnmalloc (n_indep, sizeof *indep_vars);
726 lopts.get_depvar_mean_std = 1;
727 lopts.get_indep_mean_std = xnmalloc (n_indep, sizeof (int));
731 Read from the active file. The first pass encodes categorical
732 variables and drops cases with missing values.
735 for (i = 0; i < cmd.n_variables; i++)
739 v = cmd.v_variables[i];
742 if (v->type == ALPHA)
744 /* Make a place to hold the binary vectors
745 corresponding to this variable's values. */
746 cat_stored_values_create (v);
748 for (r = casefile_get_reader (cf);
749 casereader_read (r, &c); case_destroy (&c))
751 row = casereader_cnum (r) - 1;
753 val = case_data (&c, v->fv);
754 cat_value_update (v, val);
755 if (mv_is_value_missing (&v->miss, val))
757 if (!is_missing_case[row])
759 /* Now it is missing. */
761 is_missing_case[row] = 1;
768 Y = gsl_vector_alloc (n_data);
770 design_matrix_create (n_indep, (const struct variable **) indep_vars,
772 lcache = pspp_linreg_cache_alloc (X->m->size1, X->m->size2);
773 lcache->indep_means = gsl_vector_alloc (X->m->size2);
774 lcache->indep_std = gsl_vector_alloc (X->m->size2);
777 The second pass creates the design matrix.
780 for (r2 = casefile_get_reader (cf); casereader_read (r2, &c);
782 /* Iterate over the cases. */
784 case_num = casereader_cnum (r2) - 1;
785 if (!is_missing_case[case_num])
787 for (i = 0; i < cmd.n_variables; ++i) /* Iterate over the variables
788 for the current case.
791 v = cmd.v_variables[i];
792 val = case_data (&c, v->fv);
794 Independent/dependent variable separation. The
795 'variables' subcommand specifies a varlist which contains
796 both dependent and independent variables. The dependent
797 variables are specified with the 'dependent'
798 subcommand. We need to separate the two.
802 if (v->type != NUMERIC)
805 gettext ("Dependent variable must be numeric."));
806 pspp_reg_rc = CMD_FAILURE;
809 lcache->depvar = (const struct variable *) v;
810 gsl_vector_set (Y, row, val->f);
814 if (v->type == ALPHA)
816 design_matrix_set_categorical (X, row, v, val);
818 else if (v->type == NUMERIC)
820 design_matrix_set_numeric (X, row, v, val);
823 lopts.get_indep_mean_std[i] = 1;
830 Now that we know the number of coefficients, allocate space
831 and store pointers to the variables that correspond to the
834 lcache->coeff = xnmalloc (X->m->size2 + 1, sizeof (*lcache->coeff));
835 for (i = 0; i < X->m->size2; i++)
837 j = i + 1; /* The first coeff is the intercept. */
839 (const struct variable *) design_matrix_col_to_var (X, i);
840 assert (lcache->coeff[j].v != NULL);
843 For large data sets, use QR decomposition.
845 if (n_data > sqrt (n_indep) && n_data > REG_LARGE_DATA)
847 lcache->method = PSPP_LINREG_SVD;
850 Find the least-squares estimates and other statistics.
852 pspp_linreg ((const gsl_vector *) Y, X->m, &lopts, lcache);
853 subcommand_statistics (cmd.a_statistics, lcache);
854 subcommand_export (cmd.sbc_export, lcache);
856 design_matrix_destroy (X);
857 pspp_linreg_cache_free (lcache);
858 free (lopts.get_indep_mean_std);
860 free (is_missing_case);
861 casereader_destroy (r);