1 /* pspp - a program for statistical analysis.
2 Copyright (C) 2012 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <http://www.gnu.org/licenses/>. */
20 1. "Coding Logistic Regression with Newton-Raphson", James McCaffrey
21 http://msdn.microsoft.com/en-us/magazine/jj618304.aspx
23 2. "SPSS Statistical Algorithms" Chapter LOGISTIC REGRESSION Algorithms
26 The Newton Raphson method finds successive approximations to $\bf b$ where
27 approximation ${\bf b}_t$ is (hopefully) better than the previous ${\bf b}_{t-1}$.
29 $ {\bf b}_t = {\bf b}_{t -1} + ({\bf X}^T{\bf W}_{t-1}{\bf X})^{-1}{\bf X}^T({\bf y} - {\bf \pi}_{t-1})$
32 $\bf X$ is the $n \times p$ design matrix, $n$ being the number of cases,
33 $p$ the number of parameters, \par
34 $\bf W$ is the diagonal matrix whose diagonal elements are
35 $\hat{\pi}_0(1 - \hat{\pi}_0), \, \hat{\pi}_1(1 - \hat{\pi}_2)\dots \hat{\pi}_{n-1}(1 - \hat{\pi}_{n-1})$
42 #include <gsl/gsl_blas.h>
44 #include <gsl/gsl_linalg.h>
45 #include <gsl/gsl_cdf.h>
46 #include <gsl/gsl_matrix.h>
47 #include <gsl/gsl_vector.h>
50 #include "data/case.h"
51 #include "data/casegrouper.h"
52 #include "data/casereader.h"
53 #include "data/dataset.h"
54 #include "data/dictionary.h"
55 #include "data/format.h"
56 #include "data/value.h"
57 #include "language/command.h"
58 #include "language/dictionary/split-file.h"
59 #include "language/lexer/lexer.h"
60 #include "language/lexer/value-parser.h"
61 #include "language/lexer/variable-parser.h"
62 #include "libpspp/assertion.h"
63 #include "libpspp/ll.h"
64 #include "libpspp/message.h"
65 #include "libpspp/misc.h"
66 #include "math/categoricals.h"
67 #include "math/interaction.h"
68 #include "libpspp/hmap.h"
69 #include "libpspp/hash-functions.h"
71 #include "output/tab.h"
74 #define _(msgid) gettext (msgid)
79 #define PRINT_EACH_STEP 0x01
80 #define PRINT_SUMMARY 0x02
81 #define PRINT_CORR 0x04
82 #define PRINT_ITER 0x08
83 #define PRINT_GOODFIT 0x10
87 #define PRINT_DEFAULT (PRINT_SUMMARY | PRINT_EACH_STEP)
90 The constant parameters of the procedure.
91 That is, those which are set by the user.
95 /* The dependent variable */
96 const struct variable *dep_var;
98 /* The predictor variables (excluding categorical ones) */
99 const struct variable **predictor_vars;
100 size_t n_predictor_vars;
102 /* The categorical predictors */
103 struct interaction **cat_predictors;
104 size_t n_cat_predictors;
107 /* The union of the categorical and non-categorical variables */
108 const struct variable **indep_vars;
112 /* Which classes of missing vars are to be excluded */
113 enum mv_class exclude;
115 /* The weight variable */
116 const struct variable *wv;
118 /* The dictionary of the dataset */
119 const struct dictionary *dict;
121 /* True iff the constant (intercept) is to be included in the model */
124 /* Ths maximum number of iterations */
127 /* Other iteration limiting conditions */
132 /* The confidence interval (in percent) */
135 /* What results should be presented */
142 /* The results and intermediate result of the procedure.
143 These are mutated as the procedure runs. Used for
144 temporary variables etc.
148 /* Used to indicate if a pass should flag a warning when
149 invalid (ie negative or missing) weight values are encountered */
150 bool warn_bad_weight;
152 /* The two values of the dependent variable. */
157 /* The sum of caseweights */
160 /* The number of missing and nonmissing cases */
161 casenumber n_missing;
162 casenumber n_nonmissing;
167 /* The categoricals and their payload. Null if the analysis has no
168 categorical predictors */
169 struct categoricals *cats;
173 /* The estimates of the predictor coefficients */
174 gsl_vector *beta_hat;
179 Convert INPUT into a dichotomous scalar, according to how the dependent variable's
181 For simple cases, this is a 1:1 mapping
182 The return value is always either 0 or 1
185 map_dependent_var (const struct lr_spec *cmd, const struct lr_result *res, const union value *input)
187 const int width = var_get_width (cmd->dep_var);
188 if (value_equal (input, &res->y0, width))
191 if (value_equal (input, &res->y1, width))
194 /* This should never happen. If it does, then y0 and/or y1 have probably not been set */
201 static void output_categories (const struct lr_spec *cmd, const struct lr_result *res);
203 static void output_depvarmap (const struct lr_spec *cmd, const struct lr_result *);
205 static void output_variables (const struct lr_spec *cmd,
206 const struct lr_result *);
208 static void output_model_summary (const struct lr_result *,
209 double initial_likelihood, double likelihood);
211 static void case_processing_summary (const struct lr_result *);
214 /* Return the value of case C corresponding to the INDEX'th entry in the
217 predictor_value (const struct ccase *c,
218 const struct variable **x, size_t n_x,
219 const struct categoricals *cats,
222 /* Values of the scalar predictor variables */
224 return case_data (c, x[index])->f;
226 /* Coded values of categorical predictor variables (or interactions) */
227 if (cats && index - n_x < categoricals_df_total (cats))
229 double x = categoricals_get_dummy_code_for_case (cats, index - n_x, c);
233 /* The constant term */
239 Return the probability beta_hat (that is the estimator logit(y) )
240 corresponding to the coefficient estimator for case C
243 pi_hat (const struct lr_spec *cmd,
244 const struct lr_result *res,
245 const struct variable **x, size_t n_x,
246 const struct ccase *c)
250 size_t n_coeffs = res->beta_hat->size;
254 pi += gsl_vector_get (res->beta_hat, res->beta_hat->size - 1);
258 for (v0 = 0; v0 < n_coeffs; ++v0)
260 pi += gsl_vector_get (res->beta_hat, v0) *
261 predictor_value (c, x, n_x, res->cats, v0);
264 pi = 1.0 / (1.0 + exp(-pi));
271 Calculates the Hessian matrix X' V X,
272 where: X is the n by N_X matrix comprising the n cases in INPUT
273 V is a diagonal matrix { (pi_hat_0)(1 - pi_hat_0), (pi_hat_1)(1 - pi_hat_1), ... (pi_hat_{N-1})(1 - pi_hat_{N-1})}
274 (the partial derivative of the predicted values)
276 If ALL predicted values derivatives are close to zero or one, then CONVERGED
280 hessian (const struct lr_spec *cmd,
281 struct lr_result *res,
282 struct casereader *input,
283 const struct variable **x, size_t n_x,
286 struct casereader *reader;
289 double max_w = -DBL_MAX;
291 gsl_matrix_set_zero (res->hessian);
293 for (reader = casereader_clone (input);
294 (c = casereader_read (reader)) != NULL; case_unref (c))
297 double pi = pi_hat (cmd, res, x, n_x, c);
299 double weight = dict_get_case_weight (cmd->dict, c, &res->warn_bad_weight);
300 double w = pi * (1 - pi);
305 for (v0 = 0; v0 < res->beta_hat->size; ++v0)
307 double in0 = predictor_value (c, x, n_x, res->cats, v0);
308 for (v1 = 0; v1 < res->beta_hat->size; ++v1)
310 double in1 = predictor_value (c, x, n_x, res->cats, v1);
311 double *o = gsl_matrix_ptr (res->hessian, v0, v1);
316 casereader_destroy (reader);
318 if ( max_w < cmd->min_epsilon)
321 msg (MN, _("All predicted values are either 1 or 0"));
326 /* Calculates the value X' (y - pi)
327 where X is the design model,
328 y is the vector of observed independent variables
329 pi is the vector of estimates for y
331 As a side effect, the likelihood is stored in LIKELIHOOD
334 xt_times_y_pi (const struct lr_spec *cmd,
335 struct lr_result *res,
336 struct casereader *input,
337 const struct variable **x, size_t n_x,
338 const struct variable *y_var,
341 struct casereader *reader;
343 gsl_vector *output = gsl_vector_calloc (res->beta_hat->size);
346 for (reader = casereader_clone (input);
347 (c = casereader_read (reader)) != NULL; case_unref (c))
350 double pi = pi_hat (cmd, res, x, n_x, c);
351 double weight = dict_get_case_weight (cmd->dict, c, &res->warn_bad_weight);
354 double y = map_dependent_var (cmd, res, case_data (c, y_var));
356 *likelihood *= pow (pi, weight * y) * pow (1 - pi, weight * (1 - y));
358 for (v0 = 0; v0 < res->beta_hat->size; ++v0)
360 double in0 = predictor_value (c, x, n_x, res->cats, v0);
361 double *o = gsl_vector_ptr (output, v0);
362 *o += in0 * (y - pi) * weight;
366 casereader_destroy (reader);
373 /* "payload" functions for the categoricals.
374 The only function is to accumulate the frequency of each
379 frq_create (const void *aux1 UNUSED, void *aux2 UNUSED)
381 return xzalloc (sizeof (double));
385 frq_update (const void *aux1 UNUSED, void *aux2 UNUSED,
386 void *ud, const struct ccase *c UNUSED , double weight)
393 frq_destroy (const void *aux1 UNUSED, void *aux2 UNUSED, void *user_data UNUSED)
401 Makes an initial pass though the data, doing the following:
403 * Checks that the dependent variable is dichotomous,
404 * Creates and initialises the categoricals,
405 * Accumulates summary results,
406 * Calculates necessary initial values.
407 * Creates an initial value for \hat\beta the vector of beta_hats of \beta
409 Returns true if successful
412 initial_pass (const struct lr_spec *cmd, struct lr_result *res, struct casereader *input)
414 const int width = var_get_width (cmd->dep_var);
417 struct casereader *reader;
426 size_t n_coefficients = cmd->n_predictor_vars;
430 /* Create categoricals if appropriate */
431 if (cmd->n_cat_predictors > 0)
433 res->cp.create = frq_create;
434 res->cp.update = frq_update;
435 res->cp.calculate = NULL;
436 res->cp.destroy = frq_destroy;
438 res->cats = categoricals_create (cmd->cat_predictors, cmd->n_cat_predictors,
439 cmd->wv, cmd->exclude, MV_ANY);
441 categoricals_set_payload (res->cats, &res->cp, cmd, res);
445 for (reader = casereader_clone (input);
446 (c = casereader_read (reader)) != NULL; case_unref (c))
449 bool missing = false;
450 double weight = dict_get_case_weight (cmd->dict, c, &res->warn_bad_weight);
451 const union value *depval = case_data (c, cmd->dep_var);
453 for (v = 0; v < cmd->n_indep_vars; ++v)
455 const union value *val = case_data (c, cmd->indep_vars[v]);
456 if (var_is_value_missing (cmd->indep_vars[v], val, cmd->exclude))
463 /* Accumulate the missing and non-missing counts */
471 /* Find the values of the dependent variable */
474 value_clone (&res->y0, depval, width);
479 if ( !value_equal (&res->y0, depval, width))
481 value_clone (&res->y1, depval, width);
487 if (! value_equal (&res->y0, depval, width)
489 ! value_equal (&res->y1, depval, width)
492 msg (ME, _("Dependent variable's values are not dichotomous."));
497 if (v0set && value_equal (&res->y0, depval, width))
500 if (v1set && value_equal (&res->y1, depval, width))
506 categoricals_update (res->cats, c);
508 casereader_destroy (reader);
510 categoricals_done (res->cats);
514 /* Ensure that Y0 is less than Y1. Otherwise the mapping gets
515 inverted, which is confusing to users */
516 if (var_is_numeric (cmd->dep_var) && value_compare_3way (&res->y0, &res->y1, width) > 0)
519 value_clone (&tmp, &res->y0, width);
520 value_copy (&res->y0, &res->y1, width);
521 value_copy (&res->y1, &tmp, width);
522 value_destroy (&tmp, width);
526 n_coefficients += categoricals_df_total (res->cats);
527 res->beta_hat = gsl_vector_calloc (n_coefficients);
531 double mean = sum / res->cc;
532 gsl_vector_set (res->beta_hat, res->beta_hat->size - 1, log (mean / (1 - mean)));
538 casereader_destroy (reader);
544 /* Start of the logistic regression routine proper */
546 run_lr (const struct lr_spec *cmd, struct casereader *input,
547 const struct dataset *ds UNUSED)
551 bool converged = false;
553 /* Set the likelihoods to a negative sentinel value */
554 double likelihood = -1;
555 double prev_likelihood = -1;
556 double initial_likelihood = -1;
558 struct lr_result work;
560 work.n_nonmissing = 0;
561 work.warn_bad_weight = true;
563 work.beta_hat = NULL;
565 /* Get the initial estimates of \beta and their standard errors.
566 And perform other auxilliary initialisation. */
567 if (! initial_pass (cmd, &work, input))
570 for (i = 0; i < cmd->n_cat_predictors; ++i)
572 if (1 >= categoricals_n_count (work.cats, i))
575 ds_init_empty (&str);
577 interaction_to_string (cmd->cat_predictors[i], &str);
579 msg (ME, _("Category %s does not have at least two distinct values. Logistic regression will not be run."),
586 output_depvarmap (cmd, &work);
588 case_processing_summary (&work);
591 input = casereader_create_filter_missing (input,
599 work.hessian = gsl_matrix_calloc (work.beta_hat->size, work.beta_hat->size);
601 /* Start the Newton Raphson iteration process... */
602 for( i = 0 ; i < cmd->max_iter ; ++i)
608 hessian (cmd, &work, input,
609 cmd->predictor_vars, cmd->n_predictor_vars,
612 gsl_linalg_cholesky_decomp (work.hessian);
613 gsl_linalg_cholesky_invert (work.hessian);
615 v = xt_times_y_pi (cmd, &work, input,
616 cmd->predictor_vars, cmd->n_predictor_vars,
622 gsl_vector *delta = gsl_vector_alloc (v->size);
623 gsl_blas_dgemv (CblasNoTrans, 1.0, work.hessian, v, 0, delta);
627 gsl_vector_add (work.beta_hat, delta);
629 gsl_vector_minmax (delta, &min, &max);
631 if ( fabs (min) < cmd->bcon && fabs (max) < cmd->bcon)
633 msg (MN, _("Estimation terminated at iteration number %d because parameter estimates changed by less than %g"),
638 gsl_vector_free (delta);
641 if ( prev_likelihood >= 0)
643 if (-log (likelihood) > -(1.0 - cmd->lcon) * log (prev_likelihood))
645 msg (MN, _("Estimation terminated at iteration number %d because Log Likelihood decreased by less than %g%%"), i + 1, 100 * cmd->lcon);
650 initial_likelihood = likelihood;
651 prev_likelihood = likelihood;
656 casereader_destroy (input);
657 assert (initial_likelihood >= 0);
660 msg (MW, _("Estimation terminated at iteration number %d because maximum iterations has been reached"), i );
663 output_model_summary (&work, initial_likelihood, likelihood);
666 output_categories (cmd, &work);
668 output_variables (cmd, &work);
670 gsl_matrix_free (work.hessian);
671 gsl_vector_free (work.beta_hat);
673 categoricals_destroy (work.cats);
680 struct hmap_node node; /* Node in hash map. */
681 const struct variable *var; /* The variable */
684 static struct variable_node *
685 lookup_variable (const struct hmap *map, const struct variable *var, unsigned int hash)
687 struct variable_node *vn = NULL;
688 HMAP_FOR_EACH_WITH_HASH (vn, struct variable_node, node, hash, map)
693 fprintf (stderr, "Warning: Hash table collision\n");
700 /* Parse the LOGISTIC REGRESSION command syntax */
702 cmd_logistic (struct lexer *lexer, struct dataset *ds)
704 /* Temporary location for the predictor variables.
705 These may or may not include the categorical predictors */
706 const struct variable **pred_vars;
711 lr.dict = dataset_dict (ds);
712 lr.n_predictor_vars = 0;
713 lr.predictor_vars = NULL;
715 lr.wv = dict_get_weight (lr.dict);
719 lr.min_epsilon = 0.00000001;
723 lr.print = PRINT_DEFAULT;
724 lr.cat_predictors = NULL;
725 lr.n_cat_predictors = 0;
726 lr.indep_vars = NULL;
729 if (lex_match_id (lexer, "VARIABLES"))
730 lex_match (lexer, T_EQUALS);
732 if (! (lr.dep_var = parse_variable_const (lexer, lr.dict)))
735 lex_force_match (lexer, T_WITH);
737 if (!parse_variables_const (lexer, lr.dict,
738 &pred_vars, &n_pred_vars,
743 while (lex_token (lexer) != T_ENDCMD)
745 lex_match (lexer, T_SLASH);
747 if (lex_match_id (lexer, "MISSING"))
749 lex_match (lexer, T_EQUALS);
750 while (lex_token (lexer) != T_ENDCMD
751 && lex_token (lexer) != T_SLASH)
753 if (lex_match_id (lexer, "INCLUDE"))
755 lr.exclude = MV_SYSTEM;
757 else if (lex_match_id (lexer, "EXCLUDE"))
763 lex_error (lexer, NULL);
768 else if (lex_match_id (lexer, "ORIGIN"))
772 else if (lex_match_id (lexer, "NOORIGIN"))
776 else if (lex_match_id (lexer, "NOCONST"))
780 else if (lex_match_id (lexer, "EXTERNAL"))
782 /* This is for compatibility. It does nothing */
784 else if (lex_match_id (lexer, "CATEGORICAL"))
786 lex_match (lexer, T_EQUALS);
789 lr.cat_predictors = xrealloc (lr.cat_predictors,
790 sizeof (*lr.cat_predictors) * ++lr.n_cat_predictors);
791 lr.cat_predictors[lr.n_cat_predictors - 1] = 0;
793 while (parse_design_interaction (lexer, lr.dict,
794 lr.cat_predictors + lr.n_cat_predictors - 1));
795 lr.n_cat_predictors--;
797 else if (lex_match_id (lexer, "PRINT"))
799 lex_match (lexer, T_EQUALS);
800 while (lex_token (lexer) != T_ENDCMD && lex_token (lexer) != T_SLASH)
802 if (lex_match_id (lexer, "DEFAULT"))
804 lr.print |= PRINT_DEFAULT;
806 else if (lex_match_id (lexer, "SUMMARY"))
808 lr.print |= PRINT_SUMMARY;
811 else if (lex_match_id (lexer, "CORR"))
813 lr.print |= PRINT_CORR;
815 else if (lex_match_id (lexer, "ITER"))
817 lr.print |= PRINT_ITER;
819 else if (lex_match_id (lexer, "GOODFIT"))
821 lr.print |= PRINT_GOODFIT;
824 else if (lex_match_id (lexer, "CI"))
826 lr.print |= PRINT_CI;
827 if (lex_force_match (lexer, T_LPAREN))
829 if (! lex_force_int (lexer))
831 lex_error (lexer, NULL);
834 lr.confidence = lex_integer (lexer);
836 if ( ! lex_force_match (lexer, T_RPAREN))
838 lex_error (lexer, NULL);
843 else if (lex_match_id (lexer, "ALL"))
849 lex_error (lexer, NULL);
854 else if (lex_match_id (lexer, "CRITERIA"))
856 lex_match (lexer, T_EQUALS);
857 while (lex_token (lexer) != T_ENDCMD && lex_token (lexer) != T_SLASH)
859 if (lex_match_id (lexer, "BCON"))
861 if (lex_force_match (lexer, T_LPAREN))
863 if (! lex_force_num (lexer))
865 lex_error (lexer, NULL);
868 lr.bcon = lex_number (lexer);
870 if ( ! lex_force_match (lexer, T_RPAREN))
872 lex_error (lexer, NULL);
877 else if (lex_match_id (lexer, "ITERATE"))
879 if (lex_force_match (lexer, T_LPAREN))
881 if (! lex_force_int (lexer))
883 lex_error (lexer, NULL);
886 lr.max_iter = lex_integer (lexer);
888 if ( ! lex_force_match (lexer, T_RPAREN))
890 lex_error (lexer, NULL);
895 else if (lex_match_id (lexer, "LCON"))
897 if (lex_force_match (lexer, T_LPAREN))
899 if (! lex_force_num (lexer))
901 lex_error (lexer, NULL);
904 lr.lcon = lex_number (lexer);
906 if ( ! lex_force_match (lexer, T_RPAREN))
908 lex_error (lexer, NULL);
913 else if (lex_match_id (lexer, "EPS"))
915 if (lex_force_match (lexer, T_LPAREN))
917 if (! lex_force_num (lexer))
919 lex_error (lexer, NULL);
922 lr.min_epsilon = lex_number (lexer);
924 if ( ! lex_force_match (lexer, T_RPAREN))
926 lex_error (lexer, NULL);
933 lex_error (lexer, NULL);
940 lex_error (lexer, NULL);
945 /* Copy the predictor variables from the temporary location into the
946 final one, dropping any categorical variables which appear there.
947 FIXME: This is O(NxM).
951 struct variable_node *vn, *next;
953 hmap_init (&allvars);
954 for (v = x = 0; v < n_pred_vars; ++v)
957 const struct variable *var = pred_vars[v];
960 unsigned int hash = hash_pointer (var, 0);
961 struct variable_node *vn = lookup_variable (&allvars, var, hash);
964 vn = xmalloc (sizeof *vn);
966 hmap_insert (&allvars, &vn->node, hash);
969 for (cv = 0; cv < lr.n_cat_predictors ; ++cv)
972 const struct interaction *iact = lr.cat_predictors[cv];
973 for (iv = 0 ; iv < iact->n_vars ; ++iv)
975 const struct variable *ivar = iact->vars[iv];
976 unsigned int hash = hash_pointer (ivar, 0);
977 struct variable_node *vn = lookup_variable (&allvars, ivar, hash);
980 vn = xmalloc (sizeof *vn);
983 hmap_insert (&allvars, &vn->node, hash);
996 lr.predictor_vars = xrealloc (lr.predictor_vars, sizeof *lr.predictor_vars * (x + 1));
997 lr.predictor_vars[x++] = var;
998 lr.n_predictor_vars++;
1002 lr.n_indep_vars = hmap_count (&allvars);
1003 lr.indep_vars = xmalloc (lr.n_indep_vars * sizeof *lr.indep_vars);
1005 /* Interate over each variable and push it into the array */
1007 HMAP_FOR_EACH_SAFE (vn, next, struct variable_node, node, &allvars)
1009 lr.indep_vars[x++] = vn->var;
1012 hmap_destroy (&allvars);
1016 /* logistical regression for each split group */
1018 struct casegrouper *grouper;
1019 struct casereader *group;
1022 grouper = casegrouper_create_splits (proc_open (ds), lr.dict);
1023 while (casegrouper_get_next_group (grouper, &group))
1024 ok = run_lr (&lr, group, ds);
1025 ok = casegrouper_destroy (grouper);
1026 ok = proc_commit (ds) && ok;
1029 free (lr.predictor_vars);
1030 free (lr.cat_predictors);
1031 free (lr.indep_vars);
1037 free (lr.predictor_vars);
1038 free (lr.cat_predictors);
1039 free (lr.indep_vars);
1047 /* Show the Dependent Variable Encoding box.
1048 This indicates how the dependent variable
1049 is mapped to the internal zero/one values.
1052 output_depvarmap (const struct lr_spec *cmd, const struct lr_result *res)
1054 const int heading_columns = 0;
1055 const int heading_rows = 1;
1056 struct tab_table *t;
1060 int nr = heading_rows + 2;
1062 t = tab_create (nc, nr);
1063 tab_title (t, _("Dependent Variable Encoding"));
1065 tab_headers (t, heading_columns, 0, heading_rows, 0);
1067 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, nc - 1, nr - 1);
1069 tab_hline (t, TAL_2, 0, nc - 1, heading_rows);
1070 tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
1072 tab_text (t, 0, 0, TAB_CENTER | TAT_TITLE, _("Original Value"));
1073 tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("Internal Value"));
1077 ds_init_empty (&str);
1078 var_append_value_name (cmd->dep_var, &res->y0, &str);
1079 tab_text (t, 0, 0 + heading_rows, 0, ds_cstr (&str));
1082 var_append_value_name (cmd->dep_var, &res->y1, &str);
1083 tab_text (t, 0, 1 + heading_rows, 0, ds_cstr (&str));
1086 tab_double (t, 1, 0 + heading_rows, 0, map_dependent_var (cmd, res, &res->y0), &F_8_0);
1087 tab_double (t, 1, 1 + heading_rows, 0, map_dependent_var (cmd, res, &res->y1), &F_8_0);
1094 /* Show the Variables in the Equation box */
1096 output_variables (const struct lr_spec *cmd,
1097 const struct lr_result *res)
1100 const int heading_columns = 1;
1101 int heading_rows = 1;
1102 struct tab_table *t;
1108 int idx_correction = 0;
1110 if (cmd->print & PRINT_CI)
1116 nr = heading_rows + cmd->n_predictor_vars;
1121 nr += categoricals_df_total (res->cats) + cmd->n_cat_predictors;
1123 t = tab_create (nc, nr);
1124 tab_title (t, _("Variables in the Equation"));
1126 tab_headers (t, heading_columns, 0, heading_rows, 0);
1128 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, nc - 1, nr - 1);
1130 tab_hline (t, TAL_2, 0, nc - 1, heading_rows);
1131 tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
1133 tab_text (t, 0, row + 1, TAB_CENTER | TAT_TITLE, _("Step 1"));
1135 tab_text (t, 2, row, TAB_CENTER | TAT_TITLE, _("B"));
1136 tab_text (t, 3, row, TAB_CENTER | TAT_TITLE, _("S.E."));
1137 tab_text (t, 4, row, TAB_CENTER | TAT_TITLE, _("Wald"));
1138 tab_text (t, 5, row, TAB_CENTER | TAT_TITLE, _("df"));
1139 tab_text (t, 6, row, TAB_CENTER | TAT_TITLE, _("Sig."));
1140 tab_text (t, 7, row, TAB_CENTER | TAT_TITLE, _("Exp(B)"));
1142 if (cmd->print & PRINT_CI)
1144 tab_joint_text_format (t, 8, 0, 9, 0,
1145 TAB_CENTER | TAT_TITLE, _("%d%% CI for Exp(B)"), cmd->confidence);
1147 tab_text (t, 8, row, TAB_CENTER | TAT_TITLE, _("Lower"));
1148 tab_text (t, 9, row, TAB_CENTER | TAT_TITLE, _("Upper"));
1151 for (row = heading_rows ; row < nr; ++row)
1153 const int idx = row - heading_rows - idx_correction;
1155 const double b = gsl_vector_get (res->beta_hat, idx);
1156 const double sigma2 = gsl_matrix_get (res->hessian, idx, idx);
1157 const double wald = pow2 (b) / sigma2;
1158 const double df = 1;
1160 if (idx < cmd->n_predictor_vars)
1162 tab_text (t, 1, row, TAB_LEFT | TAT_TITLE,
1163 var_to_string (cmd->predictor_vars[idx]));
1165 else if (i < cmd->n_cat_predictors)
1168 bool summary = false;
1170 const struct interaction *cat_predictors = cmd->cat_predictors[i];
1171 const int df = categoricals_df (res->cats, i);
1173 ds_init_empty (&str);
1174 interaction_to_string (cat_predictors, &str);
1178 /* Calculate the Wald statistic,
1179 which is \beta' C^-1 \beta .
1180 where \beta is the vector of the coefficient estimates comprising this
1181 categorial variable. and C is the corresponding submatrix of the
1184 gsl_matrix_const_view mv =
1185 gsl_matrix_const_submatrix (res->hessian, idx, idx, df, df);
1186 gsl_matrix *subhessian = gsl_matrix_alloc (mv.matrix.size1, mv.matrix.size2);
1187 gsl_vector_const_view vv = gsl_vector_const_subvector (res->beta_hat, idx, df);
1188 gsl_vector *temp = gsl_vector_alloc (df);
1190 gsl_matrix_memcpy (subhessian, &mv.matrix);
1191 gsl_linalg_cholesky_decomp (subhessian);
1192 gsl_linalg_cholesky_invert (subhessian);
1194 gsl_blas_dgemv (CblasTrans, 1.0, subhessian, &vv.vector, 0, temp);
1195 gsl_blas_ddot (temp, &vv.vector, &wald);
1197 tab_double (t, 4, row, 0, wald, 0);
1198 tab_double (t, 5, row, 0, df, &F_8_0);
1199 tab_double (t, 6, row, 0, gsl_cdf_chisq_Q (wald, df), 0);
1203 gsl_matrix_free (subhessian);
1204 gsl_vector_free (temp);
1208 ds_put_format (&str, "(%d)", ivar);
1211 tab_text (t, 1, row, TAB_LEFT | TAT_TITLE, ds_cstr (&str));
1214 ++i; /* next interaction */
1225 tab_text (t, 1, row, TAB_LEFT | TAT_TITLE, _("Constant"));
1228 tab_double (t, 2, row, 0, b, 0);
1229 tab_double (t, 3, row, 0, sqrt (sigma2), 0);
1230 tab_double (t, 4, row, 0, wald, 0);
1231 tab_double (t, 5, row, 0, df, &F_8_0);
1232 tab_double (t, 6, row, 0, gsl_cdf_chisq_Q (wald, df), 0);
1233 tab_double (t, 7, row, 0, exp (b), 0);
1235 if (cmd->print & PRINT_CI)
1237 double wc = gsl_cdf_ugaussian_Pinv (0.5 + cmd->confidence / 200.0);
1238 wc *= sqrt (sigma2);
1240 if (idx < cmd->n_predictor_vars)
1242 tab_double (t, 8, row, 0, exp (b - wc), 0);
1243 tab_double (t, 9, row, 0, exp (b + wc), 0);
1252 /* Show the model summary box */
1254 output_model_summary (const struct lr_result *res,
1255 double initial_likelihood, double likelihood)
1257 const int heading_columns = 0;
1258 const int heading_rows = 1;
1259 struct tab_table *t;
1262 int nr = heading_rows + 1;
1265 t = tab_create (nc, nr);
1266 tab_title (t, _("Model Summary"));
1268 tab_headers (t, heading_columns, 0, heading_rows, 0);
1270 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, nc - 1, nr - 1);
1272 tab_hline (t, TAL_2, 0, nc - 1, heading_rows);
1273 tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
1275 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("Step 1"));
1276 tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("-2 Log likelihood"));
1277 tab_double (t, 1, 1, 0, -2 * log (likelihood), 0);
1280 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Cox & Snell R Square"));
1281 cox = 1.0 - pow (initial_likelihood /likelihood, 2 / res->cc);
1282 tab_double (t, 2, 1, 0, cox, 0);
1284 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Nagelkerke R Square"));
1285 tab_double (t, 3, 1, 0, cox / ( 1.0 - pow (initial_likelihood, 2 / res->cc)), 0);
1291 /* Show the case processing summary box */
1293 case_processing_summary (const struct lr_result *res)
1295 const int heading_columns = 1;
1296 const int heading_rows = 1;
1297 struct tab_table *t;
1300 const int nr = heading_rows + 3;
1303 t = tab_create (nc, nr);
1304 tab_title (t, _("Case Processing Summary"));
1306 tab_headers (t, heading_columns, 0, heading_rows, 0);
1308 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, nc - 1, nr - 1);
1310 tab_hline (t, TAL_2, 0, nc - 1, heading_rows);
1311 tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
1313 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("Unweighted Cases"));
1314 tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("N"));
1315 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Percent"));
1318 tab_text (t, 0, 1, TAB_LEFT | TAT_TITLE, _("Included in Analysis"));
1319 tab_text (t, 0, 2, TAB_LEFT | TAT_TITLE, _("Missing Cases"));
1320 tab_text (t, 0, 3, TAB_LEFT | TAT_TITLE, _("Total"));
1322 tab_double (t, 1, 1, 0, res->n_nonmissing, &F_8_0);
1323 tab_double (t, 1, 2, 0, res->n_missing, &F_8_0);
1325 total = res->n_nonmissing + res->n_missing;
1326 tab_double (t, 1, 3, 0, total , &F_8_0);
1328 tab_double (t, 2, 1, 0, 100 * res->n_nonmissing / (double) total, 0);
1329 tab_double (t, 2, 2, 0, 100 * res->n_missing / (double) total, 0);
1330 tab_double (t, 2, 3, 0, 100 * total / (double) total, 0);
1336 output_categories (const struct lr_spec *cmd, const struct lr_result *res)
1338 const struct fmt_spec *wfmt =
1339 cmd->wv ? var_get_print_format (cmd->wv) : &F_8_0;
1343 const int heading_columns = 2;
1344 const int heading_rows = 2;
1345 struct tab_table *t;
1355 for (i = 0; i < cmd->n_cat_predictors; ++i)
1357 size_t n = categoricals_n_count (res->cats, i);
1358 size_t df = categoricals_df (res->cats, i);
1364 nc = heading_columns + 1 + max_df;
1365 nr = heading_rows + total_cats;
1367 t = tab_create (nc, nr);
1368 tab_title (t, _("Categorical Variables' Codings"));
1370 tab_headers (t, heading_columns, 0, heading_rows, 0);
1372 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, nc - 1, nr - 1);
1374 tab_hline (t, TAL_2, 0, nc - 1, heading_rows);
1375 tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
1378 tab_text (t, heading_columns, 1, TAB_CENTER | TAT_TITLE, _("Frequency"));
1380 tab_joint_text_format (t, heading_columns + 1, 0, nc - 1, 0,
1381 TAB_CENTER | TAT_TITLE, _("Parameter coding"));
1384 for (i = 0; i < max_df; ++i)
1386 int c = heading_columns + 1 + i;
1387 tab_text_format (t, c, 1, TAB_CENTER | TAT_TITLE, _("(%d)"), i + 1);
1391 for (v = 0; v < cmd->n_cat_predictors; ++v)
1394 const struct interaction *cat_predictors = cmd->cat_predictors[v];
1395 int df = categoricals_df (res->cats, v);
1397 ds_init_empty (&str);
1399 interaction_to_string (cat_predictors, &str);
1401 tab_text (t, 0, heading_rows + r, TAB_LEFT | TAT_TITLE, ds_cstr (&str) );
1405 for (cat = 0; cat < categoricals_n_count (res->cats, v) ; ++cat)
1408 const struct ccase *c = categoricals_get_case_by_category_real (res->cats, v, cat);
1409 const double *freq = categoricals_get_user_data_by_category_real (res->cats, v, cat);
1412 ds_init_empty (&str);
1414 for (x = 0; x < cat_predictors->n_vars; ++x)
1416 const union value *val = case_data (c, cat_predictors->vars[x]);
1417 var_append_value_name (cat_predictors->vars[x], val, &str);
1419 if (x < cat_predictors->n_vars - 1)
1420 ds_put_cstr (&str, " ");
1423 tab_text (t, 1, heading_rows + r, 0, ds_cstr (&str));
1425 tab_double (t, 2, heading_rows + r, 0, *freq, wfmt);
1427 for (x = 0; x < df; ++x)
1429 tab_double (t, heading_columns + 1 + x, heading_rows + r, 0, (cat == x), &F_8_0);
1433 cumulative_df += df;