1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 2009 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 #include <data/procedure.h>
21 #include <language/lexer/variable-parser.h>
22 #include <language/lexer/value-parser.h>
23 #include <language/lexer/lexer.h>
25 #include <data/casegrouper.h>
26 #include <data/casereader.h>
27 #include <data/casewriter.h>
28 #include <data/dictionary.h>
29 #include <data/format.h>
30 #include <math/sort.h>
31 #include <data/subcase.h>
34 #include <libpspp/misc.h>
36 #include <gsl/gsl_cdf.h>
37 #include <output/table.h>
39 #include <output/charts/plot-chart.h>
40 #include <output/charts/cartesian.h>
43 #define _(msgid) gettext (msgid)
44 #define N_(msgid) msgid
49 const struct variable **vars;
50 const struct dictionary *dict;
52 const struct variable *state_var ;
53 union value state_value;
55 /* Plot the roc curve */
57 /* Plot the reference line */
64 bool bi_neg_exp; /* True iff the bi-negative exponential critieria
66 enum mv_class exclude;
68 bool invert ; /* True iff a smaller test result variable indicates
77 static int run_roc (struct dataset *ds, struct cmd_roc *roc);
80 cmd_roc (struct lexer *lexer, struct dataset *ds)
83 const struct dictionary *dict = dataset_dict (ds);
88 roc.print_coords = false;
91 roc.reference = false;
93 roc.bi_neg_exp = false;
95 roc.pos = roc.pos_weighted = 0;
96 roc.neg = roc.neg_weighted = 0;
97 roc.dict = dataset_dict (ds);
99 if (!parse_variables_const (lexer, dict, &roc.vars, &roc.n_vars,
100 PV_APPEND | PV_NO_DUPLICATE | PV_NUMERIC))
103 if ( ! lex_force_match (lexer, T_BY))
108 roc.state_var = parse_variable (lexer, dict);
110 if ( !lex_force_match (lexer, '('))
115 parse_value (lexer, &roc.state_value, var_get_width (roc.state_var));
118 if ( !lex_force_match (lexer, ')'))
124 while (lex_token (lexer) != '.')
126 lex_match (lexer, '/');
127 if (lex_match_id (lexer, "MISSING"))
129 lex_match (lexer, '=');
130 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
132 if (lex_match_id (lexer, "INCLUDE"))
134 roc.exclude = MV_SYSTEM;
136 else if (lex_match_id (lexer, "EXCLUDE"))
138 roc.exclude = MV_ANY;
142 lex_error (lexer, NULL);
147 else if (lex_match_id (lexer, "PLOT"))
149 lex_match (lexer, '=');
150 if (lex_match_id (lexer, "CURVE"))
153 if (lex_match (lexer, '('))
155 roc.reference = true;
156 lex_force_match_id (lexer, "REFERENCE");
157 lex_force_match (lexer, ')');
160 else if (lex_match_id (lexer, "NONE"))
166 lex_error (lexer, NULL);
170 else if (lex_match_id (lexer, "PRINT"))
172 lex_match (lexer, '=');
173 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
175 if (lex_match_id (lexer, "SE"))
179 else if (lex_match_id (lexer, "COORDINATES"))
181 roc.print_coords = true;
185 lex_error (lexer, NULL);
190 else if (lex_match_id (lexer, "CRITERIA"))
192 lex_match (lexer, '=');
193 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
195 if (lex_match_id (lexer, "CUTOFF"))
197 lex_force_match (lexer, '(');
198 if (lex_match_id (lexer, "INCLUDE"))
200 roc.exclude = MV_SYSTEM;
202 else if (lex_match_id (lexer, "EXCLUDE"))
204 roc.exclude = MV_USER | MV_SYSTEM;
208 lex_error (lexer, NULL);
211 lex_force_match (lexer, ')');
213 else if (lex_match_id (lexer, "TESTPOS"))
215 lex_force_match (lexer, '(');
216 if (lex_match_id (lexer, "LARGE"))
220 else if (lex_match_id (lexer, "SMALL"))
226 lex_error (lexer, NULL);
229 lex_force_match (lexer, ')');
231 else if (lex_match_id (lexer, "CI"))
233 lex_force_match (lexer, '(');
234 lex_force_num (lexer);
235 roc.ci = lex_number (lexer);
237 lex_force_match (lexer, ')');
239 else if (lex_match_id (lexer, "DISTRIBUTION"))
241 lex_force_match (lexer, '(');
242 if (lex_match_id (lexer, "FREE"))
244 roc.bi_neg_exp = false;
246 else if (lex_match_id (lexer, "NEGEXPO"))
248 roc.bi_neg_exp = true;
252 lex_error (lexer, NULL);
255 lex_force_match (lexer, ')');
259 lex_error (lexer, NULL);
266 lex_error (lexer, NULL);
280 do_roc (struct cmd_roc *roc, struct casereader *group, struct dictionary *dict);
284 run_roc (struct dataset *ds, struct cmd_roc *roc)
286 struct dictionary *dict = dataset_dict (ds);
288 struct casereader *group;
290 struct casegrouper *grouper = casegrouper_create_splits (proc_open (ds), dict);
291 while (casegrouper_get_next_group (grouper, &group))
293 do_roc (roc, group, dataset_dict (ds));
295 ok = casegrouper_destroy (grouper);
296 ok = proc_commit (ds) && ok;
303 dump_casereader (struct casereader *reader)
306 struct casereader *r = casereader_clone (reader);
308 for ( ; (c = casereader_read (r) ); case_unref (c))
311 for (i = 0 ; i < case_get_value_cnt (c); ++i)
313 printf ("%g ", case_data_idx (c, i)->f);
318 casereader_destroy (r);
323 match_positives (const struct ccase *c, void *aux)
325 struct cmd_roc *roc = aux;
326 const struct variable *wv = dict_get_weight (roc->dict);
327 const double weight = wv ? case_data (c, wv)->f : 1.0;
329 bool positive = ( 0 == value_compare_3way (case_data (c, roc->state_var),
331 var_get_width (roc->state_var)));
336 roc->pos_weighted += weight;
341 roc->neg_weighted += weight;
362 struct casewriter *cutpoint_wtr;
363 struct casereader *cutpoint_rdr;
378 static struct casereader *
379 accumulate_counts (struct casereader *cutpoint_rdr,
380 double result, double weight,
381 bool (*pos_cond) (double, double),
382 int true_index, int false_index)
384 const struct caseproto *proto = casereader_get_proto (cutpoint_rdr);
385 struct casewriter *w =
386 autopaging_writer_create (proto);
387 struct casereader *r = casereader_clone (cutpoint_rdr);
389 double prev_cp = SYSMIS;
392 for ( ; (cpc = casereader_read (r) ); case_unref (cpc))
394 struct ccase *new_case;
395 const double cp = case_data_idx (cpc, CUTPOINT)->f;
397 /* We don't want duplicates here */
401 new_case = case_clone (cpc);
403 if ( pos_cond (result, cp))
405 case_data_rw_idx (new_case, true_index)->f += weight;
409 case_data_rw_idx (new_case, false_index)->f += weight;
414 casewriter_write (w, new_case);
416 casereader_destroy (r);
418 return casewriter_make_reader (w);
423 static void output_roc (struct roc_state *rs, const struct cmd_roc *roc);
426 static struct casereader *
427 process_group (const struct variable *var, struct casereader *reader,
428 bool (*pred) (double, double),
429 const struct dictionary *dict,
431 struct casereader **cutpoint_rdr,
432 bool (*pos_cond) (double, double),
436 const struct variable *w = dict_get_weight (dict);
438 struct casereader *r1 =
439 casereader_create_distinct (sort_execute_1var (reader, var), var, w);
441 const int weight_idx = w ? var_get_case_index (w) :
442 caseproto_get_n_widths (casereader_get_proto (r1)) - 1;
446 struct casereader *rclone = casereader_clone (r1);
447 struct casewriter *wtr;
448 struct caseproto *proto = caseproto_create ();
450 proto = caseproto_add_width (proto, 0);
451 proto = caseproto_add_width (proto, 0);
452 proto = caseproto_add_width (proto, 0);
454 wtr = autopaging_writer_create (proto);
458 for ( ; (c1 = casereader_read (r1) ); case_unref (c1))
460 struct ccase *new_case = case_create (proto);
462 struct casereader *r2 = casereader_clone (rclone);
464 const double weight1 = case_data_idx (c1, weight_idx)->f;
465 const double d1 = case_data (c1, var)->f;
469 *cutpoint_rdr = accumulate_counts (*cutpoint_rdr, d1, weight1,
471 true_index, false_index);
475 for ( ; (c2 = casereader_read (r2) ); case_unref (c2))
477 const double d2 = case_data (c2, var)->f;
478 const double weight2 = case_data_idx (c2, weight_idx)->f;
485 else if ( pred (d2, d1))
491 case_data_rw_idx (new_case, VALUE)->f = d1;
492 case_data_rw_idx (new_case, N_EQ)->f = n_eq;
493 case_data_rw_idx (new_case, N_PRED)->f = n_pred;
495 casewriter_write (wtr, new_case);
497 casereader_destroy (r2);
500 casereader_destroy (r1);
501 casereader_destroy (rclone);
503 return casewriter_make_reader (wtr);
507 gt (double d1, double d2)
514 ge (double d1, double d2)
520 lt (double d1, double d2)
525 static struct casereader *
526 process_positive_group (const struct variable *var, struct casereader *reader,
527 const struct dictionary *dict,
528 struct roc_state *rs)
530 return process_group (var, reader, gt, dict, &rs->n1,
537 static struct casereader *
538 process_negative_group (const struct variable *var, struct casereader *reader,
539 const struct dictionary *dict,
540 struct roc_state *rs)
542 return process_group (var, reader, lt, dict, &rs->n2,
552 append_cutpoint (struct casewriter *writer, double cutpoint)
554 struct ccase *cc = case_create (casewriter_get_proto (writer));
556 case_data_rw_idx (cc, CUTPOINT)->f = cutpoint;
557 case_data_rw_idx (cc, TP)->f = 0;
558 case_data_rw_idx (cc, FN)->f = 0;
559 case_data_rw_idx (cc, TN)->f = 0;
560 case_data_rw_idx (cc, FP)->f = 0;
563 casewriter_write (writer, cc);
567 /* Prepare the cutpoints */
569 prepare_cutpoints (struct cmd_roc *roc, struct roc_state *rs, struct casereader *input)
572 struct casereader *r = casereader_clone (input);
574 struct caseproto *proto = caseproto_create ();
576 struct subcase ordering;
577 subcase_init (&ordering, CUTPOINT, 0, SC_ASCEND);
579 proto = caseproto_add_width (proto, 0); /* cutpoint */
580 proto = caseproto_add_width (proto, 0); /* TP */
581 proto = caseproto_add_width (proto, 0); /* FN */
582 proto = caseproto_add_width (proto, 0); /* TN */
583 proto = caseproto_add_width (proto, 0); /* FP */
585 for (i = 0 ; i < roc->n_vars; ++i)
587 rs[i].cutpoint_wtr = sort_create_writer (&ordering, proto);
588 rs[i].prev_result = SYSMIS;
589 rs[i].max = -DBL_MAX;
593 for (; (c = casereader_read (r)) != NULL; case_unref (c))
595 for (i = 0 ; i < roc->n_vars; ++i)
597 const union value *v = case_data (c, roc->vars[i]);
598 const double result = v->f;
600 if ( mv_is_value_missing (var_get_missing_values (roc->vars[i]), v, roc->exclude))
603 minimize (&rs[i].min, result);
604 maximize (&rs[i].max, result);
606 if ( rs[i].prev_result != SYSMIS && rs[i].prev_result != result )
608 const double mean = (result + rs[i].prev_result ) / 2.0;
609 append_cutpoint (rs[i].cutpoint_wtr, mean);
612 rs[i].prev_result = result;
615 casereader_destroy (r);
618 /* Append the min and max cutpoints */
619 for (i = 0 ; i < roc->n_vars; ++i)
621 append_cutpoint (rs[i].cutpoint_wtr, rs[i].min - 1);
622 append_cutpoint (rs[i].cutpoint_wtr, rs[i].max + 1);
624 rs[i].cutpoint_rdr = casewriter_make_reader (rs[i].cutpoint_wtr);
629 do_roc (struct cmd_roc *roc, struct casereader *reader, struct dictionary *dict)
633 struct roc_state *rs = xcalloc (roc->n_vars, sizeof *rs);
635 struct casereader *negatives = NULL;
636 struct casereader *positives = NULL;
638 struct caseproto *n_proto = caseproto_create ();
640 struct subcase up_ordering;
641 struct subcase down_ordering;
643 struct casewriter *neg_wtr = NULL;
645 struct casereader *input = casereader_create_filter_missing (reader,
646 roc->vars, roc->n_vars,
651 input = casereader_create_filter_missing (input,
657 neg_wtr = autopaging_writer_create (casereader_get_proto (input));
659 prepare_cutpoints (roc, rs, input);
662 casereader_create_filter_func (input,
668 n_proto = caseproto_create ();
670 n_proto = caseproto_add_width (n_proto, 0);
671 n_proto = caseproto_add_width (n_proto, 0);
672 n_proto = caseproto_add_width (n_proto, 0);
673 n_proto = caseproto_add_width (n_proto, 0);
674 n_proto = caseproto_add_width (n_proto, 0);
676 subcase_init (&up_ordering, VALUE, 0, SC_ASCEND);
677 subcase_init (&down_ordering, VALUE, 0, SC_DESCEND);
679 for (i = 0 ; i < roc->n_vars; ++i)
681 struct casewriter *w = NULL;
682 struct casereader *r = NULL;
687 struct casereader *n_neg ;
688 const struct variable *var = roc->vars[i];
690 struct casereader *neg ;
691 struct casereader *pos = casereader_clone (positives);
693 struct casereader *n_pos =
694 process_positive_group (var, pos, dict, &rs[i]);
696 if ( negatives == NULL)
698 negatives = casewriter_make_reader (neg_wtr);
701 neg = casereader_clone (negatives);
703 n_neg = process_negative_group (var, neg, dict, &rs[i]);
705 w = sort_create_writer (&up_ordering, n_proto);
706 for ( ; (cpos = casereader_read (n_pos) ); case_unref (cpos))
708 struct ccase *pos_case = case_create (n_proto);
710 const double jpos = case_data_idx (cpos, VALUE)->f;
712 while ((cneg = casereader_read (n_neg)))
714 struct ccase *nc = case_create (n_proto);
716 const double jneg = case_data_idx (cneg, VALUE)->f;
718 case_data_rw_idx (nc, VALUE)->f = jneg;
719 case_data_rw_idx (nc, N_EQ)->f = 0;
721 case_data_rw_idx (nc, N_PRED)->f = SYSMIS;
723 *case_data_rw_idx (nc, 3) = *case_data_idx (cneg, N_EQ);
724 *case_data_rw_idx (nc, 4) = *case_data_idx (cneg, N_PRED);
726 casewriter_write (w, nc);
733 case_data_rw_idx (pos_case, VALUE)->f = jpos;
734 *case_data_rw_idx (pos_case, N_EQ) = *case_data_idx (cpos, N_EQ);
735 *case_data_rw_idx (pos_case, N_PRED) = *case_data_idx (cpos, N_PRED);
736 case_data_rw_idx (pos_case, 3)->f = 0;
737 case_data_rw_idx (pos_case, 4)->f = SYSMIS;
739 casewriter_write (w, pos_case);
742 r = casewriter_make_reader (w);
745 double prev_pos_gt = rs[i].n1;
746 w = sort_create_writer (&down_ordering, n_proto);
748 for ( ; (c = casereader_read (r) ); case_unref (c))
750 double n_pos_gt = case_data_idx (c, N_PRED)->f;
751 struct ccase *nc = case_clone (c);
753 if ( n_pos_gt == SYSMIS)
755 n_pos_gt = prev_pos_gt;
756 case_data_rw_idx (nc, N_PRED)->f = n_pos_gt;
759 casewriter_write (w, nc);
760 prev_pos_gt = n_pos_gt;
763 r = casewriter_make_reader (w);
767 double prev_neg_lt = rs[i].n2;
768 w = sort_create_writer (&up_ordering, n_proto);
770 for ( ; (c = casereader_read (r) ); case_unref (c))
772 double n_neg_lt = case_data_idx (c, 4)->f;
773 struct ccase *nc = case_clone (c);
775 if ( n_neg_lt == SYSMIS)
777 n_neg_lt = prev_neg_lt;
778 case_data_rw_idx (nc, 4)->f = n_neg_lt;
781 casewriter_write (w, nc);
782 prev_neg_lt = n_neg_lt;
785 r = casewriter_make_reader (w);
789 struct ccase *prev_case = NULL;
790 for ( ; (c = casereader_read (r) ); case_unref (c))
792 const struct ccase *next_case = casereader_peek (r, 0);
794 const double j = case_data_idx (c, VALUE)->f;
795 double n_pos_eq = case_data_idx (c, N_EQ)->f;
796 double n_pos_gt = case_data_idx (c, N_PRED)->f;
797 double n_neg_eq = case_data_idx (c, 3)->f;
798 double n_neg_lt = case_data_idx (c, 4)->f;
800 if ( prev_case && j == case_data_idx (prev_case, VALUE)->f)
802 if ( 0 == case_data_idx (c, N_EQ)->f)
804 n_pos_eq = case_data_idx (prev_case, N_EQ)->f;
805 n_pos_gt = case_data_idx (prev_case, N_PRED)->f;
808 if ( 0 == case_data_idx (c, 3)->f)
810 n_neg_eq = case_data_idx (prev_case, 3)->f;
811 n_neg_lt = case_data_idx (prev_case, 4)->f;
815 if ( NULL == next_case || j != case_data_idx (next_case, VALUE)->f)
817 rs[i].auc += n_pos_gt * n_neg_eq + (n_pos_eq * n_neg_eq) / 2.0;
820 n_neg_eq * ( pow2 (n_pos_gt) + n_pos_gt * n_pos_eq + pow2 (n_pos_eq) / 3.0);
822 n_pos_eq * ( pow2 (n_neg_lt) + n_neg_lt * n_neg_eq + pow2 (n_neg_eq) / 3.0);
826 case_unref (prev_case);
827 prev_case = case_clone (c);
830 rs[i].auc /= rs[i].n1 * rs[i].n2;
832 rs[i].auc = 1 - rs[i].auc;
834 if ( roc->bi_neg_exp )
836 rs[i].q1hat = rs[i].auc / ( 2 - rs[i].auc);
837 rs[i].q2hat = 2 * pow2 (rs[i].auc) / ( 1 + rs[i].auc);
841 rs[i].q1hat /= rs[i].n2 * pow2 (rs[i].n1);
842 rs[i].q2hat /= rs[i].n1 * pow2 (rs[i].n2);
847 casereader_destroy (positives);
848 casereader_destroy (negatives);
850 output_roc (rs, roc);
856 show_auc (struct roc_state *rs, const struct cmd_roc *roc)
859 const int n_fields = roc->print_se ? 5 : 1;
860 const int n_cols = roc->n_vars > 1 ? n_fields + 1: n_fields;
861 const int n_rows = 2 + roc->n_vars;
862 struct tab_table *tbl = tab_create (n_cols, n_rows, 0);
864 if ( roc->n_vars > 1)
865 tab_title (tbl, _("Area Under the Curve"));
867 tab_title (tbl, _("Area Under the Curve (%s)"), var_to_string (roc->vars[0]));
869 tab_headers (tbl, n_cols - n_fields, 0, 1, 0);
871 tab_dim (tbl, tab_natural_dimensions, NULL);
873 tab_text (tbl, n_cols - n_fields, 1, TAT_TITLE, _("Area"));
875 tab_hline (tbl, TAL_2, 0, n_cols - 1, 2);
886 tab_text (tbl, n_cols - 4, 1, TAT_TITLE, _("Std. Error"));
887 tab_text (tbl, n_cols - 3, 1, TAT_TITLE, _("Asymptotic Sig."));
889 tab_text (tbl, n_cols - 2, 1, TAT_TITLE, _("Lower Bound"));
890 tab_text (tbl, n_cols - 1, 1, TAT_TITLE, _("Upper Bound"));
892 tab_joint_text (tbl, n_cols - 2, 0, 4, 0,
893 TAT_TITLE | TAB_CENTER | TAT_PRINTF,
894 _("Asymp. %g%% Confidence Interval"), roc->ci);
895 tab_vline (tbl, 0, n_cols - 1, 0, 0);
896 tab_hline (tbl, TAL_1, n_cols - 2, n_cols - 1, 1);
899 if ( roc->n_vars > 1)
900 tab_text (tbl, 0, 1, TAT_TITLE, _("Variable under test"));
902 if ( roc->n_vars > 1)
903 tab_vline (tbl, TAL_2, 1, 0, n_rows - 1);
906 for ( i = 0 ; i < roc->n_vars ; ++i )
908 tab_text (tbl, 0, 2 + i, TAT_TITLE, var_to_string (roc->vars[i]));
910 tab_double (tbl, n_cols - n_fields, 2 + i, 0, rs[i].auc, NULL);
915 const double sd_0_5 = sqrt ((rs[i].n1 + rs[i].n2 + 1) /
916 (12 * rs[i].n1 * rs[i].n2));
920 se = rs[i].auc * (1 - rs[i].auc) + (rs[i].n1 - 1) * (rs[i].q1hat - pow2 (rs[i].auc)) +
921 (rs[i].n2 - 1) * (rs[i].q2hat - pow2 (rs[i].auc));
923 se /= rs[i].n1 * rs[i].n2;
927 tab_double (tbl, n_cols - 4, 2 + i, 0,
931 ci = 1 - roc->ci / 100.0;
932 yy = gsl_cdf_gaussian_Qinv (ci, se) ;
934 tab_double (tbl, n_cols - 2, 2 + i, 0,
938 tab_double (tbl, n_cols - 1, 2 + i, 0,
942 tab_double (tbl, n_cols - 3, 2 + i, 0,
943 2.0 * gsl_cdf_ugaussian_Q (fabs ((rs[i].auc - 0.5 ) / sd_0_5)),
953 show_summary (const struct cmd_roc *roc)
955 const int n_cols = 3;
956 const int n_rows = 4;
957 struct tab_table *tbl = tab_create (n_cols, n_rows, 0);
959 tab_title (tbl, _("Case Summary"));
961 tab_headers (tbl, 1, 0, 2, 0);
963 tab_dim (tbl, tab_natural_dimensions, NULL);
972 tab_hline (tbl, TAL_2, 0, n_cols - 1, 2);
973 tab_vline (tbl, TAL_2, 1, 0, n_rows - 1);
976 tab_hline (tbl, TAL_2, 1, n_cols - 1, 1);
977 tab_vline (tbl, TAL_1, 2, 1, n_rows - 1);
980 tab_text (tbl, 0, 1, TAT_TITLE | TAB_LEFT, var_to_string (roc->state_var));
981 tab_text (tbl, 1, 1, TAT_TITLE, _("Unweighted"));
982 tab_text (tbl, 2, 1, TAT_TITLE, _("Weighted"));
984 tab_joint_text (tbl, 1, 0, 2, 0,
985 TAT_TITLE | TAB_CENTER,
986 _("Valid N (listwise)"));
989 tab_text (tbl, 0, 2, TAB_LEFT, _("Positive"));
990 tab_text (tbl, 0, 3, TAB_LEFT, _("Negative"));
993 tab_double (tbl, 1, 2, 0, roc->pos, &F_8_0);
994 tab_double (tbl, 1, 3, 0, roc->neg, &F_8_0);
996 tab_double (tbl, 2, 2, 0, roc->pos_weighted, 0);
997 tab_double (tbl, 2, 3, 0, roc->neg_weighted, 0);
1004 show_coords (struct roc_state *rs, const struct cmd_roc *roc)
1008 const int n_cols = roc->n_vars > 1 ? 4 : 3;
1010 struct tab_table *tbl ;
1012 for (i = 0; i < roc->n_vars; ++i)
1013 n_rows += casereader_count_cases (rs[i].cutpoint_rdr);
1015 tbl = tab_create (n_cols, n_rows, 0);
1017 if ( roc->n_vars > 1)
1018 tab_title (tbl, _("Coordinates of the Curve"));
1020 tab_title (tbl, _("Coordinates of the Curve (%s)"), var_to_string (roc->vars[0]));
1023 tab_headers (tbl, 1, 0, 1, 0);
1025 tab_dim (tbl, tab_natural_dimensions, NULL);
1027 tab_hline (tbl, TAL_2, 0, n_cols - 1, 1);
1029 if ( roc->n_vars > 1)
1030 tab_text (tbl, 0, 0, TAT_TITLE, _("Test variable"));
1032 tab_text (tbl, n_cols - 3, 0, TAT_TITLE, _("Positive if greater than or equal to"));
1033 tab_text (tbl, n_cols - 2, 0, TAT_TITLE, _("Sensitivity"));
1034 tab_text (tbl, n_cols - 1, 0, TAT_TITLE, _("1 - Specificity"));
1043 if ( roc->n_vars > 1)
1044 tab_vline (tbl, TAL_2, 1, 0, n_rows - 1);
1046 for (i = 0; i < roc->n_vars; ++i)
1049 struct casereader *r = casereader_clone (rs[i].cutpoint_rdr);
1051 if ( roc->n_vars > 1)
1052 tab_text (tbl, 0, x, TAT_TITLE, var_to_string (roc->vars[i]));
1055 tab_hline (tbl, TAL_1, 0, n_cols - 1, x);
1058 for (; (cc = casereader_read (r)) != NULL;
1059 case_unref (cc), x++)
1061 const double se = case_data_idx (cc, TP)->f /
1063 case_data_idx (cc, TP)->f
1065 case_data_idx (cc, FN)->f
1068 const double sp = case_data_idx (cc, TN)->f /
1070 case_data_idx (cc, TN)->f
1072 case_data_idx (cc, FP)->f
1075 tab_double (tbl, n_cols - 3, x, 0, case_data_idx (cc, CUTPOINT)->f,
1076 var_get_print_format (roc->vars[i]));
1078 tab_double (tbl, n_cols - 2, x, 0, se, NULL);
1079 tab_double (tbl, n_cols - 1, x, 0, 1 - sp, NULL);
1082 casereader_destroy (r);
1090 draw_roc (struct roc_state *rs, const struct cmd_roc *roc)
1094 struct chart *roc_chart = chart_create ();
1096 chart_write_title (roc_chart, _("ROC Curve"));
1097 chart_write_xlabel (roc_chart, _("1 - Specificity"));
1098 chart_write_ylabel (roc_chart, _("Sensitivity"));
1100 chart_write_xscale (roc_chart, 0, 1, 5);
1101 chart_write_yscale (roc_chart, 0, 1, 5);
1103 if ( roc->reference )
1105 chart_line (roc_chart, 1.0, 0,
1110 for (i = 0; i < roc->n_vars; ++i)
1113 struct casereader *r = casereader_clone (rs[i].cutpoint_rdr);
1115 chart_vector_start (roc_chart, var_get_name (roc->vars[i]));
1116 for (; (cc = casereader_read (r)) != NULL;
1119 double se = case_data_idx (cc, TP)->f;
1120 double sp = case_data_idx (cc, TN)->f;
1122 se /= case_data_idx (cc, FN)->f +
1123 case_data_idx (cc, TP)->f ;
1125 sp /= case_data_idx (cc, TN)->f +
1126 case_data_idx (cc, FP)->f ;
1128 chart_vector (roc_chart, 1 - sp, se);
1130 chart_vector_end (roc_chart);
1131 casereader_destroy (r);
1134 chart_write_legend (roc_chart);
1136 chart_submit (roc_chart);
1141 output_roc (struct roc_state *rs, const struct cmd_roc *roc)
1151 if ( roc->print_coords )
1152 show_coords (rs, roc);