X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Flanguage%2Fstats%2Froc.c;h=7d13c2c2cb7cf32c13c9d5c5051cf1328441517a;hb=f790dbda9d498eef9c9c0a49078adbeecf768d56;hp=1e8ac4e021a75334cb55a0257f3093e93e487daf;hpb=aa937be5e1ac11fadf7a3c6705416f9b4ad35c1f;p=pspp diff --git a/src/language/stats/roc.c b/src/language/stats/roc.c index 1e8ac4e021..7d13c2c2cb 100644 --- a/src/language/stats/roc.c +++ b/src/language/stats/roc.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 @@ -16,28 +16,26 @@ #include -#include "roc.h" -#include -#include -#include -#include - -#include -#include -#include -#include -#include -#include -#include - - -#include +#include "language/stats/roc.h" #include -#include -#include -#include +#include "data/casegrouper.h" +#include "data/casereader.h" +#include "data/casewriter.h" +#include "data/dataset.h" +#include "data/dictionary.h" +#include "data/format.h" +#include "data/subcase.h" +#include "language/command.h" +#include "language/lexer/lexer.h" +#include "language/lexer/value-parser.h" +#include "language/lexer/variable-parser.h" +#include "libpspp/misc.h" +#include "math/sort.h" +#include "output/chart-item.h" +#include "output/charts/roc-chart.h" +#include "output/tab.h" #include "gettext.h" #define _(msgid) gettext (msgid) @@ -49,8 +47,9 @@ struct cmd_roc const struct variable **vars; const struct dictionary *dict; - const struct variable *state_var ; + const struct variable *state_var; union value state_value; + size_t state_var_width; /* Plot the roc curve */ bool curve; @@ -95,39 +94,47 @@ cmd_roc (struct lexer *lexer, struct dataset *ds) roc.pos = roc.pos_weighted = 0; roc.neg = roc.neg_weighted = 0; roc.dict = dataset_dict (ds); + roc.state_var = NULL; + roc.state_var_width = -1; + lex_match (lexer, T_SLASH); if (!parse_variables_const (lexer, dict, &roc.vars, &roc.n_vars, PV_APPEND | PV_NO_DUPLICATE | PV_NUMERIC)) - return 2; + goto error; if ( ! lex_force_match (lexer, T_BY)) { - return 2; + goto error; } roc.state_var = parse_variable (lexer, dict); + if (! roc.state_var) + { + goto error; + } - if ( !lex_force_match (lexer, '(')) + if ( !lex_force_match (lexer, T_LPAREN)) { - return 2; + goto error; } - parse_value (lexer, &roc.state_value, var_get_width (roc.state_var)); + roc.state_var_width = var_get_width (roc.state_var); + value_init (&roc.state_value, roc.state_var_width); + parse_value (lexer, &roc.state_value, roc.state_var); - if ( !lex_force_match (lexer, ')')) + if ( !lex_force_match (lexer, T_RPAREN)) { - return 2; + goto error; } - - while (lex_token (lexer) != '.') + while (lex_token (lexer) != T_ENDCMD) { - lex_match (lexer, '/'); + lex_match (lexer, T_SLASH); if (lex_match_id (lexer, "MISSING")) { - lex_match (lexer, '='); - while (lex_token (lexer) != '.' && lex_token (lexer) != '/') + lex_match (lexer, T_EQUALS); + while (lex_token (lexer) != T_ENDCMD && lex_token (lexer) != T_SLASH) { if (lex_match_id (lexer, "INCLUDE")) { @@ -140,21 +147,23 @@ cmd_roc (struct lexer *lexer, struct dataset *ds) else { lex_error (lexer, NULL); - return 2; + goto error; } } } else if (lex_match_id (lexer, "PLOT")) { - lex_match (lexer, '='); + lex_match (lexer, T_EQUALS); if (lex_match_id (lexer, "CURVE")) { roc.curve = true; - if (lex_match (lexer, '(')) + if (lex_match (lexer, T_LPAREN)) { roc.reference = true; - lex_force_match_id (lexer, "REFERENCE"); - lex_force_match (lexer, ')'); + if (! lex_force_match_id (lexer, "REFERENCE")) + goto error; + if (! lex_force_match (lexer, T_RPAREN)) + goto error; } } else if (lex_match_id (lexer, "NONE")) @@ -164,13 +173,13 @@ cmd_roc (struct lexer *lexer, struct dataset *ds) else { lex_error (lexer, NULL); - return 2; + goto error; } } else if (lex_match_id (lexer, "PRINT")) { - lex_match (lexer, '='); - while (lex_token (lexer) != '.' && lex_token (lexer) != '/') + lex_match (lexer, T_EQUALS); + while (lex_token (lexer) != T_ENDCMD && lex_token (lexer) != T_SLASH) { if (lex_match_id (lexer, "SE")) { @@ -183,18 +192,19 @@ cmd_roc (struct lexer *lexer, struct dataset *ds) else { lex_error (lexer, NULL); - return 2; + goto error; } } } else if (lex_match_id (lexer, "CRITERIA")) { - lex_match (lexer, '='); - while (lex_token (lexer) != '.' && lex_token (lexer) != '/') + lex_match (lexer, T_EQUALS); + while (lex_token (lexer) != T_ENDCMD && lex_token (lexer) != T_SLASH) { if (lex_match_id (lexer, "CUTOFF")) { - lex_force_match (lexer, '('); + if (! lex_force_match (lexer, T_LPAREN)) + goto error; if (lex_match_id (lexer, "INCLUDE")) { roc.exclude = MV_SYSTEM; @@ -206,13 +216,15 @@ cmd_roc (struct lexer *lexer, struct dataset *ds) else { lex_error (lexer, NULL); - return 2; + goto error; } - lex_force_match (lexer, ')'); + if (! lex_force_match (lexer, T_RPAREN)) + goto error; } else if (lex_match_id (lexer, "TESTPOS")) { - lex_force_match (lexer, '('); + if (! lex_force_match (lexer, T_LPAREN)) + goto error; if (lex_match_id (lexer, "LARGE")) { roc.invert = false; @@ -224,21 +236,26 @@ cmd_roc (struct lexer *lexer, struct dataset *ds) else { lex_error (lexer, NULL); - return 2; + goto error; } - lex_force_match (lexer, ')'); + if (! lex_force_match (lexer, T_RPAREN)) + goto error; } else if (lex_match_id (lexer, "CI")) { - lex_force_match (lexer, '('); - lex_force_num (lexer); + if (!lex_force_match (lexer, T_LPAREN)) + goto error; + if (! lex_force_num (lexer)) + goto error; roc.ci = lex_number (lexer); lex_get (lexer); - lex_force_match (lexer, ')'); + if (!lex_force_match (lexer, T_RPAREN)) + goto error; } else if (lex_match_id (lexer, "DISTRIBUTION")) { - lex_force_match (lexer, '('); + if (!lex_force_match (lexer, T_LPAREN)) + goto error; if (lex_match_id (lexer, "FREE")) { roc.bi_neg_exp = false; @@ -250,14 +267,15 @@ cmd_roc (struct lexer *lexer, struct dataset *ds) else { lex_error (lexer, NULL); - return 2; + goto error; } - lex_force_match (lexer, ')'); + if (!lex_force_match (lexer, T_RPAREN)) + goto error; } else { lex_error (lexer, NULL); - return 2; + goto error; } } } @@ -268,9 +286,19 @@ cmd_roc (struct lexer *lexer, struct dataset *ds) } } - run_roc (ds, &roc); + if ( ! run_roc (ds, &roc)) + goto error; + + if ( roc.state_var) + value_destroy (&roc.state_value, roc.state_var_width); + free (roc.vars); + return CMD_SUCCESS; - return 1; + error: + if ( roc.state_var) + value_destroy (&roc.state_value, roc.state_var_width); + free (roc.vars); + return CMD_FAILURE; } @@ -319,6 +347,13 @@ dump_casereader (struct casereader *reader) } #endif + +/* + Return true iff the state variable indicates that C has positive actual state. + + As a side effect, this function also accumulates the roc->{pos,neg} and + roc->{pos,neg}_weighted counts. + */ static bool match_positives (const struct ccase *c, void *aux) { @@ -326,9 +361,9 @@ match_positives (const struct ccase *c, void *aux) const struct variable *wv = dict_get_weight (roc->dict); const double weight = wv ? case_data (c, wv)->f : 1.0; - bool positive = ( 0 == value_compare_3way (case_data (c, roc->state_var), - &roc->state_value, - var_get_width (roc->state_var))); + const bool positive = + ( 0 == value_compare_3way (case_data (c, roc->state_var), &roc->state_value, + var_get_width (roc->state_var))); if ( positive ) { @@ -349,16 +384,20 @@ match_positives (const struct ccase *c, void *aux) #define N_EQ 1 #define N_PRED 2 +/* Some intermediate state for calculating the cutpoints and the + standard error values */ struct roc_state { - double auc; + double auc; /* Area under the curve */ - double n1; - double n2; + double n1; /* total weight of positives */ + double n2; /* total weight of negatives */ + /* intermediates for standard error */ double q1hat; double q2hat; + /* intermediates for cutpoints */ struct casewriter *cutpoint_wtr; struct casereader *cutpoint_rdr; double prev_result; @@ -366,33 +405,33 @@ struct roc_state double max; }; - - -#define CUTPOINT 0 -#define TP 1 -#define FN 2 -#define TN 3 -#define FP 4 - - +/* + Return a new casereader based upon CUTPOINT_RDR. + The number of "positive" cases are placed into + the position TRUE_INDEX, and the number of "negative" cases + into FALSE_INDEX. + POS_COND and RESULT determine the semantics of what is + "positive". + WEIGHT is the value of a single count. + */ static struct casereader * -accumulate_counts (struct casereader *cutpoint_rdr, - double result, double weight, +accumulate_counts (struct casereader *input, + double result, double weight, bool (*pos_cond) (double, double), int true_index, int false_index) { - const struct caseproto *proto = casereader_get_proto (cutpoint_rdr); + const struct caseproto *proto = casereader_get_proto (input); struct casewriter *w = autopaging_writer_create (proto); - struct casereader *r = casereader_clone (cutpoint_rdr); struct ccase *cpc; double prev_cp = SYSMIS; - - for ( ; (cpc = casereader_read (r) ); case_unref (cpc)) + for ( ; (cpc = casereader_read (input) ); case_unref (cpc)) { struct ccase *new_case; - const double cp = case_data_idx (cpc, CUTPOINT)->f; + const double cp = case_data_idx (cpc, ROC_CUTPOINT)->f; + + assert (cp != SYSMIS); /* We don't want duplicates here */ if ( cp == prev_cp ) @@ -401,19 +440,15 @@ accumulate_counts (struct casereader *cutpoint_rdr, new_case = case_clone (cpc); if ( pos_cond (result, cp)) - { - case_data_rw_idx (new_case, true_index)->f += weight; - } + case_data_rw_idx (new_case, true_index)->f += weight; else - { - case_data_rw_idx (new_case, false_index)->f += weight; - } + case_data_rw_idx (new_case, false_index)->f += weight; prev_cp = cp; casewriter_write (w, new_case); } - casereader_destroy (r); + casereader_destroy (input); return casewriter_make_reader (w); } @@ -422,13 +457,27 @@ accumulate_counts (struct casereader *cutpoint_rdr, static void output_roc (struct roc_state *rs, const struct cmd_roc *roc); +/* + This function does 3 things: + 1. Counts the number of cases which are equal to every other case in READER, + and those cases for which the relationship between it and every other case + satifies PRED (normally either > or <). VAR is variable defining a case's value + for this purpose. + + 2. Counts the number of true and false cases in reader, and populates + CUTPOINT_RDR accordingly. TRUE_INDEX and FALSE_INDEX are the indices + which receive these values. POS_COND is the condition defining true + and false. + + 3. CC is filled with the cumulative weight of all cases of READER. +*/ static struct casereader * process_group (const struct variable *var, struct casereader *reader, bool (*pred) (double, double), const struct dictionary *dict, double *cc, - struct casereader **cutpoint_rdr, + struct casereader **cutpoint_rdr, bool (*pos_cond) (double, double), int true_index, int false_index) @@ -440,7 +489,7 @@ process_group (const struct variable *var, struct casereader *reader, const int weight_idx = w ? var_get_case_index (w) : caseproto_get_n_widths (casereader_get_proto (r1)) - 1; - + struct ccase *c1; struct casereader *rclone = casereader_clone (r1); @@ -451,7 +500,7 @@ process_group (const struct variable *var, struct casereader *reader, proto = caseproto_add_width (proto, 0); proto = caseproto_add_width (proto, 0); - wtr = autopaging_writer_create (proto); + wtr = autopaging_writer_create (proto); *cc = 0; @@ -497,12 +546,21 @@ process_group (const struct variable *var, struct casereader *reader, casereader_destroy (r2); } + casereader_destroy (r1); casereader_destroy (rclone); + caseproto_unref (proto); + return casewriter_make_reader (wtr); } +/* Some more indeces into case data */ +#define N_POS_EQ 1 /* number of positive cases with values equal to n */ +#define N_POS_GT 2 /* number of positive cases with values greater than n */ +#define N_NEG_EQ 3 /* number of negative cases with values equal to n */ +#define N_NEG_LT 4 /* number of negative cases with values less than n */ + static bool gt (double d1, double d2) { @@ -522,6 +580,14 @@ lt (double d1, double d2) return d1 < d2; } + +/* + Return a casereader with width 3, + populated with cases based upon READER. + The cases will have the values: + (N, number of cases equal to N, number of cases greater than N) + As a side effect, update RS->n1 with the number of positive cases. +*/ static struct casereader * process_positive_group (const struct variable *var, struct casereader *reader, const struct dictionary *dict, @@ -530,10 +596,16 @@ process_positive_group (const struct variable *var, struct casereader *reader, return process_group (var, reader, gt, dict, &rs->n1, &rs->cutpoint_rdr, ge, - TP, FN); + ROC_TP, ROC_FN); } - +/* + Return a casereader with width 3, + populated with cases based upon READER. + The cases will have the values: + (N, number of cases equal to N, number of cases less than N) + As a side effect, update RS->n2 with the number of negative cases. +*/ static struct casereader * process_negative_group (const struct variable *var, struct casereader *reader, const struct dictionary *dict, @@ -542,7 +614,7 @@ process_negative_group (const struct variable *var, struct casereader *reader, return process_group (var, reader, lt, dict, &rs->n2, &rs->cutpoint_rdr, lt, - TN, FP); + ROC_TN, ROC_FP); } @@ -553,48 +625,58 @@ append_cutpoint (struct casewriter *writer, double cutpoint) { struct ccase *cc = case_create (casewriter_get_proto (writer)); - case_data_rw_idx (cc, CUTPOINT)->f = cutpoint; - case_data_rw_idx (cc, TP)->f = 0; - case_data_rw_idx (cc, FN)->f = 0; - case_data_rw_idx (cc, TN)->f = 0; - case_data_rw_idx (cc, FP)->f = 0; - + case_data_rw_idx (cc, ROC_CUTPOINT)->f = cutpoint; + case_data_rw_idx (cc, ROC_TP)->f = 0; + case_data_rw_idx (cc, ROC_FN)->f = 0; + case_data_rw_idx (cc, ROC_TN)->f = 0; + case_data_rw_idx (cc, ROC_FP)->f = 0; casewriter_write (writer, cc); } -/* Prepare the cutpoints */ +/* + Create and initialise the rs[x].cutpoint_rdr casereaders. That is, the readers will + be created with width 5, ready to take the values (cutpoint, ROC_TP, ROC_FN, ROC_TN, ROC_FP), and the + reader will be populated with its final number of cases. + However on exit from this function, only ROC_CUTPOINT entries will be set to their final + value. The other entries will be initialised to zero. +*/ static void prepare_cutpoints (struct cmd_roc *roc, struct roc_state *rs, struct casereader *input) { int i; struct casereader *r = casereader_clone (input); struct ccase *c; - struct caseproto *proto = caseproto_create (); - struct subcase ordering; - subcase_init (&ordering, CUTPOINT, 0, SC_ASCEND); + { + struct caseproto *proto = caseproto_create (); + struct subcase ordering; + subcase_init (&ordering, ROC_CUTPOINT, 0, SC_ASCEND); - proto = caseproto_add_width (proto, 0); /* cutpoint */ - proto = caseproto_add_width (proto, 0); /* TP */ - proto = caseproto_add_width (proto, 0); /* FN */ - proto = caseproto_add_width (proto, 0); /* TN */ - proto = caseproto_add_width (proto, 0); /* FP */ + proto = caseproto_add_width (proto, 0); /* cutpoint */ + proto = caseproto_add_width (proto, 0); /* ROC_TP */ + proto = caseproto_add_width (proto, 0); /* ROC_FN */ + proto = caseproto_add_width (proto, 0); /* ROC_TN */ + proto = caseproto_add_width (proto, 0); /* ROC_FP */ - for (i = 0 ; i < roc->n_vars; ++i) - { - rs[i].cutpoint_wtr = sort_create_writer (&ordering, proto); - rs[i].prev_result = SYSMIS; - rs[i].max = -DBL_MAX; - rs[i].min = DBL_MAX; - } + for (i = 0 ; i < roc->n_vars; ++i) + { + rs[i].cutpoint_wtr = sort_create_writer (&ordering, proto); + rs[i].prev_result = SYSMIS; + rs[i].max = -DBL_MAX; + rs[i].min = DBL_MAX; + } + + caseproto_unref (proto); + subcase_destroy (&ordering); + } for (; (c = casereader_read (r)) != NULL; case_unref (c)) { for (i = 0 ; i < roc->n_vars; ++i) { - const union value *v = case_data (c, roc->vars[i]); + const union value *v = case_data (c, roc->vars[i]); const double result = v->f; if ( mv_is_value_missing (var_get_missing_values (roc->vars[i]), v, roc->exclude)) @@ -635,7 +717,7 @@ do_roc (struct cmd_roc *roc, struct casereader *reader, struct dictionary *dict) struct casereader *negatives = NULL; struct casereader *positives = NULL; - struct caseproto *n_proto = caseproto_create (); + struct caseproto *n_proto = NULL; struct subcase up_ordering; struct subcase down_ordering; @@ -658,7 +740,9 @@ do_roc (struct cmd_roc *roc, struct casereader *reader, struct dictionary *dict) prepare_cutpoints (roc, rs, input); - positives = + + /* Separate the positive actual state cases from the negative ones */ + positives = casereader_create_filter_func (input, match_positives, NULL, @@ -666,7 +750,7 @@ do_roc (struct cmd_roc *roc, struct casereader *reader, struct dictionary *dict) neg_wtr); n_proto = caseproto_create (); - + n_proto = caseproto_add_width (n_proto, 0); n_proto = caseproto_add_width (n_proto, 0); n_proto = caseproto_add_width (n_proto, 0); @@ -684,13 +768,13 @@ do_roc (struct cmd_roc *roc, struct casereader *reader, struct dictionary *dict) struct ccase *c; struct ccase *cpos; - struct casereader *n_neg ; + struct casereader *n_neg_reader ; const struct variable *var = roc->vars[i]; struct casereader *neg ; struct casereader *pos = casereader_clone (positives); - struct casereader *n_pos = + struct casereader *n_pos_reader = process_positive_group (var, pos, dict, &rs[i]); if ( negatives == NULL) @@ -700,28 +784,29 @@ do_roc (struct cmd_roc *roc, struct casereader *reader, struct dictionary *dict) neg = casereader_clone (negatives); - n_neg = process_negative_group (var, neg, dict, &rs[i]); + n_neg_reader = process_negative_group (var, neg, dict, &rs[i]); + /* Merge the n_pos and n_neg casereaders */ w = sort_create_writer (&up_ordering, n_proto); - for ( ; (cpos = casereader_read (n_pos) ); case_unref (cpos)) + for ( ; (cpos = casereader_read (n_pos_reader) ); case_unref (cpos)) { struct ccase *pos_case = case_create (n_proto); struct ccase *cneg; const double jpos = case_data_idx (cpos, VALUE)->f; - while ((cneg = casereader_read (n_neg))) + while ((cneg = casereader_read (n_neg_reader))) { struct ccase *nc = case_create (n_proto); const double jneg = case_data_idx (cneg, VALUE)->f; case_data_rw_idx (nc, VALUE)->f = jneg; - case_data_rw_idx (nc, N_EQ)->f = 0; + case_data_rw_idx (nc, N_POS_EQ)->f = 0; - case_data_rw_idx (nc, N_PRED)->f = SYSMIS; + case_data_rw_idx (nc, N_POS_GT)->f = SYSMIS; - *case_data_rw_idx (nc, 3) = *case_data_idx (cneg, N_EQ); - *case_data_rw_idx (nc, 4) = *case_data_idx (cneg, N_PRED); + *case_data_rw_idx (nc, N_NEG_EQ) = *case_data_idx (cneg, N_EQ); + *case_data_rw_idx (nc, N_NEG_LT) = *case_data_idx (cneg, N_PRED); casewriter_write (w, nc); @@ -731,57 +816,70 @@ do_roc (struct cmd_roc *roc, struct casereader *reader, struct dictionary *dict) } case_data_rw_idx (pos_case, VALUE)->f = jpos; - *case_data_rw_idx (pos_case, N_EQ) = *case_data_idx (cpos, N_EQ); - *case_data_rw_idx (pos_case, N_PRED) = *case_data_idx (cpos, N_PRED); - case_data_rw_idx (pos_case, 3)->f = 0; - case_data_rw_idx (pos_case, 4)->f = SYSMIS; + *case_data_rw_idx (pos_case, N_POS_EQ) = *case_data_idx (cpos, N_EQ); + *case_data_rw_idx (pos_case, N_POS_GT) = *case_data_idx (cpos, N_PRED); + case_data_rw_idx (pos_case, N_NEG_EQ)->f = 0; + case_data_rw_idx (pos_case, N_NEG_LT)->f = SYSMIS; casewriter_write (w, pos_case); } + casereader_destroy (n_pos_reader); + casereader_destroy (n_neg_reader); + +/* These aren't used anymore */ +#undef N_EQ +#undef N_PRED + r = casewriter_make_reader (w); + /* Propagate the N_POS_GT values from the positive cases + to the negative ones */ { double prev_pos_gt = rs[i].n1; w = sort_create_writer (&down_ordering, n_proto); for ( ; (c = casereader_read (r) ); case_unref (c)) { - double n_pos_gt = case_data_idx (c, N_PRED)->f; + double n_pos_gt = case_data_idx (c, N_POS_GT)->f; struct ccase *nc = case_clone (c); if ( n_pos_gt == SYSMIS) { n_pos_gt = prev_pos_gt; - case_data_rw_idx (nc, N_PRED)->f = n_pos_gt; + case_data_rw_idx (nc, N_POS_GT)->f = n_pos_gt; } - + casewriter_write (w, nc); prev_pos_gt = n_pos_gt; } + casereader_destroy (r); r = casewriter_make_reader (w); } + /* Propagate the N_NEG_LT values from the negative cases + to the positive ones */ { double prev_neg_lt = rs[i].n2; w = sort_create_writer (&up_ordering, n_proto); for ( ; (c = casereader_read (r) ); case_unref (c)) { - double n_neg_lt = case_data_idx (c, 4)->f; + double n_neg_lt = case_data_idx (c, N_NEG_LT)->f; struct ccase *nc = case_clone (c); if ( n_neg_lt == SYSMIS) { n_neg_lt = prev_neg_lt; - case_data_rw_idx (nc, 4)->f = n_neg_lt; + case_data_rw_idx (nc, N_NEG_LT)->f = n_neg_lt; } - + casewriter_write (w, nc); prev_neg_lt = n_neg_lt; } + casereader_destroy (r); r = casewriter_make_reader (w); } @@ -789,26 +887,26 @@ do_roc (struct cmd_roc *roc, struct casereader *reader, struct dictionary *dict) struct ccase *prev_case = NULL; for ( ; (c = casereader_read (r) ); case_unref (c)) { - const struct ccase *next_case = casereader_peek (r, 0); + struct ccase *next_case = casereader_peek (r, 0); const double j = case_data_idx (c, VALUE)->f; - double n_pos_eq = case_data_idx (c, N_EQ)->f; - double n_pos_gt = case_data_idx (c, N_PRED)->f; - double n_neg_eq = case_data_idx (c, 3)->f; - double n_neg_lt = case_data_idx (c, 4)->f; + double n_pos_eq = case_data_idx (c, N_POS_EQ)->f; + double n_pos_gt = case_data_idx (c, N_POS_GT)->f; + double n_neg_eq = case_data_idx (c, N_NEG_EQ)->f; + double n_neg_lt = case_data_idx (c, N_NEG_LT)->f; if ( prev_case && j == case_data_idx (prev_case, VALUE)->f) { - if ( 0 == case_data_idx (c, N_EQ)->f) + if ( 0 == case_data_idx (c, N_POS_EQ)->f) { - n_pos_eq = case_data_idx (prev_case, N_EQ)->f; - n_pos_gt = case_data_idx (prev_case, N_PRED)->f; + n_pos_eq = case_data_idx (prev_case, N_POS_EQ)->f; + n_pos_gt = case_data_idx (prev_case, N_POS_GT)->f; } - if ( 0 == case_data_idx (c, 3)->f) + if ( 0 == case_data_idx (c, N_NEG_EQ)->f) { - n_neg_eq = case_data_idx (prev_case, 3)->f; - n_neg_lt = case_data_idx (prev_case, 4)->f; + n_neg_eq = case_data_idx (prev_case, N_NEG_EQ)->f; + n_neg_lt = case_data_idx (prev_case, N_NEG_LT)->f; } } @@ -823,12 +921,15 @@ do_roc (struct cmd_roc *roc, struct casereader *reader, struct dictionary *dict) } + case_unref (next_case); case_unref (prev_case); prev_case = case_clone (c); } + casereader_destroy (r); + case_unref (prev_case); - rs[i].auc /= rs[i].n1 * rs[i].n2; - if ( roc->invert ) + rs[i].auc /= rs[i].n1 * rs[i].n2; + if ( roc->invert ) rs[i].auc = 1 - rs[i].auc; if ( roc->bi_neg_exp ) @@ -847,8 +948,15 @@ do_roc (struct cmd_roc *roc, struct casereader *reader, struct dictionary *dict) casereader_destroy (positives); casereader_destroy (negatives); + caseproto_unref (n_proto); + subcase_destroy (&up_ordering); + subcase_destroy (&down_ordering); + output_roc (rs, roc); + for (i = 0 ; i < roc->n_vars; ++i) + casereader_destroy (rs[i].cutpoint_rdr); + free (rs); } @@ -859,7 +967,7 @@ show_auc (struct roc_state *rs, const struct cmd_roc *roc) const int n_fields = roc->print_se ? 5 : 1; const int n_cols = roc->n_vars > 1 ? n_fields + 1: n_fields; const int n_rows = 2 + roc->n_vars; - struct tab_table *tbl = tab_create (n_cols, n_rows, 0); + struct tab_table *tbl = tab_create (n_cols, n_rows); if ( roc->n_vars > 1) tab_title (tbl, _("Area Under the Curve")); @@ -868,7 +976,6 @@ show_auc (struct roc_state *rs, const struct cmd_roc *roc) tab_headers (tbl, n_cols - n_fields, 0, 1, 0); - tab_dim (tbl, tab_natural_dimensions, NULL); tab_text (tbl, n_cols - n_fields, 1, TAT_TITLE, _("Area")); @@ -889,9 +996,9 @@ show_auc (struct roc_state *rs, const struct cmd_roc *roc) tab_text (tbl, n_cols - 2, 1, TAT_TITLE, _("Lower Bound")); tab_text (tbl, n_cols - 1, 1, TAT_TITLE, _("Upper Bound")); - tab_joint_text (tbl, n_cols - 2, 0, 4, 0, - TAT_TITLE | TAB_CENTER | TAT_PRINTF, - _("Asymp. %g%% Confidence Interval"), roc->ci); + tab_joint_text_format (tbl, n_cols - 2, 0, 4, 0, + TAT_TITLE | TAB_CENTER, + _("Asymp. %g%% Confidence Interval"), roc->ci); tab_vline (tbl, 0, n_cols - 1, 0, 0); tab_hline (tbl, TAL_1, n_cols - 2, n_cols - 1, 1); } @@ -907,7 +1014,7 @@ show_auc (struct roc_state *rs, const struct cmd_roc *roc) { tab_text (tbl, 0, 2 + i, TAT_TITLE, var_to_string (roc->vars[i])); - tab_double (tbl, n_cols - n_fields, 2 + i, 0, rs[i].auc, NULL); + tab_double (tbl, n_cols - n_fields, 2 + i, 0, rs[i].auc, NULL, RC_OTHER); if ( roc->print_se ) { @@ -926,22 +1033,22 @@ show_auc (struct roc_state *rs, const struct cmd_roc *roc) tab_double (tbl, n_cols - 4, 2 + i, 0, se, - NULL); + NULL, RC_OTHER); ci = 1 - roc->ci / 100.0; yy = gsl_cdf_gaussian_Qinv (ci, se) ; tab_double (tbl, n_cols - 2, 2 + i, 0, rs[i].auc - yy, - NULL); + NULL, RC_OTHER); tab_double (tbl, n_cols - 1, 2 + i, 0, rs[i].auc + yy, - NULL); + NULL, RC_OTHER); tab_double (tbl, n_cols - 3, 2 + i, 0, 2.0 * gsl_cdf_ugaussian_Q (fabs ((rs[i].auc - 0.5 ) / sd_0_5)), - NULL); + NULL, RC_PVALUE); } } @@ -954,14 +1061,12 @@ show_summary (const struct cmd_roc *roc) { const int n_cols = 3; const int n_rows = 4; - struct tab_table *tbl = tab_create (n_cols, n_rows, 0); + struct tab_table *tbl = tab_create (n_cols, n_rows); tab_title (tbl, _("Case Summary")); tab_headers (tbl, 1, 0, 2, 0); - tab_dim (tbl, tab_natural_dimensions, NULL); - tab_box (tbl, TAL_2, TAL_2, -1, -1, @@ -990,11 +1095,11 @@ show_summary (const struct cmd_roc *roc) tab_text (tbl, 0, 3, TAB_LEFT, _("Negative")); - tab_double (tbl, 1, 2, 0, roc->pos, &F_8_0); - tab_double (tbl, 1, 3, 0, roc->neg, &F_8_0); + tab_double (tbl, 1, 2, 0, roc->pos, NULL, RC_INTEGER); + tab_double (tbl, 1, 3, 0, roc->neg, NULL, RC_INTEGER); - tab_double (tbl, 2, 2, 0, roc->pos_weighted, 0); - tab_double (tbl, 2, 3, 0, roc->neg_weighted, 0); + tab_double (tbl, 2, 2, 0, roc->pos_weighted, NULL, RC_OTHER); + tab_double (tbl, 2, 3, 0, roc->neg_weighted, NULL, RC_OTHER); tab_submit (tbl); } @@ -1012,7 +1117,7 @@ show_coords (struct roc_state *rs, const struct cmd_roc *roc) for (i = 0; i < roc->n_vars; ++i) n_rows += casereader_count_cases (rs[i].cutpoint_rdr); - tbl = tab_create (n_cols, n_rows, 0); + tbl = tab_create (n_cols, n_rows); if ( roc->n_vars > 1) tab_title (tbl, _("Coordinates of the Curve")); @@ -1022,8 +1127,6 @@ show_coords (struct roc_state *rs, const struct cmd_roc *roc) tab_headers (tbl, 1, 0, 1, 0); - tab_dim (tbl, tab_natural_dimensions, NULL); - tab_hline (tbl, TAL_2, 0, n_cols - 1, 1); if ( roc->n_vars > 1) @@ -1058,25 +1161,25 @@ show_coords (struct roc_state *rs, const struct cmd_roc *roc) for (; (cc = casereader_read (r)) != NULL; case_unref (cc), x++) { - const double se = case_data_idx (cc, TP)->f / + const double se = case_data_idx (cc, ROC_TP)->f / ( - case_data_idx (cc, TP)->f + case_data_idx (cc, ROC_TP)->f + - case_data_idx (cc, FN)->f + case_data_idx (cc, ROC_FN)->f ); - const double sp = case_data_idx (cc, TN)->f / + const double sp = case_data_idx (cc, ROC_TN)->f / ( - case_data_idx (cc, TN)->f + case_data_idx (cc, ROC_TN)->f + - case_data_idx (cc, FP)->f + case_data_idx (cc, ROC_FP)->f ); - tab_double (tbl, n_cols - 3, x, 0, case_data_idx (cc, CUTPOINT)->f, - var_get_print_format (roc->vars[i])); + tab_double (tbl, n_cols - 3, x, 0, case_data_idx (cc, ROC_CUTPOINT)->f, + var_get_print_format (roc->vars[i]), RC_OTHER); - tab_double (tbl, n_cols - 2, x, 0, se, NULL); - tab_double (tbl, n_cols - 1, x, 0, 1 - sp, NULL); + tab_double (tbl, n_cols - 2, x, 0, se, NULL, RC_OTHER); + tab_double (tbl, n_cols - 1, x, 0, 1 - sp, NULL, RC_OTHER); } casereader_destroy (r); @@ -1086,68 +1189,25 @@ show_coords (struct roc_state *rs, const struct cmd_roc *roc) } -static void -draw_roc (struct roc_state *rs, const struct cmd_roc *roc) -{ - int i; - - struct chart *roc_chart = chart_create (); - - chart_write_title (roc_chart, _("ROC Curve")); - chart_write_xlabel (roc_chart, _("1 - Specificity")); - chart_write_ylabel (roc_chart, _("Sensitivity")); - - chart_write_xscale (roc_chart, 0, 1, 5); - chart_write_yscale (roc_chart, 0, 1, 5); - - if ( roc->reference ) - { - chart_line (roc_chart, 1.0, 0, - 0.0, 1.0, - CHART_DIM_X); - } - - for (i = 0; i < roc->n_vars; ++i) - { - struct ccase *cc; - struct casereader *r = casereader_clone (rs[i].cutpoint_rdr); - - chart_vector_start (roc_chart, var_get_name (roc->vars[i])); - for (; (cc = casereader_read (r)) != NULL; - case_unref (cc)) - { - double se = case_data_idx (cc, TP)->f; - double sp = case_data_idx (cc, TN)->f; - - se /= case_data_idx (cc, FN)->f + - case_data_idx (cc, TP)->f ; - - sp /= case_data_idx (cc, TN)->f + - case_data_idx (cc, FP)->f ; - - chart_vector (roc_chart, 1 - sp, se); - } - chart_vector_end (roc_chart); - casereader_destroy (r); - } - - chart_write_legend (roc_chart); - - chart_submit (roc_chart); -} - - static void output_roc (struct roc_state *rs, const struct cmd_roc *roc) { show_summary (roc); if ( roc->curve ) - draw_roc (rs, roc); + { + struct roc_chart *rc; + size_t i; + + rc = roc_chart_create (roc->reference); + for (i = 0; i < roc->n_vars; i++) + roc_chart_add_var (rc, var_get_name (roc->vars[i]), + rs[i].cutpoint_rdr); + roc_chart_submit (rc); + } show_auc (rs, roc); - if ( roc->print_coords ) show_coords (rs, roc); }