/* 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
#include <config.h>
-#include <language/stats/roc.h>
-
-#include <data/procedure.h>
-#include <language/lexer/variable-parser.h>
-#include <language/lexer/value-parser.h>
-#include <language/command.h>
-#include <language/lexer/lexer.h>
-
-#include <data/casegrouper.h>
-#include <data/casereader.h>
-#include <data/casewriter.h>
-#include <data/dictionary.h>
-#include <data/format.h>
-#include <math/sort.h>
-#include <data/subcase.h>
-
-
-#include <libpspp/misc.h>
+#include "language/stats/roc.h"
#include <gsl/gsl_cdf.h>
-#include <output/table.h>
-#include <output/chart.h>
-#include <output/charts/roc-chart.h>
+#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)
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;
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))
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))
{
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))
{
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"))
{
}
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"))
}
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"))
{
}
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;
lex_error (lexer, NULL);
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;
lex_error (lexer, NULL);
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;
lex_error (lexer, NULL);
goto error;
}
- lex_force_match (lexer, ')');
+ if (!lex_force_match (lexer, T_RPAREN))
+ goto error;
}
else
{
}
}
- if ( ! 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;
error:
+ if ( roc.state_var)
+ value_destroy (&roc.state_value, roc.state_var_width);
free (roc.vars);
return CMD_FAILURE;
}
#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
+ As a side effect, this function also accumulates the roc->{pos,neg} and
roc->{pos,neg}_weighted counts.
*/
static bool
#define N_EQ 1
#define N_PRED 2
-/* Some intermediate state for calculating the cutpoints and the
+/* Some intermediate state for calculating the cutpoints and the
standard error values */
struct roc_state
{
double n2; /* total weight of negatives */
/* intermediates for standard error */
- double q1hat;
+ double q1hat;
double q2hat;
/* intermediates for cutpoints */
double max;
};
-/*
+/*
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
+ 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, ROC_CUTPOINT)->f;
casewriter_write (w, new_case);
}
- casereader_destroy (r);
+ casereader_destroy (input);
return casewriter_make_reader (w);
}
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 *
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)
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);
proto = caseproto_add_width (proto, 0);
proto = caseproto_add_width (proto, 0);
- wtr = autopaging_writer_create (proto);
+ wtr = autopaging_writer_create (proto);
*cc = 0;
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 postive cases with values greater than 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 */
}
-/*
+/*
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.
int i;
struct casereader *r = casereader_clone (input);
struct ccase *c;
- struct caseproto *proto = caseproto_create ();
- struct subcase ordering;
- subcase_init (&ordering, ROC_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); /* 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 */
+ 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))
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;
/* Separate the positive actual state cases from the negative ones */
- positives =
+ positives =
casereader_create_filter_func (input,
match_positives,
NULL,
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);
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)
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);
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
n_pos_gt = prev_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);
}
n_neg_lt = prev_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);
}
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_POS_EQ)->f;
}
+ 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 )
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);
}
tab_headers (tbl, n_cols - n_fields, 0, 1, 0);
- tab_dim (tbl, tab_natural_dimensions, NULL, NULL);
tab_text (tbl, n_cols - n_fields, 1, TAT_TITLE, _("Area"));
{
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 )
{
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);
}
}
tab_headers (tbl, 1, 0, 2, 0);
- tab_dim (tbl, tab_natural_dimensions, NULL, NULL);
-
tab_box (tbl,
TAL_2, TAL_2,
-1, -1,
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);
}
tab_headers (tbl, 1, 0, 1, 0);
- tab_dim (tbl, tab_natural_dimensions, NULL, NULL);
-
tab_hline (tbl, TAL_2, 0, n_cols - 1, 1);
if ( roc->n_vars > 1)
);
tab_double (tbl, n_cols - 3, x, 0, case_data_idx (cc, ROC_CUTPOINT)->f,
- var_get_print_format (roc->vars[i]));
+ 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);
for (i = 0; i < roc->n_vars; i++)
roc_chart_add_var (rc, var_get_name (roc->vars[i]),
rs[i].cutpoint_rdr);
- chart_submit (roc_chart_get_chart (rc));
+ roc_chart_submit (rc);
}
show_auc (rs, roc);