/* PSPP - a program for statistical analysis.
- Copyright (C) 2005, 2006, 2007, 2009, 2010, 2011, 2012 Free Software Foundation, Inc
+ Copyright (C) 2005, 2006, 2007, 2009, 2010, 2011, 2012, 2013, 2014, 2016 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 "language/stats/sort-criteria.h"
#include "math/sort.h"
#include "libpspp/assertion.h"
+#include "libpspp/i18n.h"
#include "libpspp/message.h"
#include "libpspp/misc.h"
#include "libpspp/pool.h"
+#include "libpspp/string-set.h"
#include "libpspp/taint.h"
-
-#include "output/tab.h"
+#include "output/pivot-table.h"
#include "gettext.h"
#define _(msgid) gettext (msgid)
};
static const struct fmt_spec dest_format[n_RANK_FUNCS] = {
- {FMT_F, 9, 3}, /* rank */
- {FMT_F, 6, 4}, /* normal */
- {FMT_F, 6, 2}, /* percent */
- {FMT_F, 6, 4}, /* rfraction */
- {FMT_F, 6, 4}, /* proportion */
- {FMT_F, 6, 0}, /* n */
- {FMT_F, 3, 0}, /* ntiles */
- {FMT_F, 8, 4} /* savage */
+ [RANK] = { .type = FMT_F, .w = 9, .d = 3 },
+ [NORMAL] = { .type = FMT_F, .w = 6, .d = 4 },
+ [PERCENT] = { .type = FMT_F, .w = 6, .d = 2 },
+ [RFRACTION] = { .type = FMT_F, .w = 6, .d = 4 },
+ [PROPORTION] = { .type = FMT_F, .w = 6, .d = 4 },
+ [N] = { .type = FMT_F, .w = 6, .d = 0 },
+ [NTILES] = { .type = FMT_F, .w = 3, .d = 0 },
+ [SAVAGE] = { .type = FMT_F, .w = 8, .d = 4 }
};
static const char * const function_name[n_RANK_FUNCS] = {
rank_savage
};
+static enum measure rank_measures[n_RANK_FUNCS] = {
+ [RANK] = MEASURE_ORDINAL,
+ [NORMAL] = MEASURE_ORDINAL,
+ [PERCENT] = MEASURE_ORDINAL,
+ [RFRACTION] = MEASURE_ORDINAL,
+ [PROPORTION] = MEASURE_ORDINAL,
+ [N] = MEASURE_SCALE,
+ [NTILES] = MEASURE_ORDINAL,
+ [SAVAGE] = MEASURE_ORDINAL,
+};
enum ties
{
struct rank_spec
{
enum rank_func rfunc;
- struct variable **destvars;
+ const char **dest_names;
+ const char **dest_labels;
};
-
-/* Create and return a new variable in which to store the ranks of SRC_VAR
- accoring to the rank function F.
- VNAME is the name of the variable to be created.
- If VNAME is NULL, then a name will be automatically chosen.
-*/
-static struct variable *
-create_rank_variable (struct dictionary *dict, enum rank_func f,
- const struct variable *src_var,
- const char *vname)
+/* If NEW_NAME exists in DICT or NEW_NAMES, returns NULL without changing
+ anything. Otherwise, inserts NEW_NAME in NEW_NAMES and returns the copy of
+ NEW_NAME now in NEW_NAMES. */
+static const char *
+try_new_name (const char *new_name,
+ const struct dictionary *dict, struct string_set *new_names)
{
- int i;
- struct variable *var = NULL;
- char name[SHORT_NAME_LEN + 1];
-
- if ( vname )
- var = dict_create_var(dict, vname, 0);
+ return (!dict_lookup_var (dict, new_name)
+ && string_set_insert (new_names, new_name)
+ ? string_set_find_node (new_names, new_name)->string
+ : NULL);
+}
- if ( NULL == var )
- {
- snprintf (name, SHORT_NAME_LEN + 1, "%c%s",
- function_name[f][0], var_get_name (src_var));
+/* Returns a variable name for storing ranks of a variable named SRC_NAME
+ according to the rank function F. The name chosen will not be one already in
+ DICT or NEW_NAMES.
- var = dict_create_var(dict, name, 0);
- }
- i = 1;
- while( NULL == var )
- {
- char func_abb[4];
- snprintf(func_abb, 4, "%s", function_name[f]);
- snprintf(name, SHORT_NAME_LEN + 1, "%s%03d", func_abb,
- i);
-
- var = dict_create_var(dict, name, 0);
- if (i++ >= 999)
- break;
- }
+ If successful, adds the new name to NEW_NAMES and returns the name added.
+ If no name can be generated, returns NULL. */
+static const char *
+rank_choose_dest_name (struct dictionary *dict, struct string_set *new_names,
+ enum rank_func f, const char *src_name)
+{
+ char *src_name_7;
+ char name[128];
+ const char *s;
+ int i;
- i = 1;
- while ( NULL == var )
+ /* Try the first character of the ranking function followed by the first 7
+ bytes of the srcinal variable name. */
+ src_name_7 = utf8_encoding_trunc (src_name, dict_get_encoding (dict), 7);
+ snprintf (name, sizeof name, "%c%s", function_name[f][0], src_name_7);
+ free (src_name_7);
+ s = try_new_name (name, dict, new_names);
+ if (s != NULL)
+ return s;
+
+ /* Try "fun###". */
+ for (i = 1; i <= 999; i++)
{
- char func_abb[3];
- snprintf(func_abb, 3, "%s", function_name[f]);
-
- snprintf(name, SHORT_NAME_LEN + 1,
- "RNK%s%02d", func_abb, i);
-
- var = dict_create_var(dict, name, 0);
- if ( i++ >= 99 )
- break;
+ sprintf (name, "%.3s%03d", function_name[f], i);
+ s = try_new_name (name, dict, new_names);
+ if (s != NULL)
+ return s;
}
- if ( NULL == var )
+ /* Try "RNKfn##". */
+ for (i = 1; i <= 99; i++)
{
- msg(ME, _("Cannot create new rank variable. All candidates in use."));
- return NULL;
+ sprintf (name, "RNK%.2s%02d", function_name[f], i);
+ s = try_new_name (name, dict, new_names);
+ if (s != NULL)
+ return s;
}
- var_set_both_formats (var, &dest_format[f]);
-
- return var;
+ msg (ME, _("Cannot generate variable name for ranking %s with %s. "
+ "All candidates in use."),
+ src_name, function_name[f]);
+ return NULL;
}
struct rank
{
free (rank->vars);
free (rank->group_vars);
- subcase_destroy (&rank->sc);
+ subcase_uninit (&rank->sc);
pool_destroy (rank->pool);
}
static bool
-parse_into (struct lexer *lexer, struct rank *cmd)
+parse_into (struct lexer *lexer, struct rank *cmd,
+ struct string_set *new_names)
{
int var_count = 0;
struct rank_spec *rs = NULL;
cmd->rs = pool_realloc (cmd->pool, cmd->rs, sizeof (*cmd->rs) * (cmd->n_rs + 1));
rs = &cmd->rs[cmd->n_rs];
-
+
if (lex_match_id (lexer, "RANK"))
{
rs->rfunc = RANK;
}
else if (lex_match_id (lexer, "NTILES"))
{
- if ( !lex_force_match (lexer, T_LPAREN))
+ if (!lex_force_match (lexer, T_LPAREN))
return false;
-
- if (! lex_force_int (lexer) )
+
+ if (! lex_force_int_range (lexer, "NTILES", 1, INT_MAX))
return false;
-
+
cmd->k_ntiles = lex_integer (lexer);
lex_get (lexer);
-
- if ( !lex_force_match (lexer, T_RPAREN))
+
+ if (!lex_force_match (lexer, T_RPAREN))
return false;
rs->rfunc = NTILES;
}
else
{
+ lex_error (lexer, NULL);
return false;
}
cmd->n_rs++;
- rs->destvars = NULL;
- rs->destvars = pool_calloc (cmd->pool, cmd->n_vars, sizeof (*rs->destvars));
+ rs->dest_names = pool_calloc (cmd->pool, cmd->n_vars,
+ sizeof *rs->dest_names);
if (lex_match_id (lexer, "INTO"))
{
- while( lex_token (lexer) == T_ID )
+ while(lex_token (lexer) == T_ID)
{
const char *name = lex_tokcstr (lexer);
- if ( dict_lookup_var (cmd->dict, name) != NULL )
- {
- msg (SE, _("Variable %s already exists."), name);
- return false;
- }
-
- if ( var_count >= subcase_get_n_fields (&cmd->sc) )
- {
- msg (SE, _("Too many variables in INTO clause."));
- return false;
- }
- rs->destvars[var_count] =
- create_rank_variable (cmd->dict, rs->rfunc, cmd->vars[var_count], name);
- ++var_count;
- lex_get (lexer);
- }
+
+ if (var_count >= subcase_get_n_fields (&cmd->sc))
+ msg (SE, _("Too many variables in %s clause."), "INTO");
+ else if (dict_lookup_var (cmd->dict, name) != NULL)
+ msg (SE, _("Variable %s already exists."), name);
+ else if (string_set_contains (new_names, name))
+ msg (SE, _("Duplicate variable name %s."), name);
+ else
+ {
+ string_set_insert (new_names, name);
+ rs->dest_names[var_count++] = pool_strdup (cmd->pool, name);
+ lex_get (lexer);
+ continue;
+ }
+
+ /* Error path. */
+ return false;
+ }
}
return true;
{
double rank;
- if ( c >= 1.0 )
+ if (c >= 1.0)
{
switch (cmd->ties)
{
double f;
- switch ( cmd->fraction )
+ switch (cmd->fraction)
{
case FRAC_BLOM:
f = (r - 3.0/8.0) / (w + 0.25);
f = (r - 1.0/3.0) / (w + 1.0/3.0);
break;
case FRAC_VW:
- f = r / ( w + 1.0);
+ f = r / (w + 1.0);
break;
default:
NOT_REACHED ();
double r = rank_rank (cmd, c, cc, cc_1, i, w);
- return ( floor (( r * cmd->k_ntiles) / ( w + 1) ) + 1);
+ return (floor ((r * cmd->k_ntiles) / (w + 1)) + 1);
}
/* Expected value of the order statistics from an exponential distribution */
double sum = 0.0;
for (k = 1 ; k <= j; k++)
- sum += 1.0 / ( w_star + 1 - k );
+ sum += 1.0 / (w_star + 1 - k);
return sum;
}
const int i_1 = floor (cc_1);
const int i_2 = floor (cc);
- const double w_star = (modf (w, &int_part) == 0 ) ? w : floor (w) + 1;
+ const double w_star = (modf (w, &int_part) == 0) ? w : floor (w) + 1;
const double g_1 = cc_1 - i_1;
const double g_2 = cc - i_2;
/* The second factor is infinite, when the first is zero.
Therefore, evaluate the second, only when the first is non-zero */
- const double expr1 = (1 - g_1) ? (1 - g_1) * ee(i_1+1, w_star) : ( 1 - g_1);
+ const double expr1 = (1 - g_1) ? (1 - g_1) * ee(i_1+1, w_star) : (1 - g_1);
const double expr2 = g_2 ? g_2 * ee (i_2+1, w_star) : g_2 ;
- if ( i_1 == i_2 )
+ if (i_1 == i_2)
return ee (i_1 + 1, w_star) - 1;
- if ( i_1 + 1 == i_2 )
- return ( ( expr1 + expr2 )/c ) - 1;
+ if (i_1 + 1 == i_2)
+ return ((expr1 + expr2)/c) - 1;
- if ( i_1 + 2 <= i_2 )
+ if (i_1 + 2 <= i_2)
{
int j;
double sigma = 0.0;
- for (j = i_1 + 2 ; j <= i_2; ++j )
+ for (j = i_1 + 2 ; j <= i_2; ++j)
sigma += ee (j, w_star);
- return ( (expr1 + expr2 + sigma) / c) -1;
+ return ((expr1 + expr2 + sigma) / c) -1;
}
NOT_REACHED();
}
+static double
+sum_weights (const struct casereader *input, int weight_idx)
+{
+ if (weight_idx == -1)
+ return casereader_count_cases (input);
+ else
+ {
+ struct casereader *pass;
+ struct ccase *c;
+ double w;
+
+ w = 0.0;
+ pass = casereader_clone (input);
+ for (; (c = casereader_read (pass)) != NULL; case_unref (c))
+ w += case_num_idx (c, weight_idx);
+ casereader_destroy (pass);
+
+ return w;
+ }
+}
static void
rank_sorted_file (struct casereader *input,
struct casewriter *output,
- const struct dictionary *dict,
- int dest_idx,
- const struct rank *cmd
- )
+ int weight_idx,
+ const struct rank *cmd)
{
- struct casereader *pass1, *pass2, *pass2_1;
struct casegrouper *tie_grouper;
+ struct casereader *tied_cases;
+ struct subcase input_var;
+ int tie_group = 1;
struct ccase *c;
- double w = 0.0;
double cc = 0.0;
- int tie_group = 1;
-
- input = casereader_create_filter_missing (input, &cmd->vars[dest_idx], 1,
- cmd->exclude, NULL, output);
- input = casereader_create_filter_weight (input, dict, NULL, output);
-
- casereader_split (input, &pass1, &pass2);
+ double w;
- /* Pass 1: Get total group weight. */
- for (; (c = casereader_read (pass1)) != NULL; case_unref (c))
- w += dict_get_case_weight (dict, c, NULL);
- casereader_destroy (pass1);
+ /* Get total group weight. */
+ w = sum_weights (input, weight_idx);
- /* Pass 2: Do ranking. */
- tie_grouper = casegrouper_create_vars (pass2, &cmd->vars[dest_idx], 1);
- while (casegrouper_get_next_group (tie_grouper, &pass2_1))
+ /* Do ranking. */
+ subcase_init (&input_var, 0, 0, SC_ASCEND);
+ tie_grouper = casegrouper_create_subcase (input, &input_var);
+ subcase_uninit (&input_var);
+ for (; casegrouper_get_next_group (tie_grouper, &tied_cases);
+ casereader_destroy (tied_cases))
{
- struct casereader *pass2_2;
+ double tw = sum_weights (tied_cases, weight_idx);
double cc_1 = cc;
- double tw = 0.0;
- int i;
+ cc += tw;
- pass2_2 = casereader_clone (pass2_1);
- taint_propagate (casereader_get_taint (pass2_2),
+ taint_propagate (casereader_get_taint (tied_cases),
casewriter_get_taint (output));
- /* Pass 2.1: Sum up weight for tied cases. */
- for (; (c = casereader_read (pass2_1)) != NULL; case_unref (c))
- tw += dict_get_case_weight (dict, c, NULL);
- cc += tw;
- casereader_destroy (pass2_1);
-
- /* Pass 2.2: Rank tied cases. */
- while ((c = casereader_read (pass2_2)) != NULL)
+ /* Rank tied cases. */
+ for (; (c = casereader_read (tied_cases)) != NULL; case_unref (c))
{
- c = case_unshare (c);
+ struct ccase *out_case;
+ size_t i;
+
+ out_case = case_create (casewriter_get_proto (output));
+ *case_num_rw_idx (out_case, 0) = case_num_idx (c, 1);
for (i = 0; i < cmd->n_rs; ++i)
{
- const struct variable *dst_var = cmd->rs[i].destvars[dest_idx];
- double *dst_value = &case_data_rw (c, dst_var)->f;
- *dst_value = rank_func[cmd->rs[i].rfunc] (cmd, tw, cc, cc_1, tie_group, w);
+ rank_function_t func = rank_func[cmd->rs[i].rfunc];
+ double rank = func (cmd, tw, cc, cc_1, tie_group, w);
+ *case_num_rw_idx (out_case, i + 1) = rank;
}
- casewriter_write (output, c);
- }
- casereader_destroy (pass2_2);
+ casewriter_write (output, out_case);
+ }
tie_group++;
}
casegrouper_destroy (tie_grouper);
}
-/* Transformation function to enumerate all the cases */
-static int
-create_resort_key (void *key_var_, struct ccase **cc, casenumber case_num)
-{
- struct variable *key_var = key_var_;
-
- *cc = case_unshare (*cc);
- case_data_rw (*cc, key_var)->f = case_num;
-
- return TRNS_CONTINUE;
-}
-
static bool
rank_cmd (struct dataset *ds, const struct rank *cmd);
-
static const char *
fraction_name (const struct rank *cmd)
{
- static char name[10];
- switch (cmd->fraction )
+ switch (cmd->fraction)
{
- case FRAC_BLOM:
- strcpy (name, "BLOM");
- break;
- case FRAC_RANKIT:
- strcpy (name, "RANKIT");
- break;
- case FRAC_TUKEY:
- strcpy (name, "TUKEY");
- break;
- case FRAC_VW:
- strcpy (name, "VW");
- break;
- default:
- NOT_REACHED ();
+ case FRAC_BLOM: return "BLOM";
+ case FRAC_RANKIT: return "RANKIT";
+ case FRAC_TUKEY: return "TUKEY";
+ case FRAC_VW: return "VW";
+ default: NOT_REACHED ();
}
- return name;
}
-/* Create a label on DEST_VAR, describing its derivation from SRC_VAR and F */
-static void
-create_var_label (struct rank *cmd, struct variable *dest_var,
- const struct variable *src_var, enum rank_func f)
+/* Returns a label for a variable derived from SRC_VAR with function F. */
+static const char *
+create_var_label (struct rank *cmd, const struct variable *src_var,
+ enum rank_func f)
{
struct string label;
+ const char *pool_label;
+
ds_init_empty (&label);
- if ( cmd->n_group_vars > 0 )
+ if (cmd->n_group_vars > 0)
{
struct string group_var_str;
int g;
ds_init_empty (&group_var_str);
- for (g = 0 ; g < cmd->n_group_vars ; ++g )
+ for (g = 0 ; g < cmd->n_group_vars ; ++g)
{
- if ( g > 0 ) ds_put_cstr (&group_var_str, " ");
+ if (g > 0) ds_put_cstr (&group_var_str, " ");
ds_put_cstr (&group_var_str, var_get_name (cmd->group_vars[g]));
}
ds_put_format (&label, _("%s of %s"),
function_name[f], var_get_name (src_var));
- var_set_label (dest_var, ds_cstr (&label), false);
-
+ pool_label = pool_strdup (cmd->pool, ds_cstr (&label));
+
ds_destroy (&label);
+
+ return pool_label;
}
int
cmd_rank (struct lexer *lexer, struct dataset *ds)
{
+ struct string_set new_names;
struct rank rank;
- struct variable *order;
- bool result = true;
- int i;
+ struct rank_spec *rs;
subcase_init_empty (&rank.sc);
rank.ties = TIES_MEAN;
rank.fraction = FRAC_BLOM;
rank.print = true;
+ rank.vars = NULL;
rank.pool = pool_create ();
+ string_set_init (&new_names);
+
if (lex_match_id (lexer, "VARIABLES"))
- lex_force_match (lexer, T_EQUALS);
+ if (! lex_force_match (lexer, T_EQUALS))
+ goto error;
if (!parse_sort_criteria (lexer, rank.dict,
&rank.sc,
rank.n_vars = rank.sc.n_fields;
- if (lex_match (lexer, T_BY) )
+ if (lex_match (lexer, T_BY))
{
- if ( ! parse_variables_const (lexer, rank.dict,
+ if (! parse_variables_const (lexer, rank.dict,
&rank.group_vars, &rank.n_group_vars,
PV_NO_DUPLICATE | PV_NO_SCRATCH))
goto error;
}
- while (lex_token (lexer) != T_ENDCMD )
+ while (lex_token (lexer) != T_ENDCMD)
{
- lex_force_match (lexer, T_SLASH);
+ if (! lex_force_match (lexer, T_SLASH))
+ goto error;
if (lex_match_id (lexer, "TIES"))
{
- lex_force_match (lexer, T_EQUALS);
+ if (! lex_force_match (lexer, T_EQUALS))
+ goto error;
if (lex_match_id (lexer, "MEAN"))
{
rank.ties = TIES_MEAN;
}
else if (lex_match_id (lexer, "FRACTION"))
{
- lex_force_match (lexer, T_EQUALS);
+ if (! lex_force_match (lexer, T_EQUALS))
+ goto error;
if (lex_match_id (lexer, "BLOM"))
{
rank.fraction = FRAC_BLOM;
}
else if (lex_match_id (lexer, "PRINT"))
{
- lex_force_match (lexer, T_EQUALS);
+ if (! lex_force_match (lexer, T_EQUALS))
+ goto error;
if (lex_match_id (lexer, "YES"))
{
rank.print = true;
}
else if (lex_match_id (lexer, "MISSING"))
{
- lex_force_match (lexer, T_EQUALS);
+ if (! lex_force_match (lexer, T_EQUALS))
+ goto error;
if (lex_match_id (lexer, "INCLUDE"))
{
rank.exclude = MV_SYSTEM;
goto error;
}
}
- else if (! parse_into (lexer, &rank))
+ else if (! parse_into (lexer, &rank, &new_names))
goto error;
}
/* If no rank specs are given, then apply a default */
- if ( rank.n_rs == 0)
+ if (rank.n_rs == 0)
{
- rank.rs = pool_calloc (rank.pool, 1, sizeof (*rank.rs));
+ struct rank_spec *rs;
+
+ rs = pool_calloc (rank.pool, 1, sizeof *rs);
+ rs->rfunc = RANK;
+ rs->dest_names = pool_calloc (rank.pool, rank.n_vars,
+ sizeof *rs->dest_names);
+
+ rank.rs = rs;
rank.n_rs = 1;
- rank.rs[0].rfunc = RANK;
- rank.rs[0].destvars = pool_calloc (rank.pool, rank.n_vars, sizeof (*rank.rs[0].destvars));
}
- /* Create variables for all rank destinations which haven't
- already been created with INTO.
- Add labels to all the destination variables.
- */
- for (i = 0 ; i < rank.n_rs ; ++i )
+ /* Choose variable names for all rank destinations which haven't already been
+ created with INTO. */
+ for (rs = rank.rs; rs < &rank.rs[rank.n_rs]; rs++)
{
- int v;
- struct rank_spec *rs = &rank.rs[i];
-
- for ( v = 0 ; v < rank.n_vars ; v ++ )
- {
- if ( rs->destvars[v] == NULL )
- {
- rs->destvars[v] =
- create_rank_variable (rank.dict, rs->rfunc, rank.vars[v], NULL);
- }
+ rs->dest_labels = pool_calloc (rank.pool, rank.n_vars,
+ sizeof *rs->dest_labels);
+ for (int v = 0 ; v < rank.n_vars ; v ++)
+ {
+ const char **dst_name = &rs->dest_names[v];
+ if (*dst_name == NULL)
+ {
+ *dst_name = rank_choose_dest_name (rank.dict, &new_names,
+ rs->rfunc,
+ var_get_name (rank.vars[v]));
+ if (*dst_name == NULL)
+ goto error;
+ }
- create_var_label (&rank, rs->destvars[v],
- rank.vars[v],
- rs->rfunc);
- }
+ rs->dest_labels[v] = create_var_label (&rank, rank.vars[v],
+ rs->rfunc);
+ }
}
- if ( rank.print )
+ if (rank.print)
{
- int v;
+ struct pivot_table *table = pivot_table_create (
+ N_("Variables Created by RANK"));
+
+ pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("New Variable"),
+ N_("New Variable"), N_("Function"),
+ N_("Fraction"), N_("Grouping Variables"));
- tab_output_text (0, _("Variables Created By RANK"));
- tab_output_text (0, "");
+ struct pivot_dimension *variables = pivot_dimension_create (
+ table, PIVOT_AXIS_ROW, N_("Existing Variable"),
+ N_("Existing Variable"));
+ variables->root->show_label = true;
- for (i = 0 ; i < rank.n_rs ; ++i )
+ for (size_t i = 0 ; i < rank.n_rs ; ++i)
{
- for ( v = 0 ; v < rank.n_vars ; v ++ )
+ for (size_t v = 0 ; v < rank.n_vars ; v ++)
{
- if ( rank.n_group_vars > 0 )
- {
- struct string varlist;
- int g;
-
- ds_init_empty (&varlist);
- for ( g = 0 ; g < rank.n_group_vars ; ++g )
- {
- ds_put_cstr (&varlist, var_get_name (rank.group_vars[g]));
-
- if ( g < rank.n_group_vars - 1)
- ds_put_cstr (&varlist, " ");
- }
-
- if ( rank.rs[i].rfunc == NORMAL ||
- rank.rs[i].rfunc == PROPORTION )
- tab_output_text_format (0,
- _("%s into %s(%s of %s using %s BY %s)"),
- var_get_name (rank.vars[v]),
- var_get_name (rank.rs[i].destvars[v]),
- function_name[rank.rs[i].rfunc],
- var_get_name (rank.vars[v]),
- fraction_name (&rank),
- ds_cstr (&varlist));
-
- else
- tab_output_text_format (0,
- _("%s into %s(%s of %s BY %s)"),
- var_get_name (rank.vars[v]),
- var_get_name (rank.rs[i].destvars[v]),
- function_name[rank.rs[i].rfunc],
- var_get_name (rank.vars[v]),
- ds_cstr (&varlist));
- ds_destroy (&varlist);
- }
- else
- {
- if ( rank.rs[i].rfunc == NORMAL ||
- rank.rs[i].rfunc == PROPORTION )
- tab_output_text_format (0,
- _("%s into %s(%s of %s using %s)"),
- var_get_name (rank.vars[v]),
- var_get_name (rank.rs[i].destvars[v]),
- function_name[rank.rs[i].rfunc],
- var_get_name (rank.vars[v]),
- fraction_name (&rank));
-
- else
- tab_output_text_format (0,
- _("%s into %s(%s of %s)"),
- var_get_name (rank.vars[v]),
- var_get_name (rank.rs[i].destvars[v]),
- function_name[rank.rs[i].rfunc],
- var_get_name (rank.vars[v]));
- }
+ int row_idx = pivot_category_create_leaf (
+ variables->root, pivot_value_new_variable (rank.vars[v]));
+
+ struct string group_vars = DS_EMPTY_INITIALIZER;
+ for (int g = 0 ; g < rank.n_group_vars ; ++g)
+ {
+ if (g)
+ ds_put_byte (&group_vars, ' ');
+ ds_put_cstr (&group_vars, var_get_name (rank.group_vars[g]));
+ }
+
+ enum rank_func rfunc = rank.rs[i].rfunc;
+ bool has_fraction = rfunc == NORMAL || rfunc == PROPORTION;
+ const char *entries[] =
+ {
+ rank.rs[i].dest_names[v],
+ function_name[rank.rs[i].rfunc],
+ has_fraction ? fraction_name (&rank) : NULL,
+ rank.n_group_vars ? ds_cstr (&group_vars) : NULL,
+ };
+ for (size_t j = 0; j < sizeof entries / sizeof *entries; j++)
+ {
+ const char *entry = entries[j];
+ if (entry)
+ pivot_table_put2 (table, j, row_idx,
+ pivot_value_new_user_text (entry, -1));
+ }
+ ds_destroy (&group_vars);
}
}
- }
-
- /* Add a variable which we can sort by to get back the original
- order */
- order = dict_create_var_assert (dataset_dict (ds), "$ORDER_", 0);
- add_transformation (ds, create_resort_key, 0, order);
+ pivot_table_submit (table);
+ }
/* Do the ranking */
- result = rank_cmd (ds, &rank);
-
- /* Put the active dataset back in its original order. Delete
- our sort key, which we don't need anymore. */
- {
- struct casereader *sorted;
-
-
- /* FIXME: loses error conditions. */
-
- proc_discard_output (ds);
- sorted = sort_execute_1var (proc_open (ds), order);
- result = proc_commit (ds) && result;
-
- dict_delete_var (dataset_dict (ds), order);
- result = dataset_set_source (ds, sorted) && result;
- if ( result != true)
- goto error;
- }
+ rank_cmd (ds, &rank);
destroy_rank (&rank);
+ string_set_destroy (&new_names);
return CMD_SUCCESS;
error:
destroy_rank (&rank);
+ string_set_destroy (&new_names);
return CMD_FAILURE;
}
+/* RANK transformation. */
+struct rank_trns
+ {
+ int order_case_idx;
+
+ struct rank_trns_input_var *input_vars;
+ size_t n_input_vars;
+
+ size_t n_funcs;
+ };
+
+struct rank_trns_input_var
+ {
+ struct casereader *input;
+ struct ccase *current;
+ struct variable **output_vars;
+ };
+
+static void
+advance_ranking (struct rank_trns_input_var *iv)
+{
+ case_unref (iv->current);
+ iv->current = casereader_read (iv->input);
+}
+
+static enum trns_result
+rank_trns_proc (void *trns_, struct ccase **c, casenumber case_idx UNUSED)
+{
+ struct rank_trns *trns = trns_;
+ double order = case_num_idx (*c, trns->order_case_idx);
+ struct rank_trns_input_var *iv;
+
+ *c = case_unshare (*c);
+ for (iv = trns->input_vars; iv < &trns->input_vars[trns->n_input_vars]; iv++)
+ while (iv->current != NULL)
+ {
+ double iv_order = case_num_idx (iv->current, 0);
+ if (iv_order == order)
+ {
+ size_t i;
+
+ for (i = 0; i < trns->n_funcs; i++)
+ *case_num_rw (*c, iv->output_vars[i])
+ = case_num_idx (iv->current, i + 1);
+ advance_ranking (iv);
+ break;
+ }
+ else if (iv_order > order)
+ break;
+ else
+ advance_ranking (iv);
+ }
+ return TRNS_CONTINUE;
+}
+
+static bool
+rank_trns_free (void *trns_)
+{
+ struct rank_trns *trns = trns_;
+ struct rank_trns_input_var *iv;
+
+ for (iv = trns->input_vars; iv < &trns->input_vars[trns->n_input_vars]; iv++)
+ {
+ casereader_destroy (iv->input);
+ case_unref (iv->current);
+
+ free (iv->output_vars);
+ }
+ free (trns->input_vars);
+ free (trns);
+
+ return true;
+}
+
+static const struct trns_class rank_trns_class = {
+ .name = "RANK",
+ .execute = rank_trns_proc,
+ .destroy = rank_trns_free,
+};
static bool
rank_cmd (struct dataset *ds, const struct rank *cmd)
{
struct dictionary *d = dataset_dict (ds);
+ struct variable *weight_var = dict_get_weight (d);
+ struct casewriter **outputs;
+ struct variable *order_var;
+ struct casereader *input;
+ struct rank_trns *trns;
bool ok = true;
int i;
- for (i = 0 ; i < subcase_get_n_fields (&cmd->sc) ; ++i )
+ order_var = add_permanent_ordering_transformation (ds);
+
+ /* Create output files. */
+ {
+ struct caseproto *output_proto;
+ struct subcase by_order;
+
+ output_proto = caseproto_create ();
+ for (i = 0; i < cmd->n_rs + 1; i++)
+ output_proto = caseproto_add_width (output_proto, 0);
+
+ subcase_init (&by_order, 0, 0, SC_ASCEND);
+
+ outputs = xnmalloc (cmd->n_vars, sizeof *outputs);
+ for (i = 0; i < cmd->n_vars; i++)
+ outputs[i] = sort_create_writer (&by_order, output_proto);
+
+ subcase_uninit (&by_order);
+ caseproto_unref (output_proto);
+ }
+
+ /* Open the active file and make one pass per input variable. */
+ input = proc_open (ds);
+ input = casereader_create_filter_weight (input, d, NULL, NULL);
+ for (i = 0 ; i < cmd->n_vars ; ++i)
{
- /* Rank variable at index I in SC. */
+ const struct variable *input_var = cmd->vars[i];
+ struct casereader *input_pass;
struct casegrouper *split_grouper;
struct casereader *split_group;
- struct casewriter *output;
-
- proc_discard_output (ds);
- split_grouper = casegrouper_create_splits (proc_open (ds), d);
- output = autopaging_writer_create (dict_get_proto (d));
+ struct subcase rank_ordering;
+ struct subcase projection;
+ struct subcase split_vars;
+ struct subcase group_vars;
+ int weight_idx;
+ int j;
+ /* Discard cases that have missing values of input variable. */
+ input_pass = i == cmd->n_vars - 1 ? input : casereader_clone (input);
+ input_pass = casereader_create_filter_missing (input_pass, &input_var, 1,
+ cmd->exclude, NULL, NULL);
+
+ /* Keep only the columns we really need, to save time and space when we
+ sort them just below.
+
+ After this projection, the input_pass case indexes look like:
+
+ - 0: input_var.
+ - 1: order_var.
+ - 2 and up: cmd->n_group_vars group variables
+ - 2 + cmd->n_group_vars and up: split variables
+ - 2 + cmd->n_group_vars + n_split_vars: weight var
+ */
+ subcase_init_empty (&projection);
+ subcase_add_var_always (&projection, input_var, SC_ASCEND);
+ subcase_add_var_always (&projection, order_var, SC_ASCEND);
+ subcase_add_vars_always (&projection,
+ cmd->group_vars, cmd->n_group_vars);
+ subcase_add_vars_always (&projection, dict_get_split_vars (d),
+ dict_get_n_splits (d));
+ if (weight_var != NULL)
+ {
+ subcase_add_var_always (&projection, weight_var, SC_ASCEND);
+ weight_idx = 2 + cmd->n_group_vars + dict_get_n_splits (d);
+ }
+ else
+ weight_idx = -1;
+ input_pass = casereader_project (input_pass, &projection);
+ subcase_uninit (&projection);
+
+ /* Prepare 'group_vars' as the set of grouping variables. */
+ subcase_init_empty (&group_vars);
+ for (j = 0; j < cmd->n_group_vars; j++)
+ subcase_add_always (&group_vars,
+ j + 2, var_get_width (cmd->group_vars[j]),
+ SC_ASCEND);
+
+ /* Prepare 'rank_ordering' for sorting with the group variables as
+ primary key and the input variable as secondary key. */
+ subcase_clone (&rank_ordering, &group_vars);
+ subcase_add (&rank_ordering, 0, 0, subcase_get_direction (&cmd->sc, i));
+
+ /* Group by split variables */
+ subcase_init_empty (&split_vars);
+ for (j = 0; j < dict_get_n_splits (d); j++)
+ subcase_add_always (&split_vars, 2 + j + cmd->n_group_vars,
+ var_get_width (dict_get_split_vars (d)[j]),
+ SC_ASCEND);
+ split_grouper = casegrouper_create_subcase (input_pass, &split_vars);
+ subcase_uninit (&split_vars);
while (casegrouper_get_next_group (split_grouper, &split_group))
{
- struct subcase ordering;
struct casereader *ordered;
struct casegrouper *by_grouper;
struct casereader *by_group;
- /* Sort this split group by the BY variables as primary
- keys and the rank variable as secondary key. */
- subcase_init_vars (&ordering, cmd->group_vars, cmd->n_group_vars);
- subcase_add_var (&ordering, cmd->vars[i],
- subcase_get_direction (&cmd->sc, i));
- ordered = sort_execute (split_group, &ordering);
- subcase_destroy (&ordering);
-
- /* Rank the rank variable within this split group. */
- by_grouper = casegrouper_create_vars (ordered,
- cmd->group_vars, cmd->n_group_vars);
+ ordered = sort_execute (split_group, &rank_ordering);
+ by_grouper = casegrouper_create_subcase (ordered, &group_vars);
while (casegrouper_get_next_group (by_grouper, &by_group))
- {
- /* Rank the rank variable within this BY group
- within the split group. */
+ rank_sorted_file (by_group, outputs[i], weight_idx, cmd);
+ ok = casegrouper_destroy (by_grouper) && ok;
+ }
+ subcase_uninit (&group_vars);
+ subcase_uninit (&rank_ordering);
+
+ ok = casegrouper_destroy (split_grouper) && ok;
+ }
+ ok = proc_commit (ds) && ok;
+
+ /* Re-fetch the dictionary and order variable, because if TEMPORARY was in
+ effect then there's a new dictionary. */
+ d = dataset_dict (ds);
+ order_var = dict_lookup_var_assert (d, "$ORDER");
+
+ /* Merge the original data set with the ranks (which we already sorted on
+ $ORDER). */
+ trns = xmalloc (sizeof *trns);
+ trns->order_case_idx = var_get_case_index (order_var);
+ trns->input_vars = xnmalloc (cmd->n_vars, sizeof *trns->input_vars);
+ trns->n_input_vars = cmd->n_vars;
+ trns->n_funcs = cmd->n_rs;
+ for (i = 0; i < trns->n_input_vars; i++)
+ {
+ struct rank_trns_input_var *iv = &trns->input_vars[i];
+ int j;
- rank_sorted_file (by_group, output, d, i, cmd);
+ iv->input = casewriter_make_reader (outputs[i]);
+ iv->current = casereader_read (iv->input);
+ iv->output_vars = xnmalloc (trns->n_funcs, sizeof *iv->output_vars);
+ for (j = 0; j < trns->n_funcs; j++)
+ {
+ struct rank_spec *rs = &cmd->rs[j];
+ struct variable *var;
- }
- ok = casegrouper_destroy (by_grouper) && ok;
+ var = dict_create_var_assert (d, rs->dest_names[i], 0);
+ var_set_both_formats (var, &dest_format[rs->rfunc]);
+ var_set_label (var, rs->dest_labels[i]);
+ var_set_measure (var, rank_measures[rs->rfunc]);
+
+ iv->output_vars[j] = var;
}
- ok = casegrouper_destroy (split_grouper);
- ok = proc_commit (ds) && ok;
- ok = (dataset_set_source (ds, casewriter_make_reader (output))
- && ok);
- if (!ok)
- break;
}
+ free (outputs);
+
+ add_transformation (ds, &rank_trns_class, trns);
+
+ /* Delete our sort key, which we don't need anymore. */
+ dict_delete_var (d, order_var);
return ok;
}
projects / pspp / blobdiff