1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 2005, 2006, 2007, 2009, 2010, 2011, 2012, 2013 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 <gsl/gsl_cdf.h>
22 #include "data/case.h"
23 #include "data/casegrouper.h"
24 #include "data/casereader.h"
25 #include "data/dataset.h"
26 #include "data/dictionary.h"
27 #include "data/format.h"
28 #include "data/variable.h"
29 #include "data/subcase.h"
30 #include "data/casewriter.h"
31 #include "data/short-names.h"
32 #include "language/command.h"
33 #include "language/lexer/lexer.h"
34 #include "language/lexer/variable-parser.h"
35 #include "language/stats/sort-criteria.h"
36 #include "math/sort.h"
37 #include "libpspp/assertion.h"
38 #include "libpspp/i18n.h"
39 #include "libpspp/message.h"
40 #include "libpspp/misc.h"
41 #include "libpspp/pool.h"
42 #include "libpspp/string-set.h"
43 #include "libpspp/taint.h"
45 #include "output/tab.h"
48 #define _(msgid) gettext (msgid)
49 #define N_(msgid) (msgid)
53 typedef double (*rank_function_t) (const struct rank*, double c, double cc, double cc_1,
56 static double rank_proportion (const struct rank *, double c, double cc, double cc_1,
59 static double rank_normal (const struct rank *, double c, double cc, double cc_1,
62 static double rank_percent (const struct rank *, double c, double cc, double cc_1,
65 static double rank_rfraction (const struct rank *, double c, double cc, double cc_1,
68 static double rank_rank (const struct rank *, double c, double cc, double cc_1,
71 static double rank_n (const struct rank *, double c, double cc, double cc_1,
74 static double rank_savage (const struct rank *, double c, double cc, double cc_1,
77 static double rank_ntiles (const struct rank *, double c, double cc, double cc_1,
94 static const struct fmt_spec dest_format[n_RANK_FUNCS] = {
95 {FMT_F, 9, 3}, /* rank */
96 {FMT_F, 6, 4}, /* normal */
97 {FMT_F, 6, 2}, /* percent */
98 {FMT_F, 6, 4}, /* rfraction */
99 {FMT_F, 6, 4}, /* proportion */
100 {FMT_F, 6, 0}, /* n */
101 {FMT_F, 3, 0}, /* ntiles */
102 {FMT_F, 8, 4} /* savage */
105 static const char * const function_name[n_RANK_FUNCS] = {
116 static const rank_function_t rank_func[n_RANK_FUNCS] = {
146 enum rank_func rfunc;
147 const char **dest_names;
148 const char **dest_labels;
151 /* If NEW_NAME exists in DICT or NEW_NAMES, returns NULL without changing
152 anything. Otherwise, inserts NEW_NAME in NEW_NAMES and returns the copy of
153 NEW_NAME now in NEW_NAMES. */
155 try_new_name (const char *new_name,
156 const struct dictionary *dict, struct string_set *new_names)
158 return (!dict_lookup_var (dict, new_name)
159 && string_set_insert (new_names, new_name)
160 ? string_set_find_node (new_names, new_name)->string
164 /* Returns a variable name for storing ranks of a variable named SRC_NAME
165 accoring to the rank function F. The name chosen will not be one already in
168 If successful, adds the new name to NEW_NAMES and returns the name added.
169 If no name can be generated, returns NULL. */
171 rank_choose_dest_name (struct dictionary *dict, struct string_set *new_names,
172 enum rank_func f, const char *src_name)
179 /* Try the first character of the ranking function followed by the first 7
180 bytes of the srcinal variable name. */
181 src_name_7 = utf8_encoding_trunc (src_name, dict_get_encoding (dict), 7);
182 snprintf (name, sizeof name, "%c%s", function_name[f][0], src_name_7);
184 s = try_new_name (name, dict, new_names);
189 for (i = 1; i <= 999; i++)
191 sprintf (name, "%.3s%03d", function_name[f], i);
192 s = try_new_name (name, dict, new_names);
198 for (i = 1; i <= 99; i++)
200 sprintf (name, "RNK%.2s%02d", function_name[f], i);
201 s = try_new_name (name, dict, new_names);
206 msg (ME, _("Cannot generate variable name for ranking %s with %s. "
207 "All candidates in use."),
208 src_name, function_name[f]);
214 struct dictionary *dict;
218 const struct variable **vars;
221 const struct variable **group_vars;
225 enum mv_class exclude;
227 struct rank_spec *rs;
232 enum fraction fraction;
237 /* Pool on which cell functions may allocate data */
243 destroy_rank (struct rank *rank)
246 free (rank->group_vars);
247 subcase_destroy (&rank->sc);
248 pool_destroy (rank->pool);
252 parse_into (struct lexer *lexer, struct rank *cmd,
253 struct string_set *new_names)
256 struct rank_spec *rs = NULL;
258 cmd->rs = pool_realloc (cmd->pool, cmd->rs, sizeof (*cmd->rs) * (cmd->n_rs + 1));
259 rs = &cmd->rs[cmd->n_rs];
261 if (lex_match_id (lexer, "RANK"))
265 else if (lex_match_id (lexer, "NORMAL"))
269 else if (lex_match_id (lexer, "RFRACTION"))
271 rs->rfunc = RFRACTION;
273 else if (lex_match_id (lexer, "N"))
277 else if (lex_match_id (lexer, "SAVAGE"))
281 else if (lex_match_id (lexer, "PERCENT"))
285 else if (lex_match_id (lexer, "PROPORTION"))
287 rs->rfunc = PROPORTION;
289 else if (lex_match_id (lexer, "NTILES"))
291 if ( !lex_force_match (lexer, T_LPAREN))
294 if (! lex_force_int (lexer) )
297 cmd->k_ntiles = lex_integer (lexer);
300 if ( !lex_force_match (lexer, T_RPAREN))
311 rs->dest_names = pool_calloc (cmd->pool, cmd->n_vars,
312 sizeof *rs->dest_names);
314 if (lex_match_id (lexer, "INTO"))
316 while( lex_token (lexer) == T_ID )
318 const char *name = lex_tokcstr (lexer);
320 if ( var_count >= subcase_get_n_fields (&cmd->sc) )
321 msg (SE, _("Too many variables in INTO clause."));
322 else if ( dict_lookup_var (cmd->dict, name) != NULL )
323 msg (SE, _("Variable %s already exists."), name);
324 else if (string_set_contains (new_names, name))
325 msg (SE, _("Duplicate variable name %s."), name);
328 string_set_insert (new_names, name);
329 rs->dest_names[var_count++] = pool_strdup (cmd->pool, name);
342 /* Hardly a rank function !! */
344 rank_n (const struct rank *cmd UNUSED, double c UNUSED, double cc UNUSED, double cc_1 UNUSED,
345 int i UNUSED, double w)
352 rank_rank (const struct rank *cmd, double c, double cc, double cc_1,
353 int i, double w UNUSED)
368 rank = cc_1 + (c + 1.0)/ 2.0;
388 rank = cc_1 + c / 2.0 ;
403 rank_rfraction (const struct rank *cmd, double c, double cc, double cc_1,
406 return rank_rank (cmd, c, cc, cc_1, i, w) / w ;
411 rank_percent (const struct rank *cmd, double c, double cc, double cc_1,
414 return rank_rank (cmd, c, cc, cc_1, i, w) * 100.0 / w ;
419 rank_proportion (const struct rank *cmd, double c, double cc, double cc_1,
422 const double r = rank_rank (cmd, c, cc, cc_1, i, w) ;
426 switch ( cmd->fraction )
429 f = (r - 3.0/8.0) / (w + 0.25);
435 f = (r - 1.0/3.0) / (w + 1.0/3.0);
445 return (f > 0) ? f : SYSMIS;
449 rank_normal (const struct rank *cmd, double c, double cc, double cc_1,
452 double f = rank_proportion (cmd, c, cc, cc_1, i, w);
454 return gsl_cdf_ugaussian_Pinv (f);
458 rank_ntiles (const struct rank *cmd, double c, double cc, double cc_1,
461 double r = rank_rank (cmd, c, cc, cc_1, i, w);
464 return ( floor (( r * cmd->k_ntiles) / ( w + 1) ) + 1);
467 /* Expected value of the order statistics from an exponential distribution */
469 ee (int j, double w_star)
474 for (k = 1 ; k <= j; k++)
475 sum += 1.0 / ( w_star + 1 - k );
482 rank_savage (const struct rank *cmd UNUSED, double c, double cc, double cc_1,
483 int i UNUSED, double w)
486 const int i_1 = floor (cc_1);
487 const int i_2 = floor (cc);
489 const double w_star = (modf (w, &int_part) == 0 ) ? w : floor (w) + 1;
491 const double g_1 = cc_1 - i_1;
492 const double g_2 = cc - i_2;
494 /* The second factor is infinite, when the first is zero.
495 Therefore, evaluate the second, only when the first is non-zero */
496 const double expr1 = (1 - g_1) ? (1 - g_1) * ee(i_1+1, w_star) : ( 1 - g_1);
497 const double expr2 = g_2 ? g_2 * ee (i_2+1, w_star) : g_2 ;
500 return ee (i_1 + 1, w_star) - 1;
502 if ( i_1 + 1 == i_2 )
503 return ( ( expr1 + expr2 )/c ) - 1;
505 if ( i_1 + 2 <= i_2 )
509 for (j = i_1 + 2 ; j <= i_2; ++j )
510 sigma += ee (j, w_star);
511 return ( (expr1 + expr2 + sigma) / c) -1;
518 sum_weights (const struct casereader *input, int weight_idx)
520 if (weight_idx == -1)
521 return casereader_count_cases (input);
524 struct casereader *pass;
529 pass = casereader_clone (input);
530 for (; (c = casereader_read (pass)) != NULL; case_unref (c))
531 w += case_num_idx (c, weight_idx);
532 casereader_destroy (pass);
539 rank_sorted_file (struct casereader *input,
540 struct casewriter *output,
542 const struct rank *cmd)
544 struct casegrouper *tie_grouper;
545 struct casereader *tied_cases;
546 struct subcase input_var;
552 /* Get total group weight. */
553 w = sum_weights (input, weight_idx);
556 subcase_init (&input_var, 0, 0, SC_ASCEND);
557 tie_grouper = casegrouper_create_subcase (input, &input_var);
558 subcase_destroy (&input_var);
559 for (; casegrouper_get_next_group (tie_grouper, &tied_cases);
560 casereader_destroy (tied_cases))
562 double tw = sum_weights (tied_cases, weight_idx);
566 taint_propagate (casereader_get_taint (tied_cases),
567 casewriter_get_taint (output));
569 /* Rank tied cases. */
570 for (; (c = casereader_read (tied_cases)) != NULL; case_unref (c))
572 struct ccase *out_case;
575 out_case = case_create (casewriter_get_proto (output));
576 case_data_rw_idx (out_case, 0)->f = case_num_idx (c, 1);
577 for (i = 0; i < cmd->n_rs; ++i)
579 rank_function_t func = rank_func[cmd->rs[i].rfunc];
580 double rank = func (cmd, tw, cc, cc_1, tie_group, w);
581 case_data_rw_idx (out_case, i + 1)->f = rank;
584 casewriter_write (output, out_case);
588 casegrouper_destroy (tie_grouper);
592 /* Transformation function to enumerate all the cases */
594 create_resort_key (void *key_var_, struct ccase **cc, casenumber case_num)
596 struct variable *key_var = key_var_;
598 *cc = case_unshare (*cc);
599 case_data_rw (*cc, key_var)->f = case_num;
601 return TRNS_CONTINUE;
605 rank_cmd (struct dataset *ds, const struct rank *cmd);
609 fraction_name (const struct rank *cmd)
611 switch (cmd->fraction )
613 case FRAC_BLOM: return "BLOM";
614 case FRAC_RANKIT: return "RANKIT";
615 case FRAC_TUKEY: return "TUKEY";
616 case FRAC_VW: return "VW";
617 default: NOT_REACHED ();
621 /* Returns a label for a variable derived from SRC_VAR with function F. */
623 create_var_label (struct rank *cmd, const struct variable *src_var,
627 const char *pool_label;
629 ds_init_empty (&label);
631 if ( cmd->n_group_vars > 0 )
633 struct string group_var_str;
636 ds_init_empty (&group_var_str);
638 for (g = 0 ; g < cmd->n_group_vars ; ++g )
640 if ( g > 0 ) ds_put_cstr (&group_var_str, " ");
641 ds_put_cstr (&group_var_str, var_get_name (cmd->group_vars[g]));
644 ds_put_format (&label, _("%s of %s by %s"), function_name[f],
645 var_get_name (src_var), ds_cstr (&group_var_str));
646 ds_destroy (&group_var_str);
649 ds_put_format (&label, _("%s of %s"),
650 function_name[f], var_get_name (src_var));
652 pool_label = pool_strdup (cmd->pool, ds_cstr (&label));
660 cmd_rank (struct lexer *lexer, struct dataset *ds)
662 struct string_set new_names;
664 struct rank_spec *rs;
667 subcase_init_empty (&rank.sc);
671 rank.exclude = MV_ANY;
672 rank.n_group_vars = 0;
673 rank.group_vars = NULL;
674 rank.dict = dataset_dict (ds);
675 rank.ties = TIES_MEAN;
676 rank.fraction = FRAC_BLOM;
678 rank.pool = pool_create ();
680 string_set_init (&new_names);
682 if (lex_match_id (lexer, "VARIABLES"))
683 lex_force_match (lexer, T_EQUALS);
685 if (!parse_sort_criteria (lexer, rank.dict,
690 rank.n_vars = rank.sc.n_fields;
692 if (lex_match (lexer, T_BY) )
694 if ( ! parse_variables_const (lexer, rank.dict,
695 &rank.group_vars, &rank.n_group_vars,
696 PV_NO_DUPLICATE | PV_NO_SCRATCH))
701 while (lex_token (lexer) != T_ENDCMD )
703 lex_force_match (lexer, T_SLASH);
704 if (lex_match_id (lexer, "TIES"))
706 lex_force_match (lexer, T_EQUALS);
707 if (lex_match_id (lexer, "MEAN"))
709 rank.ties = TIES_MEAN;
711 else if (lex_match_id (lexer, "LOW"))
713 rank.ties = TIES_LOW;
715 else if (lex_match_id (lexer, "HIGH"))
717 rank.ties = TIES_HIGH;
719 else if (lex_match_id (lexer, "CONDENSE"))
721 rank.ties = TIES_CONDENSE;
725 lex_error (lexer, NULL);
729 else if (lex_match_id (lexer, "FRACTION"))
731 lex_force_match (lexer, T_EQUALS);
732 if (lex_match_id (lexer, "BLOM"))
734 rank.fraction = FRAC_BLOM;
736 else if (lex_match_id (lexer, "TUKEY"))
738 rank.fraction = FRAC_TUKEY;
740 else if (lex_match_id (lexer, "VW"))
742 rank.fraction = FRAC_VW;
744 else if (lex_match_id (lexer, "RANKIT"))
746 rank.fraction = FRAC_RANKIT;
750 lex_error (lexer, NULL);
754 else if (lex_match_id (lexer, "PRINT"))
756 lex_force_match (lexer, T_EQUALS);
757 if (lex_match_id (lexer, "YES"))
761 else if (lex_match_id (lexer, "NO"))
767 lex_error (lexer, NULL);
771 else if (lex_match_id (lexer, "MISSING"))
773 lex_force_match (lexer, T_EQUALS);
774 if (lex_match_id (lexer, "INCLUDE"))
776 rank.exclude = MV_SYSTEM;
778 else if (lex_match_id (lexer, "EXCLUDE"))
780 rank.exclude = MV_ANY;
784 lex_error (lexer, NULL);
788 else if (! parse_into (lexer, &rank, &new_names))
793 /* If no rank specs are given, then apply a default */
796 struct rank_spec *rs;
798 rs = pool_calloc (rank.pool, 1, sizeof *rs);
800 rs->dest_names = pool_calloc (rank.pool, 1, sizeof *rs->dest_names);
801 rs->dest_labels = pool_calloc (rank.pool, 1, sizeof *rs->dest_labels);
807 /* Choose variable names for all rank destinations which haven't already been
808 created with INTO. */
809 for (rs = rank.rs; rs < &rank.rs[rank.n_rs]; rs++)
813 rs->dest_labels = pool_calloc (rank.pool, rank.n_vars,
814 sizeof *rs->dest_labels);
815 for ( v = 0 ; v < rank.n_vars ; v ++ )
817 const char **dst_name = &rs->dest_names[v];
818 if ( *dst_name == NULL )
820 *dst_name = rank_choose_dest_name (rank.dict, &new_names,
822 var_get_name (rank.vars[v]));
823 if (*dst_name == NULL)
827 rs->dest_labels[v] = create_var_label (&rank, rank.vars[v],
836 tab_output_text (0, _("Variables Created By RANK"));
837 tab_output_text (0, "");
839 for (i = 0 ; i < rank.n_rs ; ++i )
841 for ( v = 0 ; v < rank.n_vars ; v ++ )
843 if ( rank.n_group_vars > 0 )
845 struct string varlist;
848 ds_init_empty (&varlist);
849 for ( g = 0 ; g < rank.n_group_vars ; ++g )
851 ds_put_cstr (&varlist, var_get_name (rank.group_vars[g]));
853 if ( g < rank.n_group_vars - 1)
854 ds_put_cstr (&varlist, " ");
857 if ( rank.rs[i].rfunc == NORMAL ||
858 rank.rs[i].rfunc == PROPORTION )
859 tab_output_text_format (0,
860 _("%s into %s(%s of %s using %s BY %s)"),
861 var_get_name (rank.vars[v]),
862 rank.rs[i].dest_names[v],
863 function_name[rank.rs[i].rfunc],
864 var_get_name (rank.vars[v]),
865 fraction_name (&rank),
869 tab_output_text_format (0,
870 _("%s into %s(%s of %s BY %s)"),
871 var_get_name (rank.vars[v]),
872 rank.rs[i].dest_names[v],
873 function_name[rank.rs[i].rfunc],
874 var_get_name (rank.vars[v]),
876 ds_destroy (&varlist);
880 if ( rank.rs[i].rfunc == NORMAL ||
881 rank.rs[i].rfunc == PROPORTION )
882 tab_output_text_format (0,
883 _("%s into %s(%s of %s using %s)"),
884 var_get_name (rank.vars[v]),
885 rank.rs[i].dest_names[v],
886 function_name[rank.rs[i].rfunc],
887 var_get_name (rank.vars[v]),
888 fraction_name (&rank));
891 tab_output_text_format (0,
892 _("%s into %s(%s of %s)"),
893 var_get_name (rank.vars[v]),
894 rank.rs[i].dest_names[v],
895 function_name[rank.rs[i].rfunc],
896 var_get_name (rank.vars[v]));
903 rank_cmd (ds, &rank);
905 destroy_rank (&rank);
906 string_set_destroy (&new_names);
911 destroy_rank (&rank);
912 string_set_destroy (&new_names);
916 /* RANK transformation. */
921 struct rank_trns_input_var *input_vars;
927 struct rank_trns_input_var
929 struct casereader *input;
930 struct ccase *current;
932 struct variable **output_vars;
936 advance_ranking (struct rank_trns_input_var *iv)
938 case_unref (iv->current);
939 iv->current = casereader_read (iv->input);
943 rank_trns_proc (void *trns_, struct ccase **c, casenumber case_idx UNUSED)
945 struct rank_trns *trns = trns_;
946 double order = case_num_idx (*c, trns->order_case_idx);
947 struct rank_trns_input_var *iv;
949 *c = case_unshare (*c);
950 for (iv = trns->input_vars; iv < &trns->input_vars[trns->n_input_vars]; iv++)
951 while (iv->current != NULL)
953 double iv_order = case_num_idx (iv->current, 0);
954 if (iv_order == order)
958 for (i = 0; i < trns->n_funcs; i++)
959 case_data_rw (*c, iv->output_vars[i])->f
960 = case_num_idx (iv->current, i + 1);
961 advance_ranking (iv);
964 else if (iv_order > order)
967 advance_ranking (iv);
969 return TRNS_CONTINUE;
973 rank_trns_free (void *trns_)
975 struct rank_trns *trns = trns_;
976 struct rank_trns_input_var *iv;
978 for (iv = trns->input_vars; iv < &trns->input_vars[trns->n_input_vars]; iv++)
980 casereader_destroy (iv->input);
981 case_unref (iv->current);
983 free (iv->output_vars);
985 free (trns->input_vars);
992 rank_cmd (struct dataset *ds, const struct rank *cmd)
994 struct dictionary *d = dataset_dict (ds);
995 struct variable *weight_var = dict_get_weight (d);
996 struct casewriter **outputs;
997 struct variable *order_var;
998 struct casereader *input;
999 struct rank_trns *trns;
1003 /* Add a variable which we can sort by to get back the original
1005 order_var = dict_create_var_assert (dataset_dict (ds), "$ORDER", 0);
1007 add_transformation (ds, create_resort_key, 0, order_var);
1009 /* Create output files. */
1011 struct caseproto *output_proto;
1012 struct subcase by_order;
1014 output_proto = caseproto_create ();
1015 for (i = 0; i < cmd->n_rs + 1; i++)
1016 output_proto = caseproto_add_width (output_proto, 0);
1018 subcase_init (&by_order, 0, 0, SC_ASCEND);
1020 outputs = xnmalloc (cmd->n_vars, sizeof *outputs);
1021 for (i = 0; i < cmd->n_vars; i++)
1022 outputs[i] = sort_create_writer (&by_order, output_proto);
1024 subcase_destroy (&by_order);
1025 caseproto_unref (output_proto);
1028 /* Open the active file and make one pass per input variable. */
1029 input = proc_open (ds);
1030 input = casereader_create_filter_weight (input, d, NULL, NULL);
1031 for (i = 0 ; i < cmd->n_vars ; ++i )
1033 const struct variable *input_var = cmd->vars[i];
1034 struct casereader *input_pass;
1035 struct casegrouper *split_grouper;
1036 struct casereader *split_group;
1037 struct subcase rank_ordering;
1038 struct subcase projection;
1039 struct subcase split_vars;
1040 struct subcase group_vars;
1044 /* Discard cases that have missing values of input variable. */
1045 input_pass = i == cmd->n_vars - 1 ? input : casereader_clone (input);
1046 input_pass = casereader_create_filter_missing (input_pass, &input_var, 1,
1047 cmd->exclude, NULL, NULL);
1049 /* Keep only the columns we really need, to save time and space when we
1050 sort them just below.
1052 After this projection, the input_pass case indexes look like:
1056 - 2 and up: cmd->n_group_vars group variables
1057 - 2 + cmd->n_group_vars and up: split variables
1058 - 2 + cmd->n_group_vars + n_split_vars: weight var
1060 subcase_init_empty (&projection);
1061 subcase_add_var_always (&projection, input_var, SC_ASCEND);
1062 subcase_add_var_always (&projection, order_var, SC_ASCEND);
1063 subcase_add_vars_always (&projection,
1064 cmd->group_vars, cmd->n_group_vars);
1065 subcase_add_vars_always (&projection, dict_get_split_vars (d),
1066 dict_get_split_cnt (d));
1067 if (weight_var != NULL)
1069 subcase_add_var_always (&projection, weight_var, SC_ASCEND);
1070 weight_idx = 2 + cmd->n_group_vars + dict_get_split_cnt (d);
1074 input_pass = casereader_project (input_pass, &projection);
1075 subcase_destroy (&projection);
1077 /* Prepare 'group_vars' as the set of grouping variables. */
1078 subcase_init_empty (&group_vars);
1079 for (j = 0; j < cmd->n_group_vars; j++)
1080 subcase_add_always (&group_vars,
1081 j + 2, var_get_width (cmd->group_vars[j]),
1084 /* Prepare 'rank_ordering' for sorting with the group variables as
1085 primary key and the input variable as secondary key. */
1086 subcase_clone (&rank_ordering, &group_vars);
1087 subcase_add (&rank_ordering, 0, 0, subcase_get_direction (&cmd->sc, i));
1089 /* Group by split variables */
1090 subcase_init_empty (&split_vars);
1091 for (j = 0; j < dict_get_split_cnt (d); j++)
1092 subcase_add_always (&split_vars, 2 + j + cmd->n_group_vars,
1093 var_get_width (dict_get_split_vars (d)[j]),
1095 split_grouper = casegrouper_create_subcase (input_pass, &split_vars);
1096 subcase_destroy (&split_vars);
1097 while (casegrouper_get_next_group (split_grouper, &split_group))
1099 struct casereader *ordered;
1100 struct casegrouper *by_grouper;
1101 struct casereader *by_group;
1103 ordered = sort_execute (split_group, &rank_ordering);
1104 by_grouper = casegrouper_create_subcase (ordered, &group_vars);
1105 while (casegrouper_get_next_group (by_grouper, &by_group))
1106 rank_sorted_file (by_group, outputs[i], weight_idx, cmd);
1107 ok = casegrouper_destroy (by_grouper) && ok;
1109 subcase_destroy (&group_vars);
1110 subcase_destroy (&rank_ordering);
1112 ok = casegrouper_destroy (split_grouper) && ok;
1114 ok = proc_commit (ds) && ok;
1116 /* Re-fetch the dictionary and order variable, because if TEMPORARY was in
1117 effect then there's a new dictionary. */
1118 d = dataset_dict (ds);
1119 order_var = dict_lookup_var_assert (d, "$ORDER");
1121 /* Merge the original data set with the ranks (which we already sorted on
1123 trns = xmalloc (sizeof *trns);
1124 trns->order_case_idx = var_get_case_index (order_var);
1125 trns->input_vars = xnmalloc (cmd->n_vars, sizeof *trns->input_vars);
1126 trns->n_input_vars = cmd->n_vars;
1127 trns->n_funcs = cmd->n_rs;
1128 for (i = 0; i < trns->n_input_vars; i++)
1130 struct rank_trns_input_var *iv = &trns->input_vars[i];
1133 iv->input = casewriter_make_reader (outputs[i]);
1134 iv->current = casereader_read (iv->input);
1135 iv->output_vars = xnmalloc (trns->n_funcs, sizeof *iv->output_vars);
1136 for (j = 0; j < trns->n_funcs; j++)
1138 struct rank_spec *rs = &cmd->rs[j];
1139 struct variable *var;
1141 var = dict_create_var_assert (d, rs->dest_names[i], 0);
1142 var_set_both_formats (var, &dest_format[rs->rfunc]);
1143 var_set_label (var, rs->dest_labels[i], false);
1145 iv->output_vars[j] = var;
1150 add_transformation (ds, rank_trns_proc, rank_trns_free, trns);
1152 /* Delete our sort key, which we don't need anymore. */
1153 dict_delete_var (d, order_var);