X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Flanguage%2Fstats%2Frank.c;h=ad113676cb86d39996271def3f4936a0062f33a1;hb=c8adc6783134063d4ada34a58f9e7abf6a039bb2;hp=a4f81856a995ab5603b57874563f5bc322e0b284;hpb=2beb35516f8749170f786022441a676347f1074d;p=pspp diff --git a/src/language/stats/rank.c b/src/language/stats/rank.c index a4f81856a9..ad113676cb 100644 --- a/src/language/stats/rank.c +++ b/src/language/stats/rank.c @@ -1,5 +1,5 @@ /* PSPP - a program for statistical analysis. - Copyright (C) 2005, 2006, 2007, 2009, 2010, 2011, 2012, 2013 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 @@ -41,8 +41,7 @@ #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) @@ -92,14 +91,14 @@ enum rank_func }; 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] = { @@ -124,6 +123,16 @@ static const rank_function_t rank_func[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 { @@ -144,8 +153,8 @@ enum fraction struct rank_spec { enum rank_func rfunc; - struct variable **destvars; const char **dest_names; + const char **dest_labels; }; /* If NEW_NAME exists in DICT or NEW_NAMES, returns NULL without changing @@ -162,7 +171,7 @@ try_new_name (const char *new_name, } /* Returns a variable name for storing ranks of a variable named SRC_NAME - accoring to the rank function F. The name chosen will not be one already in + according to the rank function F. The name chosen will not be one already in DICT or NEW_NAMES. If successful, adds the new name to NEW_NAMES and returns the name added. @@ -244,7 +253,7 @@ destroy_rank (struct rank *rank) { free (rank->vars); free (rank->group_vars); - subcase_destroy (&rank->sc); + subcase_uninit (&rank->sc); pool_destroy (rank->pool); } @@ -257,7 +266,7 @@ parse_into (struct lexer *lexer, struct rank *cmd, 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; @@ -288,22 +297,23 @@ parse_into (struct lexer *lexer, struct rank *cmd, } 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; } @@ -313,13 +323,13 @@ parse_into (struct lexer *lexer, struct rank *cmd, 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 ( var_count >= subcase_get_n_fields (&cmd->sc) ) - msg (SE, _("Too many variables in INTO clause.")); - else if ( dict_lookup_var (cmd->dict, name) != NULL ) + 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); @@ -354,7 +364,7 @@ rank_rank (const struct rank *cmd, double c, double cc, double cc_1, { double rank; - if ( c >= 1.0 ) + if (c >= 1.0) { switch (cmd->ties) { @@ -423,7 +433,7 @@ rank_proportion (const struct rank *cmd, double c, double cc, double cc_1, double f; - switch ( cmd->fraction ) + switch (cmd->fraction) { case FRAC_BLOM: f = (r - 3.0/8.0) / (w + 0.25); @@ -435,7 +445,7 @@ rank_proportion (const struct rank *cmd, double c, double cc, double cc_1, 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 (); @@ -461,7 +471,7 @@ rank_ntiles (const struct rank *cmd, double c, double cc, double cc_1, 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 */ @@ -472,7 +482,7 @@ ee (int j, double w_star) 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; } @@ -486,29 +496,29 @@ rank_savage (const struct rank *cmd UNUSED, double c, double cc, double cc_1, 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(); @@ -538,29 +548,24 @@ sum_weights (const struct casereader *input, int weight_idx) 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 variable *weight_var = dict_get_weight (dict); - int weight_idx = weight_var ? var_get_case_index (weight_var) : -1; struct casegrouper *tie_grouper; struct casereader *tied_cases; + struct subcase input_var; + int tie_group = 1; struct ccase *c; double cc = 0.0; double w; - 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); /* Get total group weight. */ w = sum_weights (input, weight_idx); /* Do ranking. */ - tie_grouper = casegrouper_create_vars (input, &cmd->vars[dest_idx], 1); + 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)) { @@ -572,46 +577,35 @@ rank_sorted_file (struct casereader *input, casewriter_get_taint (output)); /* Rank tied cases. */ - while ((c = casereader_read (tied_cases)) != NULL) + for (; (c = casereader_read (tied_cases)) != NULL; case_unref (c)) { + struct ccase *out_case; size_t i; - c = case_unshare (c); + 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); - } + 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) { - switch (cmd->fraction ) + switch (cmd->fraction) { case FRAC_BLOM: return "BLOM"; case FRAC_RANKIT: return "RANKIT"; @@ -631,16 +625,16 @@ create_var_label (struct rank *cmd, const struct variable *src_var, 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])); } @@ -653,7 +647,7 @@ create_var_label (struct rank *cmd, const struct variable *src_var, function_name[f], var_get_name (src_var)); pool_label = pool_strdup (cmd->pool, ds_cstr (&label)); - + ds_destroy (&label); return pool_label; @@ -664,10 +658,7 @@ cmd_rank (struct lexer *lexer, struct dataset *ds) { struct string_set new_names; struct rank rank; - struct variable *order; struct rank_spec *rs; - bool result = true; - int i; subcase_init_empty (&rank.sc); @@ -680,12 +671,14 @@ cmd_rank (struct lexer *lexer, struct dataset *ds) 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, @@ -694,21 +687,23 @@ cmd_rank (struct lexer *lexer, struct dataset *ds) 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; @@ -733,7 +728,8 @@ cmd_rank (struct lexer *lexer, struct dataset *ds) } 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; @@ -758,7 +754,8 @@ cmd_rank (struct lexer *lexer, struct dataset *ds) } 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; @@ -775,7 +772,8 @@ cmd_rank (struct lexer *lexer, struct dataset *ds) } 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; @@ -796,13 +794,14 @@ cmd_rank (struct lexer *lexer, struct dataset *ds) /* If no rank specs are given, then apply a default */ - if ( rank.n_rs == 0) + if (rank.n_rs == 0) { struct rank_spec *rs; rs = pool_calloc (rank.pool, 1, sizeof *rs); rs->rfunc = RANK; - rs->dest_names = pool_calloc (rank.pool, 1, sizeof *rs->dest_names); + rs->dest_names = pool_calloc (rank.pool, rank.n_vars, + sizeof *rs->dest_names); rank.rs = rs; rank.n_rs = 1; @@ -812,12 +811,12 @@ cmd_rank (struct lexer *lexer, struct dataset *ds) created with INTO. */ for (rs = rank.rs; rs < &rank.rs[rank.n_rs]; rs++) { - int v; - - for ( v = 0 ; v < rank.n_vars ; v ++ ) + 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 ) + if (*dst_name == NULL) { *dst_name = rank_choose_dest_name (rank.dict, &new_names, rs->rfunc, @@ -825,125 +824,66 @@ cmd_rank (struct lexer *lexer, struct dataset *ds) if (*dst_name == NULL) goto error; } - } - } - - /* Create variables. */ - for (rs = rank.rs; rs < &rank.rs[rank.n_rs]; rs++) - rs->destvars = pool_nmalloc (rank.pool, rank.n_vars, sizeof *rs->destvars); - for ( i = 0 ; i < rank.n_vars ; i ++ ) - { - struct variable *var; - - for (rs = rank.rs; rs < &rank.rs[rank.n_rs]; rs++) - { - var = dict_create_var_assert (dataset_dict (ds), - rs->dest_names[i], 0); - var_set_both_formats (var, &dest_format[rs->rfunc]); - var_set_label (var, create_var_label (&rank, rank.vars[i], - rs->rfunc), - false); - rs->destvars[i] = var; + 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]), - rank.rs[i].dest_names[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]), - rank.rs[i].dest_names[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]), - rank.rs[i].dest_names[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]), - rank.rs[i].dest_names[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); @@ -956,61 +896,247 @@ cmd_rank (struct lexer *lexer, struct dataset *ds) 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); - rank_sorted_file (by_group, output, d, i, cmd); + 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; - } - ok = casegrouper_destroy (by_grouper) && ok; + 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; + + 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; }