X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Flanguage%2Fstats%2Foneway.c;h=8e2805b3b1d54ca618b91d2b23889d10a84cdd75;hb=71eea20b080f51f1aa00ef35acf4f49ce742d10a;hp=d08e3114c04d8155d6fe035000ba42745f05dc07;hpb=456454e38aae190277eabaaadc601ccf3ed97af3;p=pspp-builds.git diff --git a/src/language/stats/oneway.c b/src/language/stats/oneway.c index d08e3114..8e2805b3 100644 --- a/src/language/stats/oneway.c +++ b/src/language/stats/oneway.c @@ -1,5 +1,5 @@ /* PSPP - a program for statistical analysis. - Copyright (C) 1997-9, 2000, 2007, 2009, 2010 Free Software Foundation, Inc. + Copyright (C) 1997-9, 2000, 2007, 2009, 2010, 2011 Free Software Foundation, Inc. This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by @@ -16,45 +16,37 @@ #include -#include -#include -#include - -#include -#include -#include -#include -#include - -#include - -#include -#include -#include -#include - -#include -#include -#include - -#include -#include -#include -#include -#include -#include - -#include - #include +#include #include -#include -#include +#include "data/case.h" +#include "data/casegrouper.h" +#include "data/casereader.h" +#include "data/dataset.h" +#include "data/dictionary.h" +#include "data/format.h" +#include "data/value.h" +#include "language/command.h" +#include "language/dictionary/split-file.h" +#include "language/lexer/lexer.h" +#include "language/lexer/value-parser.h" +#include "language/lexer/variable-parser.h" +#include "libpspp/ll.h" +#include "libpspp/message.h" +#include "libpspp/misc.h" +#include "libpspp/taint.h" +#include "linreg/sweep.h" +#include "math/categoricals.h" +#include "math/covariance.h" +#include "math/levene.h" +#include "math/moments.h" +#include "output/tab.h" #include "gettext.h" #define _(msgid) gettext (msgid) + enum missing_type { MISS_LISTWISE, @@ -99,7 +91,6 @@ struct oneway_spec /* The weight variable */ const struct variable *wv; - }; /* Per category data */ @@ -115,7 +106,9 @@ struct descriptive_data /* Workspace variable for each dependent variable */ struct per_var_ws { + struct categoricals *cat; struct covariance *cov; + struct levene *nl; double sst; double sse; @@ -132,10 +125,6 @@ struct oneway_workspace missing values are disregarded */ int actual_number_of_groups; - /* A hash table containing all the distinct values of the independent - variable */ - struct hsh_table *group_hash; - struct per_var_ws *vws; /* An array of descriptive data. One for each dependent variable */ @@ -166,13 +155,13 @@ cmd_oneway (struct lexer *lexer, struct dataset *ds) ll_init (&oneway.contrast_list); - if ( lex_match (lexer, '/')) + if ( lex_match (lexer, T_SLASH)) { if (!lex_force_match_id (lexer, "VARIABLES")) { goto error; } - lex_match (lexer, '='); + lex_match (lexer, T_EQUALS); } if (!parse_variables_const (lexer, dict, @@ -184,14 +173,14 @@ cmd_oneway (struct lexer *lexer, struct dataset *ds) oneway.indep_var = parse_variable_const (lexer, dict); - while (lex_token (lexer) != '.') + while (lex_token (lexer) != T_ENDCMD) { - lex_match (lexer, '/'); + lex_match (lexer, T_SLASH); if (lex_match_id (lexer, "STATISTICS")) { - 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, "DESCRIPTIVES")) { @@ -213,11 +202,11 @@ cmd_oneway (struct lexer *lexer, struct dataset *ds) struct contrasts_node *cl = xzalloc (sizeof *cl); struct ll_list *coefficient_list = &cl->coefficient_list; - lex_match (lexer, '='); + lex_match (lexer, T_EQUALS); ll_init (coefficient_list); - while (lex_token (lexer) != '.' && lex_token (lexer) != '/') + while (lex_token (lexer) != T_ENDCMD && lex_token (lexer) != T_SLASH) { if ( lex_is_number (lexer)) { @@ -238,8 +227,8 @@ cmd_oneway (struct lexer *lexer, struct dataset *ds) } else 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")) { @@ -295,32 +284,6 @@ cmd_oneway (struct lexer *lexer, struct dataset *ds) -static int -compare_double_3way (const void *a_, const void *b_, const void *aux UNUSED) -{ - const double *a = a_; - const double *b = b_; - return *a < *b ? -1 : *a > *b; -} - -static unsigned -do_hash_double (const void *value_, const void *aux UNUSED) -{ - const double *value = value_; - return hash_double (*value, 0); -} - -static void -free_double (void *value_, const void *aux UNUSED) -{ - double *value = value_; - free (value); -} - - - -static void postcalc (const struct oneway_spec *cmd); -static void precalc (const struct oneway_spec *cmd); static struct descriptive_data * dd_create (const struct variable *var) @@ -335,6 +298,12 @@ dd_create (const struct variable *var) return dd; } +static void +dd_destroy (struct descriptive_data *dd) +{ + moments1_destroy (dd->mom); + free (dd); +} static void * makeit (void *aux1, void *aux2 UNUSED) @@ -360,21 +329,21 @@ updateit (void *user_data, const union value *valx = case_data (c, varp); + struct descriptive_data *dd_total = aux2; + + double weight; + if ( var_is_value_missing (varp, valx, exclude)) return; - struct descriptive_data *dd_total = aux2; - - double weight = 1.0; - if (wv) - weight = case_data (c, wv)->f; + weight = wv != NULL ? case_data (c, wv)->f : 1.0; moments1_add (dd->mom, valx->f, weight); - if (valx->f * weight < dd->minimum) - dd->minimum = valx->f * weight; + if (valx->f < dd->minimum) + dd->minimum = valx->f; - if (valx->f * weight > dd->maximum) - dd->maximum = valx->f * weight; + if (valx->f > dd->maximum) + dd->maximum = valx->f; { const struct variable *var = dd_total->var; @@ -384,11 +353,11 @@ updateit (void *user_data, val->f, weight); - if (val->f * weight < dd_total->minimum) - dd_total->minimum = val->f * weight; + if (val->f < dd_total->minimum) + dd_total->minimum = val->f; - if (val->f * weight > dd_total->maximum) - dd_total->maximum = val->f * weight; + if (val->f > dd_total->maximum) + dd_total->maximum = val->f; } } @@ -406,7 +375,7 @@ run_oneway (const struct oneway_spec *cmd, struct oneway_workspace ws; ws.actual_number_of_groups = 0; - ws.vws = xmalloc (cmd->n_vars * sizeof (*ws.vws)); + ws.vws = xzalloc (cmd->n_vars * sizeof (*ws.vws)); ws.dd_total = xmalloc (sizeof (struct descriptive_data) * cmd->n_vars); for (v = 0 ; v < cmd->n_vars; ++v) @@ -414,36 +383,29 @@ run_oneway (const struct oneway_spec *cmd, for (v = 0; v < cmd->n_vars; ++v) { - struct categoricals *cats = categoricals_create (&cmd->indep_var, 1, - cmd->wv, cmd->exclude, - makeit, - updateit, - cmd->vars[v], ws.dd_total[v]); + ws.vws[v].cat = categoricals_create (&cmd->indep_var, 1, cmd->wv, + cmd->exclude, makeit, updateit, + CONST_CAST (struct variable *, + cmd->vars[v]), + ws.dd_total[v]); ws.vws[v].cov = covariance_2pass_create (1, &cmd->vars[v], - cats, + ws.vws[v].cat, cmd->wv, cmd->exclude); + ws.vws[v].nl = levene_create (var_get_width (cmd->indep_var), NULL); } c = casereader_peek (input, 0); if (c == NULL) { casereader_destroy (input); - return; + goto finish; } output_split_file_values (ds, c); case_unref (c); taint = taint_clone (casereader_get_taint (input)); - ws.group_hash = hsh_create (4, - compare_double_3way, - do_hash_double, - free_double, - cmd->indep_var); - - precalc (cmd); - input = casereader_create_filter_missing (input, &cmd->indep_var, 1, cmd->exclude, NULL, NULL); if (cmd->missing_type == MISS_LISTWISE) @@ -452,23 +414,14 @@ run_oneway (const struct oneway_spec *cmd, input = casereader_create_filter_weight (input, dict, NULL, NULL); reader = casereader_clone (input); - for (; (c = casereader_read (reader)) != NULL; case_unref (c)) { - size_t i; - - const double weight = dict_get_case_weight (dict, c, NULL); - - const union value *indep_val = case_data (c, cmd->indep_var); - void **p = hsh_probe (ws.group_hash, &indep_val->f); - if (*p == NULL) - { - double *value = *p = xmalloc (sizeof *value); - *value = indep_val->f; - } + int i; + double w = dict_get_case_weight (dict, c, NULL); for (i = 0; i < cmd->n_vars; ++i) { + struct per_var_ws *pvw = &ws.vws[i]; const struct variable *v = cmd->vars[i]; const union value *val = case_data (c, v); @@ -478,67 +431,41 @@ run_oneway (const struct oneway_spec *cmd, continue; } - { - struct per_var_ws *pvw = &ws.vws[i]; - - covariance_accumulate_pass1 (pvw->cov, c); - } - - struct group_proc *gp = group_proc_get (cmd->vars[i]); - struct hsh_table *group_hash = gp->group_hash; - - struct group_statistics *gs; - - gs = hsh_find (group_hash, indep_val ); + covariance_accumulate_pass1 (pvw->cov, c); + levene_pass_one (pvw->nl, val->f, w, case_data (c, cmd->indep_var)); + } + } + casereader_destroy (reader); - if ( ! gs ) - { - gs = xmalloc (sizeof *gs); - gs->id = *indep_val; - gs->sum = 0; - gs->n = 0; - gs->ssq = 0; - gs->sum_diff = 0; - gs->minimum = DBL_MAX; - gs->maximum = -DBL_MAX; - - hsh_insert ( group_hash, gs ); - } + reader = casereader_clone (input); + for ( ; (c = casereader_read (reader) ); case_unref (c)) + { + int i; + double w = dict_get_case_weight (dict, c, NULL); + for (i = 0; i < cmd->n_vars; ++i) + { + struct per_var_ws *pvw = &ws.vws[i]; + const struct variable *v = cmd->vars[i]; + const union value *val = case_data (c, v); - if (!var_is_value_missing (v, val, cmd->exclude)) + if ( MISS_ANALYSIS == cmd->missing_type) { - struct group_statistics *totals = &gp->ugs; - - totals->n += weight; - totals->sum += weight * val->f; - totals->ssq += weight * pow2 (val->f); - - if ( val->f * weight < totals->minimum ) - totals->minimum = val->f * weight; - - if ( val->f * weight > totals->maximum ) - totals->maximum = val->f * weight; - - gs->n += weight; - gs->sum += weight * val->f; - gs->ssq += weight * pow2 (val->f); - - if ( val->f * weight < gs->minimum ) - gs->minimum = val->f * weight; - - if ( val->f * weight > gs->maximum ) - gs->maximum = val->f * weight; + if ( var_is_value_missing (v, val, cmd->exclude)) + continue; } - gp->n_groups = hsh_count (group_hash ); + covariance_accumulate_pass2 (pvw->cov, c); + levene_pass_two (pvw->nl, val->f, w, case_data (c, cmd->indep_var)); } - } casereader_destroy (reader); + reader = casereader_clone (input); for ( ; (c = casereader_read (reader) ); case_unref (c)) { int i; + double w = dict_get_case_weight (dict, c, NULL); + for (i = 0; i < cmd->n_vars; ++i) { struct per_var_ws *pvw = &ws.vws[i]; @@ -551,11 +478,12 @@ run_oneway (const struct oneway_spec *cmd, continue; } - covariance_accumulate_pass2 (pvw->cov, c); + levene_pass_three (pvw->nl, val->f, w, case_data (c, cmd->indep_var)); } } casereader_destroy (reader); + for (v = 0; v < cmd->n_vars; ++v) { struct per_var_ws *pvw = &ws.vws[v]; @@ -567,8 +495,6 @@ run_oneway (const struct oneway_spec *cmd, pvw->sst = gsl_matrix_get (cm, 0, 0); - // gsl_matrix_fprintf (stdout, cm, "%g "); - reg_sweep (cm, 0); pvw->sse = gsl_matrix_get (cm, 0, 0); @@ -578,14 +504,13 @@ run_oneway (const struct oneway_spec *cmd, pvw->n_groups = categoricals_total (cats); pvw->mse = (pvw->sst - pvw->ssa) / (n - pvw->n_groups); - } - postcalc (cmd); + gsl_matrix_free (cm); + } - for (v = 0; v < cmd->n_vars; ++v) { - struct categoricals *cats = covariance_get_categoricals (ws.vws[v].cov); + const struct categoricals *cats = covariance_get_categoricals (ws.vws[v].cov); categoricals_done (cats); @@ -593,85 +518,22 @@ run_oneway (const struct oneway_spec *cmd, ws.actual_number_of_groups = categoricals_total (cats); } - if ( cmd->stats & STATS_HOMOGENEITY ) - levene (dict, casereader_clone (input), cmd->indep_var, - cmd->n_vars, cmd->vars, cmd->exclude); - casereader_destroy (input); if (!taint_has_tainted_successor (taint)) output_oneway (cmd, &ws); taint_destroy (taint); -} - -/* Pre calculations */ -static void -precalc (const struct oneway_spec *cmd) -{ - size_t i = 0; - for (i = 0; i < cmd->n_vars; ++i) - { - struct group_proc *gp = group_proc_get (cmd->vars[i]); - struct group_statistics *totals = &gp->ugs; - - /* Create a hash for each of the dependent variables. - The hash contains a group_statistics structure, - and is keyed by value of the independent variable */ - - gp->group_hash = hsh_create (4, compare_group, hash_group, - (hsh_free_func *) free_group, - cmd->indep_var); - - totals->sum = 0; - totals->n = 0; - totals->ssq = 0; - totals->sum_diff = 0; - totals->maximum = -DBL_MAX; - totals->minimum = DBL_MAX; - } -} - -/* Post calculations for the ONEWAY command */ -static void -postcalc (const struct oneway_spec *cmd) -{ - size_t i = 0; - - for (i = 0; i < cmd->n_vars; ++i) + finish: + for (v = 0; v < cmd->n_vars; ++v) { - struct group_proc *gp = group_proc_get (cmd->vars[i]); - struct hsh_table *group_hash = gp->group_hash; - struct group_statistics *totals = &gp->ugs; - - struct hsh_iterator g; - struct group_statistics *gs; - - for (gs = hsh_first (group_hash, &g); - gs != 0; - gs = hsh_next (group_hash, &g)) - { - gs->mean = gs->sum / gs->n; - gs->s_std_dev = sqrt (gs->ssq / gs->n - pow2 (gs->mean)); - - gs->std_dev = sqrt ( - gs->n / (gs->n - 1) * - ( gs->ssq / gs->n - pow2 (gs->mean)) - ); - - gs->se_mean = gs->std_dev / sqrt (gs->n); - gs->mean_diff = gs->sum_diff / gs->n; - } - - totals->mean = totals->sum / totals->n; - totals->std_dev = sqrt ( - totals->n / (totals->n - 1) * - (totals->ssq / totals->n - pow2 (totals->mean)) - ); - - totals->se_mean = totals->std_dev / sqrt (totals->n); + covariance_destroy (ws.vws[v].cov); + levene_destroy (ws.vws[v].nl); + dd_destroy (ws.dd_total[v]); } + free (ws.vws); + free (ws.dd_total); } static void show_contrast_coeffs (const struct oneway_spec *cmd, const struct oneway_workspace *ws); @@ -719,16 +581,6 @@ output_oneway (const struct oneway_spec *cmd, struct oneway_workspace *ws) show_contrast_coeffs (cmd, ws); show_contrast_tests (cmd, ws); } - - /* Clean up */ - for (i = 0; i < cmd->n_vars; ++i ) - { - struct hsh_table *group_hash = group_proc_get (cmd->vars[i])->group_hash; - - hsh_destroy (group_hash); - } - - hsh_destroy (ws->group_hash); } @@ -764,14 +616,16 @@ show_anova_table (const struct oneway_spec *cmd, const struct oneway_workspace * for (i = 0; i < cmd->n_vars; ++i) { double n; - moments1_calculate (ws->dd_total[i]->mom, &n, NULL, NULL, NULL, NULL); - + double df1, df2; + double msa; + const char *s = var_to_string (cmd->vars[i]); const struct per_var_ws *pvw = &ws->vws[i]; - const double df1 = pvw->n_groups - 1; - const double df2 = n - pvw->n_groups; - const double msa = pvw->ssa / df1; - const char *s = var_to_string (cmd->vars[i]); + moments1_calculate (ws->dd_total[i]->mom, &n, NULL, NULL, NULL, NULL); + + df1 = pvw->n_groups - 1; + df2 = n - pvw->n_groups; + msa = pvw->ssa / df1; tab_text (t, 0, i * 3 + 1, TAB_LEFT | TAT_TITLE, s); tab_text (t, 1, i * 3 + 1, TAB_LEFT | TAT_TITLE, _("Between Groups")); @@ -889,8 +743,8 @@ show_descriptives (const struct oneway_spec *cmd, const struct oneway_workspace struct string vstr; - const union value *gval = categoricals_get_value_by_subscript (cats, count); - const struct descriptive_data *dd = categoricals_get_user_data_by_subscript (cats, count); + const union value *gval = categoricals_get_value_by_category (cats, count); + const struct descriptive_data *dd = categoricals_get_user_data_by_category (cats, count); moments1_calculate (dd->mom, &n, &mean, &variance, NULL, NULL); @@ -909,7 +763,7 @@ show_descriptives (const struct oneway_spec *cmd, const struct oneway_workspace /* Now fill in the numbers ... */ - tab_fixed (t, 2, row + count, 0, n, 8, 0); + tab_double (t, 2, row + count, 0, n, wfmt); tab_double (t, 3, row + count, 0, mean, NULL); @@ -1008,21 +862,19 @@ show_homogeneity (const struct oneway_spec *cmd, const struct oneway_workspace * for (v = 0; v < cmd->n_vars; ++v) { double n; - moments1_calculate (ws->dd_total[v]->mom, &n, NULL, NULL, NULL, NULL); - const struct per_var_ws *pvw = &ws->vws[v]; - const struct categoricals *cats = covariance_get_categoricals (pvw->cov); + double F = levene_calculate (pvw->nl); const struct variable *var = cmd->vars[v]; - const struct group_proc *gp = group_proc_get (cmd->vars[v]); const char *s = var_to_string (var); + double df1, df2; - const double df1 = pvw->n_groups - 1; - const double df2 = n - pvw->n_groups; - double F = gp->levene; + moments1_calculate (ws->dd_total[v]->mom, &n, NULL, NULL, NULL, NULL); - tab_text (t, 0, v + 1, TAB_LEFT | TAT_TITLE, s); + df1 = pvw->n_groups - 1; + df2 = n - pvw->n_groups; + tab_text (t, 0, v + 1, TAB_LEFT | TAT_TITLE, s); tab_double (t, 1, v + 1, TAB_RIGHT, F, NULL); tab_fixed (t, 2, v + 1, TAB_RIGHT, df1, 8, 0); @@ -1101,7 +953,7 @@ show_contrast_coeffs (const struct oneway_spec *cmd, const struct oneway_workspa ++count, coeffi = ll_next (coeffi)) { const struct categoricals *cats = covariance_get_categoricals (cov); - const union value *val = categoricals_get_value_by_subscript (cats, count); + const union value *val = categoricals_get_value_by_category (cats, count); struct string vstr; ds_init_empty (&vstr); @@ -1238,14 +1090,14 @@ show_contrast_tests (const struct oneway_spec *cmd, const struct oneway_workspac ++ci, coeffi = ll_next (coeffi)) { double n, mean, variance; - const struct descriptive_data *dd = categoricals_get_user_data_by_subscript (cats, ci); - - moments1_calculate (dd->mom, &n, &mean, &variance, NULL, NULL); - + const struct descriptive_data *dd = categoricals_get_user_data_by_category (cats, ci); struct coeff_node *cn = ll_data (coeffi, struct coeff_node, ll); const double coef = cn->coeff; + double winv ; + + moments1_calculate (dd->mom, &n, &mean, &variance, NULL, NULL); - const double winv = variance / n; + winv = variance / n; contrast_value += coef * mean;