X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Flanguage%2Fstats%2Fchisquare.c;h=d1d81bab39dcc26e422aa32e96354c5c16a808d5;hb=f6c6ee97c412e47eb1c7c30adc224e93b6c1586a;hp=bc1b6474ff16c167d06ea94ca55de828475d7a9e;hpb=7c08a6e1009cf60847e770a77a73c650e9326379;p=pspp diff --git a/src/language/stats/chisquare.c b/src/language/stats/chisquare.c index bc1b6474ff..d1d81bab39 100644 --- a/src/language/stats/chisquare.c +++ b/src/language/stats/chisquare.c @@ -1,5 +1,5 @@ /* PSPP - a program for statistical analysis. - Copyright (C) 2006, 2007, 2009 Free Software Foundation, Inc. + Copyright (C) 2006, 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 @@ -16,286 +16,116 @@ #include -#include - -#include -#include - -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include +#include "language/stats/chisquare.h" #include +#include +#include -#include "xalloc.h" +#include "data/case.h" +#include "data/casereader.h" +#include "data/dataset.h" +#include "data/dictionary.h" +#include "data/format.h" +#include "data/value-labels.h" +#include "data/variable.h" +#include "language/stats/freq.h" +#include "language/stats/npar.h" +#include "libpspp/array.h" +#include "libpspp/assertion.h" +#include "libpspp/cast.h" +#include "libpspp/compiler.h" +#include "libpspp/hash-functions.h" +#include "libpspp/message.h" +#include "libpspp/taint.h" +#include "output/pivot-table.h" + +#include "gl/xalloc.h" #include "gettext.h" +#define N_(msgid) msgid #define _(msgid) gettext (msgid) -/* Return a hash table containing the frequency counts of each - value of VAR in CF . - It is the caller's responsibility to free the hash table when - no longer required. -*/ -static struct hsh_table * +/* Adds frequency counts of each value of VAR in INPUT between LO and HI to + FREQ_HASH. LO and HI and each input value is truncated to an integer. + Returns true if successful, false on input error. It is the caller's + responsibility to initialize FREQ_HASH and to free it when no longer + required, even on failure. */ +static bool create_freq_hash_with_range (const struct dictionary *dict, struct casereader *input, const struct variable *var, - double lo, - double hi) + double lo_, double hi_, + struct hmap *freq_hash) { + struct freq **entries; bool warn = true; - float i_d; struct ccase *c; + double lo, hi; + double i_d; - struct hsh_table *freq_hash = - hsh_create (4, compare_freq, hash_freq, - free_freq_mutable_hash, - (void *) var); + assert (var_is_numeric (var)); + lo = trunc (lo_); + hi = trunc (hi_); /* Populate the hash with zero entries */ - for (i_d = trunc (lo); i_d <= trunc (hi); i_d += 1.0 ) + entries = xnmalloc (hi - lo + 1, sizeof *entries); + for (i_d = lo; i_d <= hi; i_d += 1.0) { - struct freq_mutable *fr = xmalloc (sizeof (*fr)); - value_init (&fr->value, 0); - fr->value.f = i_d; - fr->count = 0; - hsh_insert (freq_hash, fr); + size_t ofs = i_d - lo; + union value value = { i_d }; + entries[ofs] = freq_hmap_insert (freq_hash, &value, 0, + value_hash (&value, 0, 0)); } for (; (c = casereader_read (input)) != NULL; case_unref (c)) { - struct freq_mutable fr; - fr.value.f = trunc (case_num (c, var)); - if (fr.value.f >= lo && fr.value.f <= hi) + double x = trunc (case_num (c, var)); + if (x >= lo && x <= hi) { - struct freq_mutable *existing_fr = hsh_force_find (freq_hash, &fr); - existing_fr->count += dict_get_case_weight (dict, c, &warn); + size_t ofs = x - lo; + struct freq *fr = entries[ofs]; + fr->count += dict_get_case_weight (dict, c, &warn); } } - if (casereader_destroy (input)) - return freq_hash; - else - { - hsh_destroy (freq_hash); - return NULL; - } -} + free (entries); + + return casereader_destroy (input); +} -/* Return a hash table containing the frequency counts of each - value of VAR in INPUT . - It is the caller's responsibility to free the hash table when - no longer required. -*/ -static struct hsh_table * +/* Adds frequency counts of each value of VAR in INPUT to FREQ_HASH. LO and HI + and each input value is truncated to an integer. Returns true if + successful, false on input error. It is the caller's responsibility to + initialize FREQ_HASH and to free it when no longer required, even on + failure. */ +static bool create_freq_hash (const struct dictionary *dict, struct casereader *input, - const struct variable *var) + const struct variable *var, + struct hmap *freq_hash) { int width = var_get_width (var); bool warn = true; struct ccase *c; - struct hsh_table *freq_hash = - hsh_create (4, compare_freq, hash_freq, - free_freq_mutable_hash, - (void *) var); - for (; (c = casereader_read (input)) != NULL; case_unref (c)) { - struct freq_mutable fr; - void **p; - - fr.value = *case_data (c, var); - fr.count = dict_get_case_weight (dict, c, &warn); - - p = hsh_probe (freq_hash, &fr); - if (*p == NULL) - { - struct freq_mutable *new_fr = *p = xmalloc (sizeof *new_fr); - value_init (&new_fr->value, width); - value_copy (&new_fr->value, &fr.value, width); - new_fr->count = fr.count; - } - else - { - struct freq *existing_fr = *p; - existing_fr->count += fr.count; - } - } - if (casereader_destroy (input)) - return freq_hash; - else - { - hsh_destroy (freq_hash); - return NULL; - } -} - - - -static struct tab_table * -create_variable_frequency_table (const struct dictionary *dict, - struct casereader *input, - const struct chisquare_test *test, - int v, - struct hsh_table **freq_hash) - -{ - int i; - const struct one_sample_test *ost = (const struct one_sample_test*)test; - int n_cells; - struct tab_table *table ; - const struct variable *var = ost->vars[v]; - - *freq_hash = create_freq_hash (dict, input, var); - if (*freq_hash == NULL) - return NULL; - - n_cells = hsh_count (*freq_hash); - - if ( test->n_expected > 0 && n_cells != test->n_expected ) - { - msg(ME, _("CHISQUARE test specified %d expected values, but" - " %d distinct values were encountered in variable %s."), - test->n_expected, n_cells, - var_get_name (var) - ); - hsh_destroy (*freq_hash); - *freq_hash = NULL; - return NULL; - } - - table = tab_create(4, n_cells + 2); - tab_dim (table, tab_natural_dimensions, NULL, NULL); - - tab_title (table, var_to_string(var)); - tab_text (table, 1, 0, TAB_LEFT, _("Observed N")); - tab_text (table, 2, 0, TAB_LEFT, _("Expected N")); - tab_text (table, 3, 0, TAB_LEFT, _("Residual")); - - tab_headers (table, 1, 0, 1, 0); - - tab_box (table, TAL_1, TAL_1, -1, -1, - 0, 0, tab_nc (table) - 1, tab_nr(table) - 1 ); - - tab_hline (table, TAL_1, 0, tab_nc(table) - 1, 1); - - tab_vline (table, TAL_2, 1, 0, tab_nr(table) - 1); - for ( i = 2 ; i < 4 ; ++i ) - tab_vline (table, TAL_1, i, 0, tab_nr(table) - 1); - - - tab_text (table, 0, tab_nr (table) - 1, TAB_LEFT, _("Total")); - - return table; -} - - -static struct tab_table * -create_combo_frequency_table (const struct chisquare_test *test) -{ - int i; - const struct one_sample_test *ost = (const struct one_sample_test*)test; - - struct tab_table *table ; - - int n_cells = test->hi - test->lo + 1; - - table = tab_create(1 + ost->n_vars * 4, n_cells + 3); - tab_dim (table, tab_natural_dimensions, NULL, NULL); - - tab_title (table, _("Frequencies")); - for ( i = 0 ; i < ost->n_vars ; ++i ) - { - const struct variable *var = ost->vars[i]; - tab_text (table, i * 4 + 1, 1, TAB_LEFT, _("Category")); - tab_text (table, i * 4 + 2, 1, TAB_LEFT, _("Observed N")); - tab_text (table, i * 4 + 3, 1, TAB_LEFT, _("Expected N")); - tab_text (table, i * 4 + 4, 1, TAB_LEFT, _("Residual")); - - tab_vline (table, TAL_2, i * 4 + 1, - 0, tab_nr (table) - 1); - - tab_vline (table, TAL_1, i * 4 + 2, - 0, tab_nr (table) - 1); - - tab_vline (table, TAL_1, i * 4 + 3, - 1, tab_nr (table) - 1); + const union value *value = case_data (c, var); + size_t hash = value_hash (value, width, 0); + double weight = dict_get_case_weight (dict, c, &warn); + struct freq *f; - tab_vline (table, TAL_1, i * 4 + 4, - 1, tab_nr (table) - 1); + f = freq_hmap_search (freq_hash, value, width, hash); + if (f == NULL) + f = freq_hmap_insert (freq_hash, value, width, hash); - - tab_joint_text (table, - i * 4 + 1, 0, - i * 4 + 4, 0, - TAB_CENTER, - var_to_string (var)); + f->count += weight; } - for ( i = test->lo ; i <= test->hi ; ++i ) - tab_fixed (table, 0, 2 + i - test->lo, - TAB_LEFT, 1 + i - test->lo, 8, 0); - - tab_headers (table, 1, 0, 2, 0); - - tab_box (table, TAL_1, TAL_1, -1, -1, - 0, 0, tab_nc (table) - 1, tab_nr(table) - 1 ); - - tab_hline (table, TAL_1, 1, tab_nc(table) - 1, 1); - tab_hline (table, TAL_1, 0, tab_nc(table) - 1, 2); - - tab_text (table, 0, tab_nr (table) - 1, TAB_LEFT, _("Total")); - - return table; + return casereader_destroy (input); } - -static struct tab_table * -create_stats_table (const struct chisquare_test *test) -{ - const struct one_sample_test *ost = (const struct one_sample_test*) test; - - struct tab_table *table; - table = tab_create (1 + ost->n_vars, 4); - tab_dim (table, tab_natural_dimensions, NULL, NULL); - tab_title (table, _("Test Statistics")); - tab_headers (table, 1, 0, 1, 0); - - tab_box (table, TAL_1, TAL_1, -1, -1, - 0, 0, tab_nc(table) - 1, tab_nr(table) - 1 ); - - tab_box (table, -1, -1, -1, TAL_1, - 1, 0, tab_nc(table) - 1, tab_nr(table) - 1 ); - - - tab_vline (table, TAL_2, 1, 0, tab_nr (table) - 1); - tab_hline (table, TAL_1, 0, tab_nc (table) - 1, 1); - - - tab_text (table, 0, 1, TAB_LEFT, _("Chi-Square")); - tab_text (table, 0, 2, TAB_LEFT, _("df")); - tab_text (table, 0, 3, TAB_LEFT, _("Asymp. Sig.")); - - return table; -} - - void chisquare_execute (const struct dataset *ds, struct casereader *input, @@ -309,182 +139,208 @@ chisquare_execute (const struct dataset *ds, struct chisquare_test *cst = UP_CAST (test, struct chisquare_test, parent.parent); struct one_sample_test *ost = &cst->parent; - int n_cells = 0; double total_expected = 0.0; - const struct variable *wvar = dict_get_weight (dict); - const struct fmt_spec *wfmt = wvar ? - var_get_print_format (wvar) : & F_8_0; - double *df = xzalloc (sizeof (*df) * ost->n_vars); - double *xsq = xzalloc (sizeof (*df) * ost->n_vars); + double *df = XCALLOC (ost->n_vars, double); + double *xsq = XCALLOC (ost->n_vars, double); bool ok; - for ( i = 0 ; i < cst->n_expected ; ++i ) + for (i = 0 ; i < cst->n_expected ; ++i) total_expected += cst->expected[i]; - if ( cst->ranged == false ) + if (cst->ranged == false) { - for ( v = 0 ; v < ost->n_vars ; ++v ) + for (v = 0 ; v < ost->n_vars ; ++v) { - double total_obs = 0.0; - struct hsh_table *freq_hash = NULL; + const struct variable *var = ost->vars[v]; + + struct hmap freq_hash = HMAP_INITIALIZER (freq_hash); struct casereader *reader = casereader_create_filter_missing (casereader_clone (input), - &ost->vars[v], 1, exclude, + &var, 1, exclude, NULL, NULL); - struct tab_table *freq_table = - create_variable_frequency_table(dict, reader, cst, v, &freq_hash); - - struct freq **ff; - - if ( NULL == freq_table ) - continue; - ff = (struct freq **) hsh_sort (freq_hash); + if (!create_freq_hash (dict, reader, var, &freq_hash)) + { + freq_hmap_destroy (&freq_hash, var_get_width (var)); + return; + } + + size_t n_cells = hmap_count (&freq_hash); + if (cst->n_expected > 0 && n_cells != cst->n_expected) + { + msg (ME, _("CHISQUARE test specified %d expected values, but " + "variable %s has %zu distinct values."), + cst->n_expected, var_get_name (var), n_cells); + freq_hmap_destroy (&freq_hash, var_get_width (var)); + continue; + } + + struct pivot_table *table = pivot_table_create__ ( + pivot_value_new_variable (var), "Chisquare"); + pivot_table_set_weight_var (table, dict_get_weight (dict)); + + pivot_dimension_create ( + table, PIVOT_AXIS_COLUMN, N_("Statistics"), + N_("Observed N"), PIVOT_RC_COUNT, + N_("Expected N"), PIVOT_RC_OTHER, + N_("Residual"), PIVOT_RC_RESIDUAL); + + struct freq **ff = freq_hmap_sort (&freq_hash, var_get_width (var)); - n_cells = hsh_count (freq_hash); - - for ( i = 0 ; i < n_cells ; ++i ) + double total_obs = 0.0; + for (size_t i = 0; i < n_cells; i++) total_obs += ff[i]->count; + struct pivot_dimension *values = pivot_dimension_create ( + table, PIVOT_AXIS_ROW, N_("Value")); + values->root->show_label = true; + xsq[v] = 0.0; - for ( i = 0 ; i < n_cells ; ++i ) + for (size_t i = 0; i < n_cells; i++) { - struct string str; - double exp; - const union value *observed_value = &ff[i]->value; - - ds_init_empty (&str); - var_append_value_name (ost->vars[v], observed_value, &str); - - /* The key */ - tab_text (freq_table, 0, i + 1, TAB_LEFT, ds_cstr (&str)); - ds_destroy (&str); - - - /* The observed N */ - tab_double (freq_table, 1, i + 1, TAB_NONE, - ff[i]->count, wfmt); - - if ( cst->n_expected > 0 ) - exp = cst->expected[i] * total_obs / total_expected ; - else - exp = total_obs / (double) n_cells; - - tab_double (freq_table, 2, i + 1, TAB_NONE, - exp, NULL); - - /* The residual */ - tab_double (freq_table, 3, i + 1, TAB_NONE, - ff[i]->count - exp, NULL); + int row = pivot_category_create_leaf ( + values->root, pivot_value_new_var_value ( + var, &ff[i]->values[0])); + + double exp = (cst->n_expected > 0 + ? cst->expected[i] * total_obs / total_expected + : total_obs / (double) n_cells); + double entries[] = { + ff[i]->count, + exp, + ff[i]->count - exp, + }; + for (size_t j = 0; j < sizeof entries / sizeof *entries; j++) + pivot_table_put2 ( + table, j, row, pivot_value_new_number (entries[j])); xsq[v] += (ff[i]->count - exp) * (ff[i]->count - exp) / exp; } df[v] = n_cells - 1.0; - tab_double (freq_table, 1, i + 1, TAB_NONE, - total_obs, wfmt); + int row = pivot_category_create_leaf ( + values->root, pivot_value_new_text (N_("Total"))); + pivot_table_put2 (table, 0, row, + pivot_value_new_number (total_obs)); - tab_submit (freq_table); + pivot_table_submit (table); - hsh_destroy (freq_hash); + freq_hmap_destroy (&freq_hash, var_get_width (var)); + free (ff); } } else /* ranged == true */ { - struct tab_table *freq_table = create_combo_frequency_table (cst); - - n_cells = cst->hi - cst->lo + 1; - - for ( v = 0 ; v < ost->n_vars ; ++v ) + struct pivot_table *table = pivot_table_create (N_("Frequencies")); + pivot_table_set_weight_var (table, dict_get_weight (dict)); + + pivot_dimension_create ( + table, PIVOT_AXIS_COLUMN, N_("Statistics"), + N_("Category"), + N_("Observed N"), PIVOT_RC_COUNT, + N_("Expected N"), PIVOT_RC_OTHER, + N_("Residual"), PIVOT_RC_RESIDUAL); + + struct pivot_dimension *var_dim + = pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Variable")); + for (size_t i = 0 ; i < ost->n_vars ; ++i) + pivot_category_create_leaf (var_dim->root, + pivot_value_new_variable (ost->vars[i])); + + struct pivot_dimension *category_dim + = pivot_dimension_create (table, PIVOT_AXIS_ROW, N_("Category")); + size_t n_cells = cst->hi - cst->lo + 1; + for (size_t i = 0 ; i < n_cells; ++i) + pivot_category_create_leaf (category_dim->root, + pivot_value_new_integer (i + 1)); + pivot_category_create_leaves (category_dim->root, N_("Total")); + + for (size_t v = 0 ; v < ost->n_vars ; ++v) { - double total_obs = 0.0; + const struct variable *var = ost->vars[v]; struct casereader *reader = casereader_create_filter_missing (casereader_clone (input), - &ost->vars[v], 1, exclude, + &var, 1, exclude, NULL, NULL); - struct hsh_table *freq_hash = - create_freq_hash_with_range (dict, reader, - ost->vars[v], cst->lo, cst->hi); - - struct freq **ff; - - if (freq_hash == NULL) - continue; - - ff = (struct freq **) hsh_sort (freq_hash); - assert ( n_cells == hsh_count (freq_hash)); - - for ( i = 0 ; i < hsh_count (freq_hash) ; ++i ) + struct hmap freq_hash = HMAP_INITIALIZER (freq_hash); + if (!create_freq_hash_with_range (dict, reader, var, + cst->lo, cst->hi, &freq_hash)) + { + freq_hmap_destroy (&freq_hash, var_get_width (var)); + continue; + } + + struct freq **ff = freq_hmap_sort (&freq_hash, var_get_width (var)); + + double total_obs = 0.0; + for (size_t i = 0 ; i < hmap_count (&freq_hash) ; ++i) total_obs += ff[i]->count; xsq[v] = 0.0; - for ( i = 0 ; i < hsh_count (freq_hash) ; ++i ) + for (size_t i = 0 ; i < hmap_count (&freq_hash) ; ++i) { - struct string str; - double exp; - - const union value *observed_value = &ff[i]->value; - - ds_init_empty (&str); - var_append_value_name (ost->vars[v], observed_value, &str); - /* The key */ - tab_text (freq_table, v * 4 + 1, i + 2 , TAB_LEFT, - ds_cstr (&str)); - ds_destroy (&str); - - /* The observed N */ - tab_double (freq_table, v * 4 + 2, i + 2 , TAB_NONE, - ff[i]->count, wfmt); - - if ( cst->n_expected > 0 ) - exp = cst->expected[i] * total_obs / total_expected ; - else - exp = total_obs / (double) hsh_count (freq_hash); - - /* The expected N */ - tab_double (freq_table, v * 4 + 3, i + 2 , TAB_NONE, - exp, NULL); + /* Category. */ + pivot_table_put3 (table, 0, v, i, + pivot_value_new_var_value ( + var, &ff[i]->values[0])); + + double exp = (cst->n_expected > 0 + ? cst->expected[i] * total_obs / total_expected + : total_obs / (double) hmap_count (&freq_hash)); + double entries[] = { + ff[i]->count, + exp, + ff[i]->count - exp, + }; + for (size_t j = 0; j < sizeof entries / sizeof *entries; j++) + pivot_table_put3 (table, j + 1, v, i, + pivot_value_new_number (entries[j])); - /* The residual */ - tab_double (freq_table, v * 4 + 4, i + 2 , TAB_NONE, - ff[i]->count - exp, NULL); xsq[v] += (ff[i]->count - exp) * (ff[i]->count - exp) / exp; } - - tab_double (freq_table, v * 4 + 2, tab_nr (freq_table) - 1, TAB_NONE, - total_obs, wfmt); - df[v] = n_cells - 1.0; - hsh_destroy (freq_hash); + freq_hmap_destroy (&freq_hash, var_get_width (var)); + free (ff); + + pivot_table_put3 (table, 1, v, n_cells, + pivot_value_new_number (total_obs)); } - tab_submit (freq_table); + pivot_table_submit (table); } ok = !taint_has_tainted_successor (casereader_get_taint (input)); casereader_destroy (input); if (ok) { - struct tab_table *stats_table = create_stats_table (cst); + struct pivot_table *table = pivot_table_create (N_("Test Statistics")); - /* Populate the summary statistics table */ - for ( v = 0 ; v < ost->n_vars ; ++v ) + pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Statistics"), + N_("Chi-square"), PIVOT_RC_OTHER, + N_("df"), PIVOT_RC_INTEGER, + N_("Asymp. Sig."), PIVOT_RC_SIGNIFICANCE); + + struct pivot_dimension *variables = pivot_dimension_create ( + table, PIVOT_AXIS_ROW, N_("Variable")); + + for (size_t v = 0 ; v < ost->n_vars ; ++v) { const struct variable *var = ost->vars[v]; - tab_text (stats_table, 1 + v, 0, TAB_CENTER, var_get_name (var)); - - tab_double (stats_table, 1 + v, 1, TAB_NONE, xsq[v], NULL); - tab_fixed (stats_table, 1 + v, 2, TAB_NONE, df[v], 8, 0); + int row = pivot_category_create_leaf ( + variables->root, pivot_value_new_variable (var)); - tab_double (stats_table, 1 + v, 3, TAB_NONE, - gsl_cdf_chisq_Q (xsq[v], df[v]), NULL); + double sig = gsl_cdf_chisq_Q (xsq[v], df[v]); + double entries[] = { xsq[v], df[v], sig }; + for (size_t i = 0; i < sizeof entries / sizeof *entries; i++) + pivot_table_put2 (table, i, row, + pivot_value_new_number (entries[i])); } - tab_submit (stats_table); + pivot_table_submit (table); } free (xsq);