X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Flanguage%2Fstats%2Foneway.c;h=2423215f3ac228c5d866801fec53194d2cf94dde;hb=8d9fc72f4f9cf0bf5bbcb80bb020592cf9496a97;hp=3995a1a4cd81d2c2316afd49ed6ee8603dbe148e;hpb=c759a7875e72ad8f1523f1c8980ca5a9cb8d20c1;p=pspp diff --git a/src/language/stats/oneway.c b/src/language/stats/oneway.c index 3995a1a4cd..2423215f3a 100644 --- a/src/language/stats/oneway.c +++ b/src/language/stats/oneway.c @@ -20,6 +20,11 @@ #include #include +#include +#include +#include +#include + #include #include @@ -76,7 +81,7 @@ struct contrasts_node bool bad_count; /* True if the number of coefficients does not equal the number of groups */ }; -struct oneway +struct oneway_spec { size_t n_vars; const struct variable **vars; @@ -88,6 +93,30 @@ struct oneway enum missing_type missing_type; enum mv_class exclude; + /* List of contrasts */ + struct ll_list contrast_list; + + /* The weight variable */ + const struct variable *wv; +}; + + +/* Workspace variable for each dependent variable */ +struct per_var_ws +{ + struct covariance *cov; + + double sst; + double sse; + double ssa; + + int n_groups; + + double cc; +}; + +struct oneway_workspace +{ /* The number of distinct values of the independent variable, when all missing values are disregarded */ int actual_number_of_groups; @@ -96,32 +125,29 @@ struct oneway variable */ struct hsh_table *group_hash; - /* List of contrasts */ - struct ll_list contrast_list; + struct per_var_ws *vws; }; /* Routines to show the output tables */ -static void show_anova_table (const struct oneway *); -static void show_descriptives (const struct oneway *, const struct dictionary *dict); -static void show_homogeneity (const struct oneway *); +static void show_anova_table (const struct oneway_spec *, const struct oneway_workspace *); +static void show_descriptives (const struct oneway_spec *); +static void show_homogeneity (const struct oneway_spec *); -static void output_oneway (const struct oneway *, const struct dictionary *dict); -static void run_oneway (struct oneway *cmd, struct casereader *input, const struct dataset *ds); +static void output_oneway (const struct oneway_spec *, struct oneway_workspace *ws); +static void run_oneway (const struct oneway_spec *cmd, struct casereader *input, const struct dataset *ds); int cmd_oneway (struct lexer *lexer, struct dataset *ds) { - const struct dictionary *dict = dataset_dict (ds); - - struct oneway oneway ; + const struct dictionary *dict = dataset_dict (ds); + struct oneway_spec oneway ; oneway.n_vars = 0; oneway.vars = NULL; oneway.indep_var = NULL; oneway.stats = 0; oneway.missing_type = MISS_ANALYSIS; oneway.exclude = MV_ANY; - oneway.actual_number_of_groups = 0; - oneway.group_hash = NULL; + oneway.wv = dict_get_weight (dict); ll_init (&oneway.contrast_list); @@ -179,13 +205,13 @@ cmd_oneway (struct lexer *lexer, struct dataset *ds) while (lex_token (lexer) != '.' && lex_token (lexer) != '/') { - union value val; - if ( parse_value (lexer, &val, 0)) + if ( lex_is_number (lexer)) { struct coeff_node *cc = xmalloc (sizeof *cc); - cc->coeff = val.f; + cc->coeff = lex_number (lexer); ll_push_tail (coefficient_list, &cc->ll); + lex_get (lexer); } else { @@ -237,6 +263,8 @@ cmd_oneway (struct lexer *lexer, struct dataset *ds) struct casereader *group; bool ok; + + grouper = casegrouper_create_splits (proc_open (ds), dict); while (casegrouper_get_next_group (grouper, &group)) run_oneway (&oneway, group, ds); @@ -277,20 +305,38 @@ free_double (void *value_, const void *aux UNUSED) free (value); } -static void postcalc (const struct oneway *cmd); -static void precalc (const struct oneway *cmd); +static void postcalc (const struct oneway_spec *cmd); +static void precalc (const struct oneway_spec *cmd); static void -run_oneway (struct oneway *cmd, +run_oneway (const struct oneway_spec *cmd, struct casereader *input, const struct dataset *ds) { + int v; struct taint *taint; struct dictionary *dict = dataset_dict (ds); struct casereader *reader; struct ccase *c; + + struct oneway_workspace ws; + + ws.vws = xmalloc (cmd->n_vars * sizeof (*ws.vws)); + + + for (v = 0; v < cmd->n_vars; ++v) + { + struct categoricals *cats = categoricals_create (&cmd->indep_var, 1, + cmd->wv, cmd->exclude); + + ws.vws[v].cov = covariance_2pass_create (1, &cmd->vars[v], + cats, + cmd->wv, cmd->exclude); + ws.vws[v].cc = 0; + } + c = casereader_peek (input, 0); if (c == NULL) { @@ -302,7 +348,7 @@ run_oneway (struct oneway *cmd, taint = taint_clone (casereader_get_taint (input)); - cmd->group_hash = hsh_create (4, + ws.group_hash = hsh_create (4, compare_double_3way, do_hash_double, free_double, @@ -318,6 +364,7 @@ run_oneway (struct oneway *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; @@ -325,7 +372,7 @@ run_oneway (struct oneway *cmd, 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 (cmd->group_hash, &indep_val->f); + void **p = hsh_probe (ws.group_hash, &indep_val->f); if (*p == NULL) { double *value = *p = xmalloc (sizeof *value); @@ -334,6 +381,13 @@ run_oneway (struct oneway *cmd, for (i = 0; i < cmd->n_vars; ++i) { + { + struct per_var_ws *pvw = &ws.vws[i]; + + pvw->cc += weight; + covariance_accumulate_pass1 (pvw->cov, c); + } + const struct variable *v = cmd->vars[i]; const union value *val = case_data (c, v); @@ -389,6 +443,34 @@ run_oneway (struct oneway *cmd, } casereader_destroy (reader); + reader = casereader_clone (input); + for ( ; (c = casereader_read (reader) ); case_unref (c)) + { + int i; + for (i = 0; i < cmd->n_vars; ++i) + { + struct per_var_ws *pvw = &ws.vws[i]; + covariance_accumulate_pass2 (pvw->cov, c); + } + } + casereader_destroy (reader); + + for (v = 0; v < cmd->n_vars; ++v) + { + struct per_var_ws *pvw = &ws.vws[v]; + gsl_matrix *cm = covariance_calculate_unnormalized (pvw->cov); + const struct categoricals *cats = covariance_get_categoricals (pvw->cov); + + pvw->sst = gsl_matrix_get (cm, 0, 0); + + reg_sweep (cm, 0); + + pvw->sse = gsl_matrix_get (cm, 0, 0); + + pvw->ssa = pvw->sst - pvw->sse; + + pvw->n_groups = categoricals_total (cats); + } postcalc (cmd); @@ -398,17 +480,17 @@ run_oneway (struct oneway *cmd, casereader_destroy (input); - cmd->actual_number_of_groups = hsh_count (cmd->group_hash); + ws.actual_number_of_groups = hsh_count (ws.group_hash); if (!taint_has_tainted_successor (taint)) - output_oneway (cmd, dict); + output_oneway (cmd, &ws); taint_destroy (taint); } /* Pre calculations */ static void -precalc (const struct oneway *cmd) +precalc (const struct oneway_spec *cmd) { size_t i = 0; @@ -436,7 +518,7 @@ precalc (const struct oneway *cmd) /* Post calculations for the ONEWAY command */ static void -postcalc (const struct oneway *cmd) +postcalc (const struct oneway_spec *cmd) { size_t i = 0; @@ -475,11 +557,11 @@ postcalc (const struct oneway *cmd) } } -static void show_contrast_coeffs (const struct oneway *cmd); -static void show_contrast_tests (const struct oneway *cmd); +static void show_contrast_coeffs (const struct oneway_spec *cmd, struct oneway_workspace *ws); +static void show_contrast_tests (const struct oneway_spec *cmd, struct oneway_workspace *ws); static void -output_oneway (const struct oneway *cmd, const struct dictionary *dict) +output_oneway (const struct oneway_spec *cmd, struct oneway_workspace *ws) { size_t i = 0; @@ -492,7 +574,7 @@ output_oneway (const struct oneway *cmd, const struct dictionary *dict) struct ll_list *cl = &coeff_list->coefficient_list; ++i; - if (ll_count (cl) != cmd->actual_number_of_groups) + if (ll_count (cl) != ws->actual_number_of_groups) { msg (SW, _("Number of contrast coefficients must equal the number of groups")); @@ -508,18 +590,18 @@ output_oneway (const struct oneway *cmd, const struct dictionary *dict) } if (cmd->stats & STATS_DESCRIPTIVES) - show_descriptives (cmd, dict); + show_descriptives (cmd); if (cmd->stats & STATS_HOMOGENEITY) show_homogeneity (cmd); - show_anova_table (cmd); + show_anova_table (cmd, ws); if (ll_count (&cmd->contrast_list) > 0) { - show_contrast_coeffs (cmd); - show_contrast_tests (cmd); + show_contrast_coeffs (cmd, ws); + show_contrast_tests (cmd, ws); } @@ -531,13 +613,13 @@ output_oneway (const struct oneway *cmd, const struct dictionary *dict) hsh_destroy (group_hash); } - hsh_destroy (cmd->group_hash); + hsh_destroy (ws->group_hash); } /* Show the ANOVA table */ static void -show_anova_table (const struct oneway *cmd) +show_anova_table (const struct oneway_spec *cmd, const struct oneway_workspace *ws) { size_t i; int n_cols =7; @@ -566,21 +648,13 @@ show_anova_table (const struct oneway *cmd) for (i = 0; i < cmd->n_vars; ++i) { - struct group_statistics *totals = &group_proc_get (cmd->vars[i])->ugs; - struct hsh_table *group_hash = group_proc_get (cmd->vars[i])->group_hash; - struct hsh_iterator g; - struct group_statistics *gs; - double ssa = 0; - const char *s = var_to_string (cmd->vars[i]); - - for (gs = hsh_first (group_hash, &g); - gs != 0; - gs = hsh_next (group_hash, &g)) - { - ssa += pow2 (gs->sum) / gs->n; - } + const struct per_var_ws *pvw = &ws->vws[i]; + struct group_proc *gp = group_proc_get (cmd->vars[i]); + const double df1 = pvw->n_groups - 1; + const double df2 = pvw->cc - pvw->n_groups; + const double msa = pvw->ssa / df1; - ssa -= pow2 (totals->sum) / totals->n; + const char *s = var_to_string (cmd->vars[i]); 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")); @@ -590,44 +664,35 @@ show_anova_table (const struct oneway *cmd) if (i > 0) tab_hline (t, TAL_1, 0, n_cols - 1, i * 3 + 1); - { - struct group_proc *gp = group_proc_get (cmd->vars[i]); - const double sst = totals->ssq - pow2 (totals->sum) / totals->n; - const double df1 = gp->n_groups - 1; - const double df2 = totals->n - gp->n_groups; - const double msa = ssa / df1; - - gp->mse = (sst - ssa) / df2; + gp->mse = (pvw->sst - pvw->ssa) / df2; - /* Sums of Squares */ - tab_double (t, 2, i * 3 + 1, 0, ssa, NULL); - tab_double (t, 2, i * 3 + 3, 0, sst, NULL); - tab_double (t, 2, i * 3 + 2, 0, sst - ssa, NULL); + /* Sums of Squares */ + tab_double (t, 2, i * 3 + 1, 0, pvw->ssa, NULL); + tab_double (t, 2, i * 3 + 3, 0, pvw->sst, NULL); + tab_double (t, 2, i * 3 + 2, 0, pvw->sse, NULL); - /* Degrees of freedom */ - tab_fixed (t, 3, i * 3 + 1, 0, df1, 4, 0); - tab_fixed (t, 3, i * 3 + 2, 0, df2, 4, 0); - tab_fixed (t, 3, i * 3 + 3, 0, totals->n - 1, 4, 0); + /* Degrees of freedom */ + tab_fixed (t, 3, i * 3 + 1, 0, df1, 4, 0); + tab_fixed (t, 3, i * 3 + 2, 0, df2, 4, 0); + tab_fixed (t, 3, i * 3 + 3, 0, pvw->cc - 1, 4, 0); - /* Mean Squares */ - tab_double (t, 4, i * 3 + 1, TAB_RIGHT, msa, NULL); - tab_double (t, 4, i * 3 + 2, TAB_RIGHT, gp->mse, NULL); + /* Mean Squares */ + tab_double (t, 4, i * 3 + 1, TAB_RIGHT, msa, NULL); + tab_double (t, 4, i * 3 + 2, TAB_RIGHT, gp->mse, NULL); - { - const double F = msa / gp->mse ; + { + const double F = msa / gp->mse ; - /* The F value */ - tab_double (t, 5, i * 3 + 1, 0, F, NULL); + /* The F value */ + tab_double (t, 5, i * 3 + 1, 0, F, NULL); - /* The significance */ - tab_double (t, 6, i * 3 + 1, 0, gsl_cdf_fdist_Q (F, df1, df2), NULL); - } + /* The significance */ + tab_double (t, 6, i * 3 + 1, 0, gsl_cdf_fdist_Q (F, df1, df2), NULL); } } - tab_title (t, _("ANOVA")); tab_submit (t); } @@ -635,7 +700,7 @@ show_anova_table (const struct oneway *cmd) /* Show the descriptives table */ static void -show_descriptives (const struct oneway *cmd, const struct dictionary *dict) +show_descriptives (const struct oneway_spec *cmd) { size_t v; int n_cols = 10; @@ -645,8 +710,7 @@ show_descriptives (const struct oneway *cmd, const struct dictionary *dict) const double confidence = 0.95; const double q = (1.0 - confidence) / 2.0; - const struct variable *wv = dict_get_weight (dict); - const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : & F_8_0; + const struct fmt_spec *wfmt = cmd->wv ? var_get_print_format (cmd->wv) : & F_8_0; int n_rows = 2; @@ -790,7 +854,7 @@ show_descriptives (const struct oneway *cmd, const struct dictionary *dict) /* Show the homogeneity table */ static void -show_homogeneity (const struct oneway *cmd) +show_homogeneity (const struct oneway_spec *cmd) { size_t v; int n_cols = 5; @@ -850,19 +914,19 @@ show_homogeneity (const struct oneway *cmd) /* Show the contrast coefficients table */ static void -show_contrast_coeffs (const struct oneway *cmd) +show_contrast_coeffs (const struct oneway_spec *cmd, struct oneway_workspace *ws) { int c_num = 0; struct ll *cli; int n_contrasts = ll_count (&cmd->contrast_list); - int n_cols = 2 + cmd->actual_number_of_groups; + int n_cols = 2 + ws->actual_number_of_groups; int n_rows = 2 + n_contrasts; - void *const *group_values; - struct tab_table *t; + const struct covariance *cov = ws->vws[0].cov ; + t = tab_create (n_cols, n_rows); tab_headers (t, 2, 0, 2, 0); @@ -898,34 +962,29 @@ show_contrast_coeffs (const struct oneway *cmd) tab_joint_text (t, 2, 0, n_cols - 1, 0, TAB_CENTER | TAT_TITLE, var_to_string (cmd->indep_var)); - group_values = hsh_sort (cmd->group_hash); - for ( cli = ll_head (&cmd->contrast_list); cli != ll_null (&cmd->contrast_list); cli = ll_next (cli)) { int count = 0; struct contrasts_node *cn = ll_data (cli, struct contrasts_node, ll); - struct ll *coeffi = ll_head (&cn->coefficient_list); + struct ll *coeffi ; tab_text_format (t, 1, c_num + 2, TAB_CENTER, "%d", c_num + 1); - for (count = 0; - count < hsh_count (cmd->group_hash) && coeffi != ll_null (&cn->coefficient_list); - ++count) + for (coeffi = ll_head (&cn->coefficient_list); + coeffi != ll_null (&cn->coefficient_list); + ++count, coeffi = ll_next (coeffi)) { - double *group_value_p; - union value group_value; + const struct categoricals *cats = covariance_get_categoricals (cov); + const union value *val = categoricals_get_value_by_subscript (cats, count); struct string vstr; ds_init_empty (&vstr); - group_value_p = group_values[count]; - group_value.f = *group_value_p; - var_append_value_name (cmd->indep_var, &group_value, &vstr); + var_append_value_name (cmd->indep_var, val, &vstr); - tab_text (t, count + 2, 1, TAB_CENTER | TAT_TITLE, - ds_cstr (&vstr)); + tab_text (t, count + 2, 1, TAB_CENTER | TAT_TITLE, ds_cstr (&vstr)); ds_destroy (&vstr); @@ -937,8 +996,6 @@ show_contrast_coeffs (const struct oneway *cmd) tab_text_format (t, count + 2, c_num + 2, TAB_RIGHT, "%g", coeffn->coeff); } - - coeffi = ll_next (coeffi); } ++c_num; } @@ -949,7 +1006,7 @@ show_contrast_coeffs (const struct oneway *cmd) /* Show the results of the contrast tests */ static void -show_contrast_tests (const struct oneway *cmd) +show_contrast_tests (const struct oneway_spec *cmd, struct oneway_workspace *ws) { int n_contrasts = ll_count (&cmd->contrast_list); size_t v;