X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Flanguage%2Fstats%2Fglm.c;h=1ded71e9f1c6b69db56b6278a4ce8118bb6c5049;hb=897c1977501bad5b68f7cc6248a8b950fe15048a;hp=d01d41805758705adae7cf6f94f5ffcf20cc5c17;hpb=bae3da89755d9503a674712303117dd76311b267;p=pspp diff --git a/src/language/stats/glm.c b/src/language/stats/glm.c index d01d418057..1ded71e9f1 100644 --- a/src/language/stats/glm.c +++ b/src/language/stats/glm.c @@ -1,5 +1,5 @@ /* PSPP - a program for statistical analysis. - Copyright (C) 2010, 2011 Free Software Foundation, Inc. + Copyright (C) 2010, 2011, 2012 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 @@ -324,8 +324,6 @@ cmd_glm (struct lexer *lexer, struct dataset *ds) const_var_set_destroy (factors); free (glm.factor_vars); - for (i = 0 ; i < glm.n_interactions; ++i) - interaction_destroy (glm.interactions[i]); free (glm.interactions); free (glm.dep_vars); @@ -590,8 +588,7 @@ run_glm (struct glm_spec *cmd, struct casereader *input, struct covariance *cov; ws.cats = categoricals_create (cmd->interactions, cmd->n_interactions, - cmd->wv, cmd->exclude, - NULL, NULL, NULL, NULL); + cmd->wv, cmd->exclude, MV_ANY); cov = covariance_2pass_create (cmd->n_dep_vars, cmd->dep_vars, ws.cats, cmd->wv, cmd->exclude); @@ -716,8 +713,10 @@ output_glm (const struct glm_spec *cmd, const struct glm_workspace *ws) const struct fmt_spec *wfmt = cmd->wv ? var_get_print_format (cmd->wv) : &F_8_0; + double intercept_ssq; + double ssq_effects; double n_total, mean; - double df_corr = 0.0; + double df_corr = 1.0; double mse = 0; int f; @@ -727,9 +726,9 @@ output_glm (const struct glm_spec *cmd, const struct glm_workspace *ws) struct tab_table *t; const int nc = 6; - int nr = heading_rows + 4 + cmd->n_interactions; + int nr = heading_rows + 3 + cmd->n_interactions; if (cmd->intercept) - nr++; + nr += 2; msg (MW, "GLM is experimental. Do not rely on these results."); t = tab_create (nc, nr); @@ -755,27 +754,29 @@ output_glm (const struct glm_spec *cmd, const struct glm_workspace *ws) moments_calculate (ws->totals, &n_total, &mean, NULL, NULL, NULL); - if (cmd->intercept) - df_corr += 1.0; - df_corr += categoricals_df_total (ws->cats); - mse = gsl_vector_get (ws->ssq, 0) / (n_total - df_corr); - r = heading_rows; - tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Corrected Model")); + if (cmd->intercept) + tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Corrected Model")); + else + tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Model")); r++; + mse = gsl_vector_get (ws->ssq, 0) / (n_total - df_corr); + + intercept_ssq = pow2 (mean * n_total) / n_total; + + ssq_effects = 0.0; if (cmd->intercept) { - const double intercept = pow2 (mean * n_total) / n_total; const double df = 1.0; - const double F = intercept / df / mse; + const double F = intercept_ssq / df / mse; tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Intercept")); - tab_double (t, 1, r, 0, intercept, NULL); + tab_double (t, 1, r, 0, intercept_ssq, NULL); tab_double (t, 2, r, 0, 1.00, wfmt); - tab_double (t, 3, r, 0, intercept / df, NULL); + tab_double (t, 3, r, 0, intercept_ssq / df, NULL); tab_double (t, 4, r, 0, F, NULL); tab_double (t, 5, r, 0, gsl_cdf_fdist_Q (F, df, n_total - df_corr), NULL); @@ -785,9 +786,20 @@ output_glm (const struct glm_spec *cmd, const struct glm_workspace *ws) for (f = 0; f < cmd->n_interactions; ++f) { struct string str = DS_EMPTY_INITIALIZER; - const double df = categoricals_df (ws->cats, f); - const double ssq = gsl_vector_get (ws->ssq, f + 1); - const double F = ssq / df / mse; + double df = categoricals_df (ws->cats, f); + + double ssq = gsl_vector_get (ws->ssq, f + 1); + double F; + + ssq_effects += ssq; + + if (! cmd->intercept) + { + df++; + ssq += intercept_ssq; + } + + F = ssq / df / mse; interaction_to_string (cmd->interactions[f], &str); tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, ds_cstr (&str)); ds_destroy (&str); @@ -803,10 +815,17 @@ output_glm (const struct glm_spec *cmd, const struct glm_workspace *ws) } { - /* Corrected Model */ - const double df = df_corr - 1.0; - const double ssq = ws->total_ssq - gsl_vector_get (ws->ssq, 0); - const double F = ssq / df / mse; + /* Model / Corrected Model */ + double df = df_corr; + double ssq = ws->total_ssq - gsl_vector_get (ws->ssq, 0); + double F; + + if ( cmd->intercept ) + df --; + else + ssq += intercept_ssq; + + F = ssq / df / mse; tab_double (t, 1, heading_rows, 0, ssq, NULL); tab_double (t, 2, heading_rows, 0, df, wfmt); tab_double (t, 3, heading_rows, 0, ssq / df, NULL); @@ -826,24 +845,21 @@ output_glm (const struct glm_spec *cmd, const struct glm_workspace *ws) tab_double (t, 3, r++, 0, mse, NULL); } + { + tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Total")); + tab_double (t, 1, r, 0, ws->total_ssq + intercept_ssq, NULL); + tab_double (t, 2, r, 0, n_total, wfmt); + + r++; + } + if (cmd->intercept) { - const double intercept = pow2 (mean * n_total) / n_total; - const double ssq = intercept + ws->total_ssq; - - tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Total")); - tab_double (t, 1, r, 0, ssq, NULL); - tab_double (t, 2, r, 0, n_total, wfmt); - - r++; + tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Corrected Total")); + tab_double (t, 1, r, 0, ws->total_ssq, NULL); + tab_double (t, 2, r, 0, n_total - 1.0, wfmt); } - tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Corrected Total")); - - - tab_double (t, 1, r, 0, ws->total_ssq, NULL); - tab_double (t, 2, r, 0, n_total - 1.0, wfmt); - tab_submit (t); } @@ -867,71 +883,11 @@ dump_matrix (const gsl_matrix * m) - -/* Match a variable. - If the match succeeds, the variable will be placed in VAR. - Returns true if successful */ -static bool -lex_match_variable (struct lexer *lexer, const struct glm_spec *glm, const struct variable **var) -{ - if (lex_token (lexer) != T_ID) - return false; - - *var = parse_variable_const (lexer, glm->dict); - - if ( *var == NULL) - return false; - return true; -} - -/* An interaction is a variable followed by {*, BY} followed by an interaction */ -static bool -parse_design_interaction (struct lexer *lexer, struct glm_spec *glm, struct interaction **iact) -{ - const struct variable *v = NULL; - assert (iact); - - switch (lex_next_token (lexer, 1)) - { - case T_ENDCMD: - case T_SLASH: - case T_COMMA: - case T_ID: - case T_BY: - case T_ASTERISK: - break; - default: - return false; - break; - } - - if (! lex_match_variable (lexer, glm, &v)) - { - interaction_destroy (*iact); - *iact = NULL; - return false; - } - - assert (v); - - if ( *iact == NULL) - *iact = interaction_create (v); - else - interaction_add_variable (*iact, v); - - if ( lex_match (lexer, T_ASTERISK) || lex_match (lexer, T_BY)) - { - return parse_design_interaction (lexer, glm, iact); - } - - return true; -} - static bool parse_nested_variable (struct lexer *lexer, struct glm_spec *glm) { const struct variable *v = NULL; - if ( ! lex_match_variable (lexer, glm, &v)) + if ( ! lex_match_variable (lexer, glm->dict, &v)) return false; if (lex_match (lexer, T_LPAREN)) @@ -952,7 +908,7 @@ static bool parse_design_term (struct lexer *lexer, struct glm_spec *glm) { struct interaction *iact = NULL; - if (parse_design_interaction (lexer, glm, &iact)) + if (parse_design_interaction (lexer, glm->dict, &iact)) { /* Interaction parsing successful. Add to list of interactions */ glm->interactions = xrealloc (glm->interactions, sizeof *glm->interactions * ++glm->n_interactions);