X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Flanguage%2Fstats%2Fglm.c;h=6401defdeac8774daaf43d906819fa0d27e1650c;hb=6f1fc2655c5e4aea0b39e5d309a1cc4826f16d4e;hp=2a2cbd54f1520bc93c3ef4e829a71e55bf3baefb;hpb=b609b8bb1f90ef0e98fde74c0c58187456c2f877;p=pspp diff --git a/src/language/stats/glm.c b/src/language/stats/glm.c index 2a2cbd54f1..6401defdea 100644 --- a/src/language/stats/glm.c +++ b/src/language/stats/glm.c @@ -18,6 +18,7 @@ #include #include +#include #include #include "data/case.h" @@ -84,6 +85,37 @@ struct glm_workspace gsl_vector *ssq; }; + +/* Default design: all possible interactions */ +static void +design_full (struct glm_spec *glm) +{ + int sz; + int i = 0; + glm->n_interactions = (1 << glm->n_factor_vars) - 1; + + glm->interactions = xcalloc (glm->n_interactions, sizeof *glm->interactions); + + /* All subsets, with exception of the empty set, of [0, glm->n_factor_vars) */ + for (sz = 1; sz <= glm->n_factor_vars; ++sz) + { + gsl_combination *c = gsl_combination_calloc (glm->n_factor_vars, sz); + + do + { + struct interaction *iact = interaction_create (NULL); + int e; + for (e = 0 ; e < gsl_combination_k (c); ++e) + interaction_add_variable (iact, glm->factor_vars [gsl_combination_get (c, e)]); + + glm->interactions[i++] = iact; + } + while (gsl_combination_next (c) == GSL_SUCCESS); + + gsl_combination_free (c); + } +} + static void output_glm (const struct glm_spec *, const struct glm_workspace *ws); static void run_glm (struct glm_spec *cmd, struct casereader *input, @@ -92,6 +124,8 @@ static void run_glm (struct glm_spec *cmd, struct casereader *input, static bool parse_design_spec (struct lexer *lexer, struct glm_spec *glm); +/* Define to 1 if the /DESIGN subcommand should not be optional */ +#define DESIGN_MANDATORY 1 int cmd_glm (struct lexer *lexer, struct dataset *ds) @@ -248,7 +282,8 @@ cmd_glm (struct lexer *lexer, struct dataset *ds) if (! parse_design_spec (lexer, &glm)) goto error; - + +#if DESIGN_MANDATORY if ( glm.n_interactions == 0) { msg (ME, _("One or more design variables must be given")); @@ -256,6 +291,10 @@ cmd_glm (struct lexer *lexer, struct dataset *ds) } design = true; +#else + if (glm.n_interactions > 0) + design = true; +#endif } else { @@ -266,8 +305,12 @@ cmd_glm (struct lexer *lexer, struct dataset *ds) if ( ! design ) { +#if DESIGN_MANDATORY lex_error (lexer, _("/DESIGN is mandatory in GLM")); goto error; +#endif + + design_full (&glm); } { @@ -309,7 +352,7 @@ static void get_ssq (struct covariance *, gsl_vector *, const struct glm_spec *); static bool -not_dropped (size_t j, size_t * dropped, size_t n_dropped) +not_dropped (size_t j, const size_t *dropped, size_t n_dropped) { size_t i; @@ -322,29 +365,25 @@ not_dropped (size_t j, size_t * dropped, size_t n_dropped) } static void -get_ssq (struct covariance *cov, gsl_vector * ssq, const struct glm_spec *cmd) +get_ssq (struct covariance *cov, gsl_vector *ssq, const struct glm_spec *cmd) { - const struct variable **vars; - gsl_matrix *small_cov = NULL; gsl_matrix *cm = covariance_calculate_unnormalized (cov); size_t i; size_t j; size_t k; - size_t n; - size_t m; - size_t *dropped; - size_t n_dropped; - - dropped = xcalloc (covariance_dim (cov), sizeof (*dropped)); - vars = xcalloc (covariance_dim (cov), sizeof (*vars)); - covariance_get_var_indices (cov, vars); + size_t *dropped = xcalloc (covariance_dim (cov), sizeof (*dropped)); + const struct categoricals *cats = covariance_get_categoricals (cov); for (k = 0; k < cmd->n_interactions; k++) { - n_dropped = 0; - for (i = 1; i < covariance_dim (cov); i++) + size_t n = 0; + size_t m = 0; + gsl_matrix *small_cov = NULL; + size_t n_dropped = 0; + for (i = cmd->n_dep_vars; i < covariance_dim (cov); i++) { - if (vars[i] == cmd->interactions[k]->vars[0]) + if (categoricals_get_interaction_by_subscript (cats, i - cmd->n_dep_vars) + == cmd->interactions[k]) { assert (n_dropped < covariance_dim (cov)); dropped[n_dropped++] = i; @@ -353,8 +392,6 @@ get_ssq (struct covariance *cov, gsl_vector * ssq, const struct glm_spec *cmd) small_cov = gsl_matrix_alloc (cm->size1 - n_dropped, cm->size2 - n_dropped); gsl_matrix_set (small_cov, 0, 0, gsl_matrix_get (cm, 0, 0)); - n = 0; - m = 0; for (i = 0; i < cm->size1; i++) { if (not_dropped (i, dropped, n_dropped)) @@ -380,7 +417,6 @@ get_ssq (struct covariance *cov, gsl_vector * ssq, const struct glm_spec *cmd) } free (dropped); - free (vars); gsl_matrix_free (cm); } @@ -434,8 +470,6 @@ run_glm (struct glm_spec *cmd, struct casereader *input, } casereader_destroy (reader); - categoricals_done (ws.cats); - for (reader = input; (c = casereader_read (reader)) != NULL; case_unref (c)) { @@ -526,8 +560,7 @@ output_glm (const struct glm_spec *cmd, const struct glm_workspace *ws) if (cmd->intercept) df_corr += 1.0; - for (f = 0; f < cmd->n_interactions; ++f) - df_corr += categoricals_n_count (ws->cats, f) - 1.0; + df_corr += categoricals_df_total (ws->cats); mse = gsl_vector_get (ws->ssq, 0) / (n_total - df_corr); @@ -554,7 +587,7 @@ 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_n_count (ws->cats, f) - 1.0; + const double df = categoricals_df (ws->cats, f); const double ssq = gsl_vector_get (ws->ssq, f + 1); const double F = ssq / df / mse; interaction_to_string (cmd->interactions[f], &str); @@ -690,7 +723,7 @@ parse_design_interaction (struct lexer *lexer, struct glm_spec *glm, struct inte if ( lex_match (lexer, T_ASTERISK) || lex_match (lexer, T_BY)) { - // lex_error (lexer, "Interactions are not yet implemented"); return false; + lex_error (lexer, "Interactions are not yet implemented"); return false; return parse_design_interaction (lexer, glm, iact); }