glm.c: Allow parser to accept METHOD=SSTYPE(n) subcommand

[pspp-builds.git] / src / language / stats / glm.c

diff --git a/src/language/stats/glm.c b/src/language/stats/glm.c
index 9d16a149b372766e0eef55a67fe5801e013936b9..4b1b1a8bf3f7e233e44db8519e07da484963c23a 100644 (file)
--- a/src/language/stats/glm.c
+++ b/src/language/stats/glm.c
@@ -65,9 +65,12 @@ struct glm_spec
 
   const struct dictionary *dict;
 
+  int ss_type;
   bool intercept;
 
   double alpha;
+
+  bool dump_coding;
 };
 
 struct glm_workspace
@@ -143,6 +146,8 @@ cmd_glm (struct lexer *lexer, struct dataset *ds)
   glm.intercept = true;
   glm.wv = dict_get_weight (glm.dict);
   glm.alpha = 0.05;
+  glm.dump_coding = false;
+  glm.ss_type = 3;
 
   if (!parse_variables_const (lexer, glm.dict,
                              &glm.dep_vars, &glm.n_dep_vars,
@@ -260,9 +265,10 @@ cmd_glm (struct lexer *lexer, struct dataset *ds)
              goto error;
            }
 
-         if (3 != lex_integer (lexer))
+         glm.ss_type = lex_integer (lexer);
+         if (1 != glm.ss_type  && 2 != glm.ss_type )
            {
-             msg (ME, _("Only type 3 sum of squares are currently implemented"));
+             msg (ME, _("Only types 1 & 2 sum of squares are currently implemented"));
              goto error;
            }
 
@@ -284,6 +290,13 @@ cmd_glm (struct lexer *lexer, struct dataset *ds)
          if (glm.n_interactions > 0)
            design = true;
        }
+      else if (lex_match_id (lexer, "SHOWCODES"))
+       /* Undocumented debug option */
+       {
+         lex_match (lexer, T_EQUALS);
+
+         glm.dump_coding = true;
+       }
       else
        {
          lex_error (lexer, NULL);
@@ -406,8 +419,7 @@ get_ssq (struct covariance *cov, gsl_vector *ssq, const struct glm_spec *cmd)
        }
 
       model_cov = gsl_matrix_alloc (cm->size1 - n_dropped_model, cm->size2 - n_dropped_model);
-      gsl_matrix_set (model_cov, 0, 0, gsl_matrix_get (cm, 0, 0));
-      submodel_cov = gsl_matrix_calloc (cm->size1 - n_dropped_submodel, cm->size2 - n_dropped_submodel);
+      submodel_cov = gsl_matrix_alloc (cm->size1 - n_dropped_submodel, cm->size2 - n_dropped_submodel);
 
       fill_submatrix (cm, model_cov,    model_dropped);
       fill_submatrix (cm, submodel_cov, submodel_dropped);
@@ -478,7 +490,12 @@ run_glm (struct glm_spec *cmd, struct casereader *input,
     }
   casereader_destroy (reader);
 
-  for (reader = input;
+  if (cmd->dump_coding)
+    reader = casereader_clone (input);
+  else
+    reader = input;
+
+  for (;
        (c = casereader_read (reader)) != NULL; case_unref (c))
     {
       double weight = dict_get_case_weight (dict, c, &warn_bad_weight);
@@ -491,6 +508,21 @@ run_glm (struct glm_spec *cmd, struct casereader *input,
     }
   casereader_destroy (reader);
 
+
+  if (cmd->dump_coding)
+    {
+      struct tab_table *t =
+       covariance_dump_enc_header (cov,
+                                   1 + casereader_count_cases (input));
+      for (reader = input;
+          (c = casereader_read (reader)) != NULL; case_unref (c))
+       {
+         covariance_dump_enc (cov, c, t);
+       }
+      casereader_destroy (reader);
+      tab_submit (t);
+    }
+
   {
     gsl_matrix *cm = covariance_calculate_unnormalized (cov);
 
@@ -522,6 +554,15 @@ run_glm (struct glm_spec *cmd, struct casereader *input,
   taint_destroy (taint);
 }
 
+static const char *roman[] = 
+  {
+    "", /* The Romans had no concept of zero */
+    "I",
+    "II",
+    "III",
+    "IV"
+  };
+
 static void
 output_glm (const struct glm_spec *cmd, const struct glm_workspace *ws)
 {
@@ -543,6 +584,7 @@ output_glm (const struct glm_spec *cmd, const struct glm_workspace *ws)
   if (cmd->intercept)
     nr++;
 
+  msg (MW, "GLM is experimental.  Do not rely on these results.");
   t = tab_create (nc, nr);
   tab_title (t, _("Tests of Between-Subjects Effects"));
 
@@ -557,7 +599,8 @@ output_glm (const struct glm_spec *cmd, const struct glm_workspace *ws)
 
   /* TRANSLATORS: The parameter is a roman numeral */
   tab_text_format (t, 1, 0, TAB_CENTER | TAT_TITLE,
-                  _("Type %s Sum of Squares"), "III");
+                  _("Type %s Sum of Squares"), 
+                  roman[cmd->ss_type]);
   tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("df"));
   tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean Square"));
   tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("F"));