const struct dictionary *dict;
+ int ss_type;
bool intercept;
double alpha;
+
+ bool dump_coding;
};
struct glm_workspace
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,
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;
}
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);
static void get_ssq (struct covariance *, gsl_vector *,
const struct glm_spec *);
-static bool
-not_dropped (size_t j, const size_t *dropped, size_t n_dropped)
-{
- size_t i;
-
- for (i = 0; i < n_dropped; i++)
- {
- if (j == dropped[i])
- return false;
- }
- return true;
-}
-
-/*
- Do the variables in X->VARS constitute a proper
- subset of the variables in Y->VARS?
- */
-static bool
-is_subset (const struct interaction *x, const struct interaction *y)
-{
- size_t i;
- size_t j;
- size_t n = 0;
-
- if (x->n_vars < y->n_vars)
- {
- for (i = 0; i < x->n_vars; i++)
- {
- for (j = 0; j < y->n_vars; j++)
- {
- if (x->vars [i] == y->vars [j])
- {
- n++;
- }
- }
- }
- }
- if (n >= x->n_vars)
- return true;
- return false;
-}
-
-static bool
-drop_from_submodel (const struct interaction *x, const struct interaction *y)
+static inline bool
+not_dropped (size_t j, const bool *ff)
{
- size_t i;
- size_t j;
- size_t n = 0;
-
- if (is_subset (x, y))
- return true;
-
- for (i = 0; i < x->n_vars; i++)
- for (j = 0; j < y->n_vars; j++)
- {
- if (x->vars [i] == y->vars [j])
- n++;
- }
- if (n == x->n_vars)
- {
- return true;
- }
-
- return false;
+ return ! ff[j];
}
static void
-fill_submatrix (gsl_matrix * cov, gsl_matrix * submatrix, size_t * dropped,
- size_t n_dropped)
+fill_submatrix (const gsl_matrix * cov, gsl_matrix * submatrix, bool *dropped_f)
{
size_t i;
size_t j;
for (i = 0; i < cov->size1; i++)
{
- if (not_dropped (i, dropped, n_dropped))
+ if (not_dropped (i, dropped_f))
{
m = 0;
for (j = 0; j < cov->size2; j++)
{
- if (not_dropped (j, dropped, n_dropped))
+ if (not_dropped (j, dropped_f))
{
gsl_matrix_set (submatrix, n, m,
gsl_matrix_get (cov, i, j));
gsl_matrix *cm = covariance_calculate_unnormalized (cov);
size_t i;
size_t k;
- size_t *model_dropped = xcalloc (covariance_dim (cov), sizeof (*model_dropped));
- size_t *submodel_dropped = xcalloc (covariance_dim (cov), sizeof (*submodel_dropped));
+ bool *model_dropped = xcalloc (covariance_dim (cov), sizeof (*model_dropped));
+ bool *submodel_dropped = xcalloc (covariance_dim (cov), sizeof (*submodel_dropped));
const struct categoricals *cats = covariance_get_categoricals (cov);
for (k = 0; k < cmd->n_interactions; k++)
{
const struct interaction * x =
categoricals_get_interaction_by_subscript (cats, i - cmd->n_dep_vars);
- if (is_subset (cmd->interactions [k], x))
- {
- assert (n_dropped_model < covariance_dim (cov));
- model_dropped[n_dropped_model++] = i;
- }
- if (drop_from_submodel (cmd->interactions [k], x))
+
+ model_dropped[i] = false;
+ submodel_dropped[i] = false;
+ if (interaction_is_subset (cmd->interactions [k], x))
{
assert (n_dropped_submodel < covariance_dim (cov));
- submodel_dropped[n_dropped_submodel++] = i;
+ n_dropped_submodel++;
+ submodel_dropped[i] = true;
+
+ if ( cmd->interactions [k]->n_vars < x->n_vars)
+ {
+ assert (n_dropped_model < covariance_dim (cov));
+ n_dropped_model++;
+ model_dropped[i] = true;
+ }
}
}
- 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);
- fill_submatrix (cm, model_cov, model_dropped, n_dropped_model);
- fill_submatrix (cm, submodel_cov, submodel_dropped, n_dropped_submodel);
+
+ model_cov = gsl_matrix_alloc (cm->size1 - n_dropped_model, cm->size2 - n_dropped_model);
+ 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);
reg_sweep (model_cov, 0);
reg_sweep (submodel_cov, 0);
+
gsl_vector_set (ssq, k + 1,
- gsl_matrix_get (submodel_cov, 0, 0)
- - gsl_matrix_get (model_cov, 0, 0));
+ gsl_matrix_get (submodel_cov, 0, 0) - gsl_matrix_get (model_cov, 0, 0)
+ );
+
gsl_matrix_free (model_cov);
gsl_matrix_free (submodel_cov);
}
}
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);
}
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);
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)
{
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"));
/* 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"));