#include <math/design-matrix.h>
#include <data/dictionary.h>
#include <libpspp/message.h>
-#include <data/file-handle-def.h>
+#include <language/data-io/file-handle.h>
#include "gettext.h"
#include <language/lexer/lexer.h>
#include <math/linreg/linreg.h>
#include <math/linreg/coefficient.h>
#include <data/missing-values.h>
-#include <language/stats/regression-export.h>
+#include "regression-export.h"
#include <output/table.h>
#include <data/value-labels.h>
#include <data/variable.h>
/* (specification)
"REGRESSION" (regression_):
- *variables=varlist;
+ *variables=custom;
statistics[st_]=r,
coeff,
anova,
all;
export=custom;
^dependent=varlist;
+ save[sv_]=resid,pred;
method=enter.
*/
/* (declarations) */
/* (functions) */
static struct cmd_regression cmd;
+/* Linear regression models. */
+pspp_linreg_cache **models = NULL;
+
+/*
+ Transformations for saving predicted values
+ and residuals, etc.
+ */
+struct reg_trns
+{
+ int n_trns; /* Number of transformations. */
+ int trns_id; /* Which trns is this one? */
+ pspp_linreg_cache *c; /* Linear model for this trns. */
+};
+/*
+ Variables used (both explanatory and response).
+ */
+static struct variable **v_variables;
+
/*
- Array holding the subscripts of the independent variables.
+ Number of variables.
*/
-size_t *indep_vars;
+static size_t n_variables;
/*
File where the model will be saved if the EXPORT subcommand
tab_float (t, 2, 1, TAB_RIGHT, rsq, 10, 2);
tab_float (t, 3, 1, TAB_RIGHT, adjrsq, 10, 2);
tab_float (t, 4, 1, TAB_RIGHT, std_error, 10, 2);
- tab_title (t, 0, _("Model Summary"));
+ tab_title (t, _("Model Summary"));
tab_submit (t);
}
static void
reg_stats_coeff (pspp_linreg_cache * c)
{
- size_t i;
size_t j;
int n_cols = 7;
int n_rows;
tab_float (t, 6, 1, 0, pval, 10, 2);
for (j = 1; j <= c->n_indeps; j++)
{
- i = indep_vars[j];
v = pspp_linreg_coeff_get_var (c->coeff + j, 0);
label = var_to_string (v);
/* Do not overwrite the variable's name. */
pval = 2 * gsl_cdf_tdist_Q (fabs (t_stat), 1.0);
tab_float (t, 6, j + 1, 0, pval, 10, 2);
}
- tab_title (t, 0, _("Coefficients"));
+ tab_title (t, _("Coefficients"));
tab_submit (t);
free (tmp);
}
tab_float (t, 6, 1, 0, pval, 8, 3);
- tab_title (t, 0, _("ANOVA"));
+ tab_title (t, _("ANOVA"));
tab_submit (t);
}
static void
int n_cols;
int n_rows;
int i;
- int j;
int k;
int row;
int col;
tab_vline (t, TAL_0, 1, 0, 0);
tab_text (t, 0, 0, TAB_CENTER | TAT_TITLE, _("Model"));
tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Covariances"));
- for (i = 1; i < c->n_indeps + 1; i++)
+ for (i = 1; i < c->n_coeffs; i++)
{
- j = indep_vars[(i - 1)];
- struct variable *v = cmd.v_variables[j];
+ const struct variable *v = pspp_linreg_coeff_get_var (c->coeff + i, 0);
label = var_to_string (v);
tab_text (t, 2, i, TAB_CENTER, label);
tab_text (t, i + 2, 0, TAB_CENTER, label);
- for (k = 1; k < c->n_indeps + 1; k++)
+ for (k = 1; k < c->n_coeffs; k++)
{
col = (i <= k) ? k : i;
row = (i <= k) ? i : k;
gsl_matrix_get (c->cov, row, col), 8, 3);
}
}
- tab_title (t, 0, _("Coefficient Correlations"));
+ tab_title (t, _("Coefficient Correlations"));
tab_submit (t);
}
static void
statistics_keyword_output (reg_stats_tol, keywords[tol], c);
statistics_keyword_output (reg_stats_selection, keywords[selection], c);
}
+/*
+ Free the transformation. Free its linear model if this
+ transformation is the last one.
+ */
+static
+bool regression_trns_free (void *t_)
+{
+ bool result = true;
+ struct reg_trns *t = t_;
+
+ if (t->trns_id == t->n_trns)
+ {
+ result = pspp_linreg_cache_free (t->c);
+ }
+ free (t);
+
+ return result;
+}
+/*
+ Gets the predicted values.
+ */
+static int
+regression_trns_pred_proc (void *t_, struct ccase *c, int case_idx UNUSED)
+{
+ size_t i;
+ size_t n_vars;
+ size_t n_vals = 0;
+ union value *tmp = NULL;
+ struct reg_trns *t = t_;
+ pspp_linreg_cache *model;
+ union value *output;
+ const union value **vals = NULL;
+ struct variable **vars = NULL;
+ struct variable **model_vars = NULL;
+
+ assert (t != NULL);
+ model = t->c;
+ assert (model != NULL);
+ assert (model->depvar != NULL);
+ assert (model->pred != NULL);
+
+ dict_get_vars (default_dict, &vars, &n_vars, 1u << DC_SYSTEM);
+ vals = xnmalloc (n_vars, sizeof (*vals));
+ model_vars = xnmalloc (n_vars, sizeof (*model_vars));
+ assert (vals != NULL);
+ output = case_data_rw (c, model->pred->fv);
+ assert (output != NULL);
+
+ for (i = 0; i < n_vars; i++)
+ {
+ /* Use neither the predicted values nor the dependent variable. */
+ if (vars[i]->index != model->pred->index &&
+ vars[i]->index != model->depvar->index)
+ {
+ if (vars[i]->type == ALPHA && vars[i]->obs_vals != NULL)
+ {
+ tmp = vars[i]->obs_vals->vals;
+ }
+ else
+ {
+ tmp = NULL;
+ }
+ /*
+ Make sure the variable we use is in the linear model.
+ */
+ if (pspp_linreg_get_coeff (model, vars[i], tmp) != NULL)
+ {
+ vals[n_vals] = case_data (c, i);
+ model_vars[n_vals] = vars[i];
+ n_vals++;
+ }
+ }
+ }
+ output->f = (*model->predict) ((const struct variable **) vars,
+ vals, model, n_vals);
+ free (vals);
+ free (model_vars);
+ return TRNS_CONTINUE;
+}
+/*
+ Gets the residuals.
+ */
+static int
+regression_trns_resid_proc (void *t_, struct ccase *c, int case_idx UNUSED)
+{
+ size_t i;
+ size_t n_vars;
+ size_t n_vals = 0;
+ struct reg_trns *t = t_;
+ pspp_linreg_cache *model;
+ union value *output;
+ union value *tmp;
+ const union value **vals = NULL;
+ const union value *obs = NULL;
+ struct variable **vars = NULL;
+ struct variable **model_vars = NULL;
+
+ assert (t!= NULL);
+ model = t->c;
+ assert (model != NULL);
+ assert (model->depvar != NULL);
+ assert (model->resid != NULL);
+
+ dict_get_vars (default_dict, &vars, &n_vars, 1u << DC_SYSTEM);
+ vals = xnmalloc (n_vars, sizeof (*vals));
+ model_vars = xnmalloc (n_vars, sizeof (*model_vars));
+ assert (vals != NULL);
+ output = case_data_rw (c, model->resid->fv);
+ assert (output != NULL);
+
+ for (i = 0; i < n_vars; i++)
+ {
+ /* Use neither the predicted values nor the dependent variable. */
+ if (vars[i]->index != model->resid->index &&
+ vars[i]->index != model->depvar->index)
+ {
+ if (vars[i]->type == ALPHA && vars[i]->obs_vals != NULL)
+ {
+ tmp = vars[i]->obs_vals->vals;
+ }
+ else
+ {
+ tmp = NULL;
+ }
+ /*
+ Make sure the variable we use is in the linear model.
+ */
+ if (pspp_linreg_get_coeff (model, vars[i], tmp) != NULL)
+ {
+ vals[n_vals] = case_data (c, i);
+ model_vars[n_vals] = vars[i];
+ n_vals++;
+ }
+ }
+ if (vars[i]->index == model->depvar->index)
+ {
+ obs = case_data (c, i);
+ assert (obs != NULL);
+ }
+ }
+ output->f = (*model->residual) ((const struct variable **) vars,
+ vals, obs, model, n_vals);
+ free (vals);
+ free (model_vars);
+ return TRNS_CONTINUE;
+}
+/*
+ Returns 0 if NAME is a duplicate of any existing variable name.
+*/
+static int
+try_name (char *name)
+{
+ if (dict_lookup_var (default_dict, name) != NULL)
+ return 0;
+
+ return 1;
+}
+static
+void reg_get_name (char name[LONG_NAME_LEN], const char prefix[LONG_NAME_LEN])
+{
+ int i = 1;
+
+ snprintf (name, LONG_NAME_LEN, "%s%d", prefix, i);
+ while (!try_name(name))
+ {
+ i++;
+ snprintf (name, LONG_NAME_LEN, "%s%d", prefix, i);
+ }
+}
+static void
+reg_save_var (const char *prefix, trns_proc_func *f,
+ pspp_linreg_cache *c, struct variable **v,
+ int n_trns)
+{
+ static int trns_index = 1;
+ char name[LONG_NAME_LEN + 1];
+ struct variable *new_var;
+ struct reg_trns *t = NULL;
+
+ t = xmalloc (sizeof (*t));
+ assert (t != NULL);
+ t->trns_id = trns_index;
+ t->n_trns = n_trns;
+ t->c = c;
+ reg_get_name (name, prefix);
+ new_var = dict_create_var (default_dict, name, 0);
+ assert (new_var != NULL);
+ *v = new_var;
+ add_transformation (f, regression_trns_free, t);
+ trns_index++;
+}
+static void
+subcommand_save (int save, pspp_linreg_cache **models)
+{
+ pspp_linreg_cache **lc;
+ int n_trns = 0;
+ int i;
+
+ assert (models != NULL);
+
+ if (save)
+ {
+ /* Count the number of transformations we will need. */
+ for (i = 0; i < REGRESSION_SV_count; i++)
+ {
+ if (cmd.a_save[i])
+ {
+ n_trns++;
+ }
+ }
+ n_trns *= cmd.n_dependent;
+
+ for (lc = models; lc < models + cmd.n_dependent; lc++)
+ {
+ assert (*lc != NULL);
+ assert ((*lc)->depvar != NULL);
+ if (cmd.a_save[REGRESSION_SV_RESID])
+ {
+ reg_save_var ("RES", regression_trns_resid_proc, *lc, &(*lc)->resid, n_trns);
+ }
+ if (cmd.a_save[REGRESSION_SV_PRED])
+ {
+ reg_save_var ("PRED", regression_trns_pred_proc, *lc, &(*lc)->pred, n_trns);
+ }
+ }
+ }
+ else
+ {
+ for (lc = models; lc < models + cmd.n_dependent; lc++)
+ {
+ assert (*lc != NULL);
+ pspp_linreg_cache_free (*lc);
+ }
+ }
+}
static int
reg_inserted (const struct variable *v, struct variable **varlist, int n_vars)
{
"if (strncmp (v_name, model_depvars[i], PSPP_REG_MAXLEN) == 0)\n\t\t{\n\t\t\t");
fprintf (fp, "return i;\n\t\t}\n\t}\n}\n");
}
+static int
+reg_has_categorical (pspp_linreg_cache * c)
+{
+ int i;
+ const struct variable *v;
+
+ for (i = 1; i < c->n_coeffs; i++)
+ {
+ v = pspp_linreg_coeff_get_var (c->coeff + i, 0);
+ if (v->type == ALPHA)
+ {
+ return 1;
+ }
+ }
+ return 0;
+}
+
static void
subcommand_export (int export, pspp_linreg_cache * c)
{
+ FILE *fp;
size_t i;
size_t j;
int n_quantiles = 100;
if (export)
{
- FILE *fp;
assert (c != NULL);
assert (model_file != NULL);
+ fp = fopen (fh_get_file_name (model_file), "w");
assert (fp != NULL);
- fp = fopen (fh_get_filename (model_file), "w");
fprintf (fp, "%s", reg_preamble);
reg_print_getvar (fp, c);
- reg_print_categorical_encoding (fp, c);
+ if (reg_has_categorical (c))
+ {
+ reg_print_categorical_encoding (fp, c);
+ }
fprintf (fp, "%s", reg_export_t_quantiles_1);
increment = 0.5 / (double) increment;
for (i = 0; i < n_quantiles - 1; i++)
}
}
static int
-regression_custom_export (struct cmd_regression *cmd)
+regression_custom_export (struct cmd_regression *cmd UNUSED)
{
/* 0 on failure, 1 on success, 2 on failure that should result in syntax error */
if (!lex_force_match ('('))
{
if (!parse_regression (&cmd))
return CMD_FAILURE;
+
+ models = xnmalloc (cmd.n_dependent, sizeof *models);
if (!multipass_procedure_with_splits (run_regression, &cmd))
return CMD_CASCADING_FAILURE;
-
+ subcommand_save (cmd.sbc_save, models);
+ free (v_variables);
+ free (models);
return pspp_reg_rc;
}
/*
- Is variable k one of the dependent variables?
+ Is variable k the dependent variable?
*/
static int
-is_depvar (size_t k)
+is_depvar (size_t k, const struct variable *v)
{
- size_t j = 0;
- for (j = 0; j < cmd.n_dependent; j++)
- {
- /*
- compare_var_names returns 0 if the variable
- names match.
- */
- if (!compare_var_names (cmd.v_dependent[j], cmd.v_variables[k], NULL))
- return 1;
- }
+ /*
+ compare_var_names returns 0 if the variable
+ names match.
+ */
+ if (!compare_var_names (v, v_variables[k], NULL))
+ return 1;
+
return 0;
}
return n_data;
}
+/* Parser for the variables sub command */
+static int
+regression_custom_variables(struct cmd_regression *cmd UNUSED)
+{
+
+ lex_match('=');
+
+ if ((token != T_ID || dict_lookup_var (default_dict, tokid) == NULL)
+ && token != T_ALL)
+ return 2;
+
+
+ if (!parse_variables (default_dict, &v_variables, &n_variables,
+ PV_NONE ))
+ {
+ free (v_variables);
+ return 0;
+ }
+ assert(n_variables);
+
+ return 1;
+}
+/*
+ Count the explanatory variables. The user may or may
+ not have specified a response variable in the syntax.
+ */
+static
+int get_n_indep (const struct variable *v)
+{
+ int result;
+ int i = 0;
+
+ result = n_variables;
+ while (i < n_variables)
+ {
+ if (is_depvar (i, v))
+ {
+ result--;
+ i = n_variables;
+ }
+ i++;
+ }
+ return result;
+}
+/*
+ Read from the active file. Identify the explanatory variables in
+ v_variables. Encode categorical variables. Drop cases with missing
+ values.
+*/
+static
+int prepare_data (int n_data, int is_missing_case[],
+ struct variable **indep_vars,
+ struct variable *depvar,
+ const struct casefile *cf)
+{
+ int i;
+ int j;
+
+ assert (indep_vars != NULL);
+ j = 0;
+ for (i = 0; i < n_variables; i++)
+ {
+ if (!is_depvar (i, depvar))
+ {
+ indep_vars[j] = v_variables[i];
+ j++;
+ if (v_variables[i]->type == ALPHA)
+ {
+ /* Make a place to hold the binary vectors
+ corresponding to this variable's values. */
+ cat_stored_values_create (v_variables[i]);
+ }
+ n_data = mark_missing_cases (cf, v_variables[i], is_missing_case, n_data);
+ }
+ }
+ /*
+ Mark missing cases for the dependent variable.
+ */
+ n_data = mark_missing_cases (cf, depvar, is_missing_case, n_data);
+
+ return n_data;
+}
static bool
run_regression (const struct casefile *cf, void *cmd_ UNUSED)
{
size_t i;
- size_t n_data = 0;
+ size_t n_data = 0; /* Number of valide cases. */
+ size_t n_cases; /* Number of cases. */
size_t row;
size_t case_num;
- int n_indep;
- int j = 0;
+ int n_indep = 0;
int k;
/*
Keep track of the missing cases.
const union value *val;
struct casereader *r;
struct ccase c;
- struct variable *v;
- struct variable *depvar;
struct variable **indep_vars;
struct design_matrix *X;
gsl_vector *Y;
- pspp_linreg_cache *lcache;
+
pspp_linreg_opts lopts;
- n_data = casefile_get_case_cnt (cf);
+ assert (models != NULL);
+ if (!v_variables)
+ {
+ dict_get_vars (default_dict, &v_variables, &n_variables,
+ 1u << DC_SYSTEM);
+ }
+
+ n_cases = casefile_get_case_cnt (cf);
for (i = 0; i < cmd.n_dependent; i++)
{
}
}
- is_missing_case = xnmalloc (n_data, sizeof (*is_missing_case));
- for (i = 0; i < n_data; i++)
- is_missing_case[i] = 0;
-
- n_indep = cmd.n_variables - cmd.n_dependent;
- indep_vars = xnmalloc (n_indep, sizeof *indep_vars);
+ is_missing_case = xnmalloc (n_cases, sizeof (*is_missing_case));
lopts.get_depvar_mean_std = 1;
- lopts.get_indep_mean_std = xnmalloc (n_indep, sizeof (int));
- /*
- Read from the active file. The first pass encodes categorical
- variables and drops cases with missing values.
- */
- j = 0;
- for (i = 0; i < cmd.n_variables; i++)
+ for (k = 0; k < cmd.n_dependent; k++)
{
- if (!is_depvar (i))
+ n_indep = get_n_indep ((const struct variable *) cmd.v_dependent[k]);
+ lopts.get_indep_mean_std = xnmalloc (n_indep, sizeof (int));
+ indep_vars = xnmalloc (n_indep, sizeof *indep_vars);
+ assert (indep_vars != NULL);
+
+ for (i = 0; i < n_cases; i++)
{
- v = cmd.v_variables[i];
- indep_vars[j] = v;
- j++;
- if (v->type == ALPHA)
- {
- /* Make a place to hold the binary vectors
- corresponding to this variable's values. */
- cat_stored_values_create (v);
- }
- n_data = mark_missing_cases (cf, v, is_missing_case, n_data);
+ is_missing_case[i] = 0;
}
- }
-
- /*
- Drop cases with missing values for any dependent variable.
- */
- j = 0;
- for (i = 0; i < cmd.n_dependent; i++)
- {
- v = cmd.v_dependent[i];
- j++;
- n_data = mark_missing_cases (cf, v, is_missing_case, n_data);
- }
-
- for (k = 0; k < cmd.n_dependent; k++)
- {
- depvar = cmd.v_dependent[k];
+ n_data = prepare_data (n_cases, is_missing_case, indep_vars,
+ cmd.v_dependent[k],
+ (const struct casefile *) cf);
Y = gsl_vector_alloc (n_data);
X =
{
lopts.get_indep_mean_std[i] = 1;
}
- lcache = pspp_linreg_cache_alloc (X->m->size1, X->m->size2);
- lcache->indep_means = gsl_vector_alloc (X->m->size2);
- lcache->indep_std = gsl_vector_alloc (X->m->size2);
- lcache->depvar = (const struct variable *) depvar;
+ models[k] = pspp_linreg_cache_alloc (X->m->size1, X->m->size2);
+ models[k]->indep_means = gsl_vector_alloc (X->m->size2);
+ models[k]->indep_std = gsl_vector_alloc (X->m->size2);
+ models[k]->depvar = (const struct variable *) cmd.v_dependent[k];
/*
For large data sets, use QR decomposition.
*/
if (n_data > sqrt (n_indep) && n_data > REG_LARGE_DATA)
{
- lcache->method = PSPP_LINREG_SVD;
+ models[k]->method = PSPP_LINREG_SVD;
}
/*
- The second pass creates the design matrix.
+ The second pass fills the design matrix.
*/
row = 0;
for (r = casefile_get_reader (cf); casereader_read (r, &c);
case_num = casereader_cnum (r) - 1;
if (!is_missing_case[case_num])
{
- for (i = 0; i < cmd.n_variables; ++i) /* Iterate over the variables
- for the current case.
- */
+ for (i = 0; i < n_variables; ++i) /* Iterate over the
+ variables for the
+ current case.
+ */
{
- v = cmd.v_variables[i];
- val = case_data (&c, v->fv);
+ val = case_data (&c, v_variables[i]->fv);
/*
Independent/dependent variable separation. The
'variables' subcommand specifies a varlist which contains
subcommand, and maybe also in the 'variables' subcommand.
We need to separate the two.
*/
- if (!is_depvar (i))
+ if (!is_depvar (i, cmd.v_dependent[k]))
{
- if (v->type == ALPHA)
+ if (v_variables[i]->type == ALPHA)
{
- design_matrix_set_categorical (X, row, v, val);
+ design_matrix_set_categorical (X, row, v_variables[i], val);
}
- else if (v->type == NUMERIC)
+ else if (v_variables[i]->type == NUMERIC)
{
- design_matrix_set_numeric (X, row, v, val);
+ design_matrix_set_numeric (X, row, v_variables[i], val);
}
}
}
- val = case_data (&c, depvar->fv);
+ val = case_data (&c, cmd.v_dependent[k]->fv);
gsl_vector_set (Y, row, val->f);
row++;
}
and store pointers to the variables that correspond to the
coefficients.
*/
- pspp_linreg_coeff_init (lcache, X);
+ pspp_linreg_coeff_init (models[k], X);
/*
Find the least-squares estimates and other statistics.
*/
- pspp_linreg ((const gsl_vector *) Y, X->m, &lopts, lcache);
- subcommand_statistics (cmd.a_statistics, lcache);
- subcommand_export (cmd.sbc_export, lcache);
+ pspp_linreg ((const gsl_vector *) Y, X->m, &lopts, models[k]);
+ subcommand_statistics (cmd.a_statistics, models[k]);
+ subcommand_export (cmd.sbc_export, models[k]);
+
gsl_vector_free (Y);
design_matrix_destroy (X);
- pspp_linreg_cache_free (lcache);
+ free (indep_vars);
free (lopts.get_indep_mean_std);
casereader_destroy (r);
}
- free (indep_vars);
+
free (is_missing_case);
return true;