/* PSPP - linear regression.
Copyright (C) 2005 Free Software Foundation, Inc.
- Written by Jason H Stover <jason@sakla.net>.
This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License as
#include <math/design-matrix.h>
#include <math/coefficient.h>
#include <math/linreg/linreg.h>
+#include <math/moments.h>
#include <output/table.h>
#include "gettext.h"
/* (functions) */
static struct cmd_regression cmd;
+/*
+ Moments for each of the variables used.
+ */
+struct moments_var
+{
+ struct moments1 *m;
+ const struct variable *v;
+};
+
/* Linear regression models. */
-pspp_linreg_cache **models = NULL;
+static pspp_linreg_cache **models = NULL;
/*
Transformations for saving predicted values
/*
Variables used (both explanatory and response).
*/
-static struct variable **v_variables;
+static const struct variable **v_variables;
/*
Number of variables.
File where the model will be saved if the EXPORT subcommand
is given.
*/
-struct file_handle *model_file;
+static struct file_handle *model_file;
/*
Return value for the procedure.
*/
-int pspp_reg_rc = CMD_SUCCESS;
+static int pspp_reg_rc = CMD_SUCCESS;
static bool run_regression (const struct ccase *,
- const struct casefile *, void *);
+ const struct casefile *, void *,
+ const struct dataset *);
/*
STATISTICS subcommand output functions.
label = var_to_string (v);
/* Do not overwrite the variable's name. */
strncpy (tmp, label, MAX_STRING);
- if (v->type == ALPHA)
+ if (var_is_alpha (v))
{
/*
Append the value associated with this coefficient.
*/
val = pspp_coeff_get_value (c->coeff[j], v);
- val_s = value_to_string (val, v);
+ val_s = var_get_value_name (v, val);
strncat (tmp, val_s, MAX_STRING);
}
/*
P values for the test statistic above.
*/
- pval = 2 * gsl_cdf_tdist_Q (fabs (t_stat), 1.0);
+ pval =
+ 2 * gsl_cdf_tdist_Q (fabs (t_stat),
+ (double) (c->n_obs - c->n_coeffs));
tab_float (t, 6, j + 1, 0, pval, 10, 2);
}
tab_title (t, _("Coefficients"));
Gets the predicted values.
*/
static int
-regression_trns_pred_proc (void *t_, struct ccase *c, int case_idx UNUSED)
+regression_trns_pred_proc (void *t_, struct ccase *c,
+ casenumber case_idx UNUSED)
{
size_t i;
size_t n_vals;
n_vals = (*model->get_vars) (model, vars);
vals = xnmalloc (n_vals, sizeof (*vals));
- output = case_data_rw (c, model->pred->fv);
+ output = case_data_rw (c, model->pred);
assert (output != NULL);
for (i = 0; i < n_vals; i++)
{
- vals[i] = case_data (c, vars[i]->fv);
+ vals[i] = case_data (c, vars[i]);
}
output->f = (*model->predict) ((const struct variable **) vars,
vals, model, n_vals);
Gets the residuals.
*/
static int
-regression_trns_resid_proc (void *t_, struct ccase *c, int case_idx UNUSED)
+regression_trns_resid_proc (void *t_, struct ccase *c,
+ casenumber case_idx UNUSED)
{
size_t i;
size_t n_vals;
n_vals = (*model->get_vars) (model, vars);
vals = xnmalloc (n_vals, sizeof (*vals));
- output = case_data_rw (c, model->resid->fv);
+ output = case_data_rw (c, model->resid);
assert (output != NULL);
for (i = 0; i < n_vals; i++)
{
- vals[i] = case_data (c, vars[i]->fv);
+ vals[i] = case_data (c, vars[i]);
}
- obs = case_data (c, model->depvar->fv);
+ obs = case_data (c, model->depvar);
output->f = (*model->residual) ((const struct variable **) vars,
vals, obs, model, n_vals);
free (vals);
}
/*
- Returns 0 if NAME is a duplicate of any existing variable name.
+ Returns false if NAME is a duplicate of any existing variable name.
*/
-static int
-try_name (char *name)
+static bool
+try_name (const struct dictionary *dict, const char *name)
{
- if (dict_lookup_var (default_dict, name) != NULL)
- return 0;
+ if (dict_lookup_var (dict, name) != NULL)
+ return false;
- return 1;
+ return true;
}
+
static void
-reg_get_name (char name[LONG_NAME_LEN], const char prefix[LONG_NAME_LEN])
+reg_get_name (const struct dictionary *dict, 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))
+ while (!try_name (dict, name))
{
i++;
snprintf (name, LONG_NAME_LEN, "%s%d", prefix, i);
}
}
+
static void
-reg_save_var (const char *prefix, trns_proc_func * f,
+reg_save_var (struct dataset *ds, const char *prefix, trns_proc_func * f,
pspp_linreg_cache * c, struct variable **v, int n_trns)
{
+ struct dictionary *dict = dataset_dict (ds);
static int trns_index = 1;
char name[LONG_NAME_LEN];
struct variable *new_var;
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);
+ reg_get_name (dict, name, prefix);
+ new_var = dict_create_var (dict, name, 0);
assert (new_var != NULL);
*v = new_var;
- add_transformation (f, regression_trns_free, t);
+ add_transformation (ds, f, regression_trns_free, t);
trns_index++;
}
+
static void
-subcommand_save (int save, pspp_linreg_cache ** models)
+subcommand_save (struct dataset *ds, int save, pspp_linreg_cache ** models)
{
pspp_linreg_cache **lc;
int n_trns = 0;
assert ((*lc)->depvar != NULL);
if (cmd.a_save[REGRESSION_SV_RESID])
{
- reg_save_var ("RES", regression_trns_resid_proc, *lc,
+ reg_save_var (ds, "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,
+ reg_save_var (ds, "PRED", regression_trns_pred_proc, *lc,
&(*lc)->pred, n_trns);
}
}
{
for (lc = models; lc < models + cmd.n_dependent; lc++)
{
- assert (*lc != NULL);
- pspp_linreg_cache_free (*lc);
+ if (*lc != NULL)
+ {
+ pspp_linreg_cache_free (*lc);
+ }
}
}
}
+
static int
reg_inserted (const struct variable *v, struct variable **varlist, int n_vars)
{
for (i = 0; i < n_vars; i++)
{
- if (v->index == varlist[i]->index)
+ if (v == varlist[i])
{
return 1;
}
}
return 0;
}
+
static void
reg_print_categorical_encoding (FILE * fp, pspp_linreg_cache * c)
{
int i;
- size_t j;
int n_vars = 0;
struct variable **varlist;
- struct pspp_coeff *coeff;
- const struct variable *v;
- union value *val;
fprintf (fp, "%s", reg_export_categorical_encode_1);
varlist = xnmalloc (c->n_indeps, sizeof (*varlist));
for (i = 1; i < c->n_indeps; i++) /* c->coeff[0] is the intercept. */
{
- coeff = c->coeff[i];
- v = pspp_coeff_get_var (coeff, 0);
- if (v->type == ALPHA)
+ struct pspp_coeff *coeff = c->coeff[i];
+ const struct variable *v = pspp_coeff_get_var (coeff, 0);
+ if (var_is_alpha (v))
{
if (!reg_inserted (v, varlist, n_vars))
{
fprintf (fp, "struct pspp_reg_categorical_variable %s;\n\t",
- v->name);
+ var_get_name (v));
varlist[n_vars] = (struct variable *) v;
n_vars++;
}
n_vars);
for (i = 0; i < n_vars - 1; i++)
{
- fprintf (fp, "&%s,\n\t\t", varlist[i]->name);
+ fprintf (fp, "&%s,\n\t\t", var_get_name (varlist[i]));
}
- fprintf (fp, "&%s};\n\t", varlist[i]->name);
+ fprintf (fp, "&%s};\n\t", var_get_name (varlist[i]));
for (i = 0; i < n_vars; i++)
{
- coeff = c->coeff[i];
- fprintf (fp, "%s.name = \"%s\";\n\t", varlist[i]->name,
- varlist[i]->name);
- fprintf (fp, "%s.n_vals = %d;\n\t", varlist[i]->name,
- varlist[i]->obs_vals->n_categories);
+ int n_categories = cat_get_n_categories (varlist[i]);
+ int j;
+
+ fprintf (fp, "%s.name = \"%s\";\n\t",
+ var_get_name (varlist[i]), var_get_name (varlist[i]));
+ fprintf (fp, "%s.n_vals = %d;\n\t",
+ var_get_name (varlist[i]), n_categories);
- for (j = 0; j < varlist[i]->obs_vals->n_categories; j++)
+ for (j = 0; j < n_categories; j++)
{
- val = cat_subscript_to_value ((const size_t) j, varlist[i]);
- fprintf (fp, "%s.values[%d] = \"%s\";\n\t", varlist[i]->name, j,
- value_to_string (val, varlist[i]));
+ union value *val = cat_subscript_to_value (j, varlist[i]);
+ fprintf (fp, "%s.values[%d] = \"%s\";\n\t",
+ var_get_name (varlist[i]), j,
+ var_get_value_name (varlist[i], val));
}
}
fprintf (fp, "%s", reg_export_categorical_encode_2);
{
coeff = c->coeff[i];
v = pspp_coeff_get_var (coeff, 0);
- fprintf (fp, "\"%s\",\n\t\t", v->name);
+ fprintf (fp, "\"%s\",\n\t\t", var_get_name (v));
}
coeff = c->coeff[i];
v = pspp_coeff_get_var (coeff, 0);
- fprintf (fp, "\"%s\"};\n\t", v->name);
+ fprintf (fp, "\"%s\"};\n\t", var_get_name (v));
}
static void
reg_print_getvar (FILE * fp, pspp_linreg_cache * c)
for (i = 1; i < c->n_coeffs; i++)
{
v = pspp_coeff_get_var (c->coeff[i], 0);
- if (v->type == ALPHA)
- {
- return 1;
- }
+ if (var_is_alpha (v))
+ return 1;
}
return 0;
}
size_t i;
size_t j;
int n_quantiles = 100;
- double increment;
double tmp;
struct pspp_coeff *coeff;
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++)
{
tmp = 0.5 + 0.005 * (double) i;
fclose (fp);
}
}
+
static int
-regression_custom_export (struct cmd_regression *cmd UNUSED, void *aux UNUSED)
+regression_custom_export (struct lexer *lexer, struct dataset *ds UNUSED,
+ struct cmd_regression *cmd UNUSED, void *aux UNUSED)
{
/* 0 on failure, 1 on success, 2 on failure that should result in syntax error */
- if (!lex_force_match ('('))
+ if (!lex_force_match (lexer, '('))
return 0;
- if (lex_match ('*'))
+ if (lex_match (lexer, '*'))
model_file = NULL;
else
{
- model_file = fh_parse (FH_REF_FILE);
+ model_file = fh_parse (lexer, FH_REF_FILE);
if (model_file == NULL)
return 0;
}
- if (!lex_force_match (')'))
+ if (!lex_force_match (lexer, ')'))
return 0;
return 1;
}
int
-cmd_regression (void)
+cmd_regression (struct lexer *lexer, struct dataset *ds)
{
- if (!parse_regression (&cmd, NULL))
+ size_t i;
+
+ if (!parse_regression (lexer, ds, &cmd, NULL))
return CMD_FAILURE;
models = xnmalloc (cmd.n_dependent, sizeof *models);
- if (!multipass_procedure_with_splits (run_regression, &cmd))
+ for (i = 0; i < cmd.n_dependent; i++)
+ {
+ models[i] = NULL;
+ }
+ if (!multipass_procedure_with_splits (ds, run_regression, &cmd))
return CMD_CASCADING_FAILURE;
- subcommand_save (cmd.sbc_save, models);
+ subcommand_save (ds, cmd.sbc_save, models);
free (v_variables);
free (models);
return pspp_reg_rc;
/*
Is variable k the dependent variable?
*/
-static int
+static bool
is_depvar (size_t k, const struct variable *v)
{
- /*
- compare_var_names returns 0 if the variable
- names match.
- */
- if (!compare_var_names (v, v_variables[k], NULL))
- return 1;
-
- return 0;
+ return v == v_variables[k];
}
/*
Mark missing cases. Return the number of non-missing cases.
+ Compute the first two moments.
*/
static size_t
-mark_missing_cases (const struct casefile *cf, struct variable *v,
- int *is_missing_case, double n_data)
+mark_missing_cases (const struct casefile *cf, const struct variable *v,
+ int *is_missing_case, double n_data,
+ struct moments_var *mom)
{
struct casereader *r;
struct ccase c;
size_t row;
const union value *val;
+ double w = 1.0;
- for (r = casefile_get_reader (cf);
+ for (r = casefile_get_reader (cf, NULL);
casereader_read (r, &c); case_destroy (&c))
{
row = casereader_cnum (r) - 1;
- val = case_data (&c, v->fv);
+ val = case_data (&c, v);
+ if (mom != NULL)
+ {
+ moments1_add (mom->m, val->f, w);
+ }
cat_value_update (v, val);
- if (mv_is_value_missing (&v->miss, val))
+ if (var_is_value_missing (v, val, MV_ANY))
{
if (!is_missing_case[row])
{
/* Parser for the variables sub command */
static int
-regression_custom_variables (struct cmd_regression *cmd UNUSED,
- void *aux UNUSED)
+regression_custom_variables (struct lexer *lexer, struct dataset *ds,
+ struct cmd_regression *cmd UNUSED,
+ void *aux UNUSED)
{
+ const struct dictionary *dict = dataset_dict (ds);
- lex_match ('=');
+ lex_match (lexer, '=');
- if ((token != T_ID || dict_lookup_var (default_dict, tokid) == NULL)
- && token != T_ALL)
+ if ((lex_token (lexer) != T_ID
+ || dict_lookup_var (dict, lex_tokid (lexer)) == NULL)
+ && lex_token (lexer) != T_ALL)
return 2;
- if (!parse_variables (default_dict, &v_variables, &n_variables, PV_NONE))
+ if (!parse_variables_const
+ (lexer, dict, &v_variables, &n_variables, PV_NONE))
{
free (v_variables);
return 0;
*/
static int
prepare_data (int n_data, int is_missing_case[],
- struct variable **indep_vars,
- struct variable *depvar, const struct casefile *cf)
+ const struct variable **indep_vars,
+ const struct variable *depvar, const struct casefile *cf,
+ struct moments_var *mom)
{
int i;
int j;
{
indep_vars[j] = v_variables[i];
j++;
- if (v_variables[i]->type == ALPHA)
+ if (var_is_alpha (v_variables[i]))
{
/* 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 (cf, v_variables[i], is_missing_case, n_data,
+ mom + i);
}
}
/*
Mark missing cases for the dependent variable.
*/
- n_data = mark_missing_cases (cf, depvar, is_missing_case, n_data);
+ n_data = mark_missing_cases (cf, depvar, is_missing_case, n_data, NULL);
return n_data;
}
static void
-coeff_init (pspp_linreg_cache *c, struct design_matrix *dm)
+coeff_init (pspp_linreg_cache * c, struct design_matrix *dm)
{
c->coeff = xnmalloc (dm->m->size2 + 1, sizeof (*c->coeff));
- c->coeff[0] = xmalloc (sizeof (*(c->coeff[0]))); /* The first coefficient is the intercept. */
+ c->coeff[0] = xmalloc (sizeof (*(c->coeff[0]))); /* The first coefficient is the intercept. */
c->coeff[0]->v_info = NULL; /* Intercept has no associated variable. */
pspp_coeff_init (c->coeff + 1, dm);
}
+
+/*
+ Put the moments in the linreg cache.
+ */
+static void
+compute_moments (pspp_linreg_cache * c, struct moments_var *mom,
+ struct design_matrix *dm, size_t n)
+{
+ size_t i;
+ size_t j;
+ double weight;
+ double mean;
+ double variance;
+ double skewness;
+ double kurtosis;
+ /*
+ Scan the variable names in the columns of the design matrix.
+ When we find the variable we need, insert its mean in the cache.
+ */
+ for (i = 0; i < dm->m->size2; i++)
+ {
+ for (j = 0; j < n; j++)
+ {
+ if (design_matrix_col_to_var (dm, i) == (mom + j)->v)
+ {
+ moments1_calculate ((mom + j)->m, &weight, &mean, &variance,
+ &skewness, &kurtosis);
+ gsl_vector_set (c->indep_means, i, mean);
+ gsl_vector_set (c->indep_std, i, sqrt (variance));
+ }
+ }
+ }
+}
static bool
run_regression (const struct ccase *first,
- const struct casefile *cf, void *cmd_ UNUSED)
+ const struct casefile *cf, void *cmd_ UNUSED,
+ const struct dataset *ds)
{
size_t i;
size_t n_data = 0; /* Number of valide cases. */
const union value *val;
struct casereader *r;
struct ccase c;
- struct variable **indep_vars;
+ const struct variable **indep_vars;
struct design_matrix *X;
+ struct moments_var *mom;
gsl_vector *Y;
pspp_linreg_opts lopts;
assert (models != NULL);
- output_split_file_values (first);
+ output_split_file_values (ds, first);
if (!v_variables)
{
- dict_get_vars (default_dict, &v_variables, &n_variables,
+ dict_get_vars (dataset_dict (ds), &v_variables, &n_variables,
1u << DC_SYSTEM);
}
for (i = 0; i < cmd.n_dependent; i++)
{
- if (cmd.v_dependent[i]->type != NUMERIC)
+ if (!var_is_numeric (cmd.v_dependent[i]))
{
msg (SE, gettext ("Dependent variable must be numeric."));
pspp_reg_rc = CMD_FAILURE;
}
is_missing_case = xnmalloc (n_cases, sizeof (*is_missing_case));
-
+ mom = xnmalloc (n_variables, sizeof (*mom));
+ for (i = 0; i < n_variables; i++)
+ {
+ (mom + i)->m = moments1_create (MOMENT_VARIANCE);
+ (mom + i)->v = v_variables[i];
+ }
lopts.get_depvar_mean_std = 1;
for (k = 0; k < cmd.n_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 =
- design_matrix_create (n_indep, (const struct variable **) indep_vars,
- n_data);
- for (i = 0; i < X->m->size2; i++)
- {
- lopts.get_indep_mean_std[i] = 1;
- }
- 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)
- {
- models[k]->method = PSPP_LINREG_SVD;
- }
-
- /*
- The second pass fills the design matrix.
- */
- row = 0;
- for (r = casefile_get_reader (cf); casereader_read (r, &c);
- case_destroy (&c))
- /* Iterate over the cases. */
+ (const struct casefile *) cf, mom);
+ if (n_data > 0)
{
- case_num = casereader_cnum (r) - 1;
- if (!is_missing_case[case_num])
+ Y = gsl_vector_alloc (n_data);
+
+ X =
+ design_matrix_create (n_indep, (const struct variable **) indep_vars,
+ n_data);
+ for (i = 0; i < X->m->size2; i++)
+ {
+ lopts.get_indep_mean_std[i] = 1;
+ }
+ 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)
+ {
+ models[k]->method = PSPP_LINREG_SVD;
+ }
+
+ /*
+ The second pass fills the design matrix.
+ */
+ row = 0;
+ for (r = casefile_get_reader (cf, NULL); casereader_read (r, &c);
+ case_destroy (&c))
+ /* Iterate over the cases. */
{
- for (i = 0; i < n_variables; ++i) /* Iterate over the
- variables for the
- current case.
- */
+ case_num = casereader_cnum (r) - 1;
+ if (!is_missing_case[case_num])
{
- val = case_data (&c, v_variables[i]->fv);
- /*
- Independent/dependent variable separation. The
- 'variables' subcommand specifies a varlist which contains
- both dependent and independent variables. The dependent
- variables are specified with the 'dependent'
- subcommand, and maybe also in the 'variables' subcommand.
- We need to separate the two.
- */
- if (!is_depvar (i, cmd.v_dependent[k]))
+ for (i = 0; i < n_variables; ++i) /* Iterate over the
+ variables for the
+ current case.
+ */
{
- if (v_variables[i]->type == ALPHA)
- {
- design_matrix_set_categorical (X, row,
- v_variables[i], val);
- }
- else if (v_variables[i]->type == NUMERIC)
+ val = case_data (&c, v_variables[i]);
+ /*
+ Independent/dependent variable separation. The
+ 'variables' subcommand specifies a varlist which contains
+ both dependent and independent variables. The dependent
+ variables are specified with the 'dependent'
+ subcommand, and maybe also in the 'variables' subcommand.
+ We need to separate the two.
+ */
+ if (!is_depvar (i, cmd.v_dependent[k]))
{
- design_matrix_set_numeric (X, row, v_variables[i],
- val);
+ if (var_is_alpha (v_variables[i]))
+ {
+ design_matrix_set_categorical (X, row,
+ v_variables[i], val);
+ }
+ else
+ {
+ design_matrix_set_numeric (X, row, v_variables[i],
+ val);
+ }
}
}
+ val = case_data (&c, cmd.v_dependent[k]);
+ gsl_vector_set (Y, row, val->f);
+ row++;
}
- val = case_data (&c, cmd.v_dependent[k]->fv);
- gsl_vector_set (Y, row, val->f);
- row++;
}
+ /*
+ Now that we know the number of coefficients, allocate space
+ and store pointers to the variables that correspond to the
+ coefficients.
+ */
+ coeff_init (models[k], X);
+
+ /*
+ Find the least-squares estimates and other statistics.
+ */
+ pspp_linreg ((const gsl_vector *) Y, X->m, &lopts, models[k]);
+ compute_moments (models[k], mom, X, n_variables);
+ subcommand_statistics (cmd.a_statistics, models[k]);
+ subcommand_export (cmd.sbc_export, models[k]);
+
+ gsl_vector_free (Y);
+ design_matrix_destroy (X);
+ free (indep_vars);
+ free (lopts.get_indep_mean_std);
+ casereader_destroy (r);
}
- /*
- Now that we know the number of coefficients, allocate space
- and store pointers to the variables that correspond to the
- coefficients.
- */
- coeff_init (models[k], X);
-
- /*
- Find the least-squares estimates and other statistics.
- */
- 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);
- free (indep_vars);
- free (lopts.get_indep_mean_std);
- casereader_destroy (r);
}
-
+ for (i = 0; i < n_variables; i++)
+ {
+ moments1_destroy ((mom + i)->m);
+ }
+ free (mom);
free (is_missing_case);
return true;