/* PSPP - a program for statistical analysis.
- Copyright (C) 2005, 2009, 2010, 2011, 2012, 2013, 2014 Free Software Foundation, Inc.
+ Copyright (C) 2005, 2009, 2010, 2011, 2012, 2013, 2014,
+ 2016, 2017 Free Software Foundation, Inc.
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
#include <float.h>
#include <stdbool.h>
+#include <gsl/gsl_math.h>
#include <gsl/gsl_cdf.h>
#include <gsl/gsl_matrix.h>
#include <gl/intprops.h>
-#define REG_LARGE_DATA 1000
-
#define STATS_R 1
#define STATS_COEFF 2
#define STATS_ANOVA 4
bool resid;
bool pred;
+
+ bool origin;
};
struct regression_workspace
{
/* The new variables which will be introduced by /SAVE */
- const struct variable **predvars;
+ const struct variable **predvars;
const struct variable **residvars;
- /* A reader/writer pair to temporarily hold the
+ /* A reader/writer pair to temporarily hold the
values of the new variables */
struct casewriter *writer;
struct casereader *reader;
return true;
}
-static int
+static int
save_trans_func (void *aux, struct ccase **c, casenumber x UNUSED)
{
struct save_trans_data *save_trans_data = aux;
double pred = case_data_idx (in, ws->extras * k + ws->pred_idx)->f;
case_data_rw (*c, ws->predvars[k])->f = pred;
}
-
+
if (ws->res_idx != -1)
{
double resid = case_data_idx (in, ws->extras * k + ws->res_idx)->f;
regression.resid = false;
regression.ds = ds;
+ regression.origin = false;
- /* Accept an optional, completely pointless "/VARIABLES=" */
- lex_match (lexer, T_SLASH);
- if (lex_match_id (lexer, "VARIABLES"))
- {
- if (!lex_force_match (lexer, T_EQUALS))
- goto error;
- }
-
- if (!parse_variables_const (lexer, dict,
- ®ression.vars, ®ression.n_vars,
- PV_NO_DUPLICATE | PV_NUMERIC))
- goto error;
-
-
+ bool variables_seen = false;
+ bool method_seen = false;
+ bool dependent_seen = false;
while (lex_token (lexer) != T_ENDCMD)
{
lex_match (lexer, T_SLASH);
- if (lex_match_id (lexer, "DEPENDENT"))
+ if (lex_match_id (lexer, "VARIABLES"))
{
- if (!lex_force_match (lexer, T_EQUALS))
- goto error;
+ if (method_seen)
+ {
+ msg (SE, _("VARIABLES may not appear after %s"), "METHOD");
+ goto error;
+ }
+ if (dependent_seen)
+ {
+ msg (SE, _("VARIABLES may not appear after %s"), "DEPENDENT");
+ goto error;
+ }
+ variables_seen = true;
+ lex_match (lexer, T_EQUALS);
+
+ if (!parse_variables_const (lexer, dict,
+ ®ression.vars, ®ression.n_vars,
+ PV_NO_DUPLICATE | PV_NUMERIC))
+ goto error;
+ }
+ else if (lex_match_id (lexer, "DEPENDENT"))
+ {
+ dependent_seen = true;
+ lex_match (lexer, T_EQUALS);
free (regression.dep_vars);
regression.n_dep_vars = 0;
-
+
if (!parse_variables_const (lexer, dict,
®ression.dep_vars,
®ression.n_dep_vars,
PV_NO_DUPLICATE | PV_NUMERIC))
goto error;
}
+ else if (lex_match_id (lexer, "ORIGIN"))
+ {
+ regression.origin = true;
+ }
+ else if (lex_match_id (lexer, "NOORIGIN"))
+ {
+ regression.origin = false;
+ }
else if (lex_match_id (lexer, "METHOD"))
{
+ method_seen = true;
lex_match (lexer, T_EQUALS);
if (!lex_force_match_id (lexer, "ENTER"))
{
goto error;
}
+
+ if (! variables_seen)
+ {
+ if (!parse_variables_const (lexer, dict,
+ ®ression.vars, ®ression.n_vars,
+ PV_NO_DUPLICATE | PV_NUMERIC))
+ goto error;
+ }
}
else if (lex_match_id (lexer, "STATISTICS"))
{
+ unsigned long statistics = 0;
lex_match (lexer, T_EQUALS);
while (lex_token (lexer) != T_ENDCMD
{
if (lex_match (lexer, T_ALL))
{
- regression.stats = ~0;
+ statistics = ~0;
}
else if (lex_match_id (lexer, "DEFAULTS"))
{
- regression.stats |= STATS_DEFAULT;
+ statistics |= STATS_DEFAULT;
}
else if (lex_match_id (lexer, "R"))
{
- regression.stats |= STATS_R;
+ statistics |= STATS_R;
}
else if (lex_match_id (lexer, "COEFF"))
{
- regression.stats |= STATS_COEFF;
+ statistics |= STATS_COEFF;
}
else if (lex_match_id (lexer, "ANOVA"))
{
- regression.stats |= STATS_ANOVA;
+ statistics |= STATS_ANOVA;
}
else if (lex_match_id (lexer, "BCOV"))
{
- regression.stats |= STATS_BCOV;
+ statistics |= STATS_BCOV;
}
else if (lex_match_id (lexer, "CI"))
{
- regression.stats |= STATS_CI;
+ statistics |= STATS_CI;
- if (lex_match (lexer, T_LPAREN))
+ if (lex_match (lexer, T_LPAREN) &&
+ lex_force_num (lexer))
{
regression.ci = lex_number (lexer) / 100.0;
lex_get (lexer);
- lex_force_match (lexer, T_RPAREN);
+ if (! lex_force_match (lexer, T_RPAREN))
+ goto error;
}
}
else
goto error;
}
}
+
+ if (statistics)
+ regression.stats = statistics;
+
}
else if (lex_match_id (lexer, "SAVE"))
{
workspace.extras = 0;
workspace.res_idx = -1;
workspace.pred_idx = -1;
- workspace.writer = NULL;
+ workspace.writer = NULL;
workspace.reader = NULL;
+ workspace.residvars = NULL;
+ workspace.predvars = NULL;
if (save)
{
int i;
if (regression.resid)
{
- workspace.extras ++;
- workspace.res_idx = 0;
+ workspace.res_idx = workspace.extras ++;
workspace.residvars = xcalloc (regression.n_dep_vars, sizeof (*workspace.residvars));
for (i = 0; i < regression.n_dep_vars; ++i)
if (regression.pred)
{
- workspace.extras ++;
- workspace.pred_idx = 1;
+ workspace.pred_idx = workspace.extras ++;
workspace.predvars = xcalloc (regression.n_dep_vars, sizeof (*workspace.predvars));
for (i = 0; i < regression.n_dep_vars; ++i)
msg (SW, _("REGRESSION with SAVE ignores TEMPORARY. "
"Temporary transformations will be made permanent."));
+ if (dict_get_filter (dict))
+ msg (SW, _("REGRESSION with SAVE ignores FILTER. "
+ "All cases will be processed."));
+
workspace.writer = autopaging_writer_create (proto);
caseproto_unref (proto);
}
save_trans_data->ws = xmalloc (sizeof (workspace));
memcpy (save_trans_data->ws, &workspace, sizeof (workspace));
save_trans_data->n_dep_vars = regression.n_dep_vars;
-
+
add_transformation (ds, save_trans_func, save_trans_free, save_trans_data);
}
return n_indep_vars;
}
-
static double
fill_covariance (gsl_matrix * cov, struct covariance *all_cov,
const struct variable **vars,
/*
STATISTICS subcommand output functions.
*/
-static void reg_stats_r (const linreg *, const struct variable *);
-static void reg_stats_coeff (const linreg *, const gsl_matrix *, const struct variable *, const struct regression *);
-static void reg_stats_anova (const linreg *, const struct variable *);
-static void reg_stats_bcov (const linreg *, const struct variable *);
+static void reg_stats_r (const struct linreg *, const struct variable *);
+static void reg_stats_coeff (const struct linreg *, const gsl_matrix *, const struct variable *, const struct regression *);
+static void reg_stats_anova (const struct linreg *, const struct variable *);
+static void reg_stats_bcov (const struct linreg *, const struct variable *);
static void
-subcommand_statistics (const struct regression *cmd, const linreg * c, const gsl_matrix * cm,
+subcommand_statistics (const struct regression *cmd, const struct linreg * c, const gsl_matrix * cm,
const struct variable *var)
{
- if (cmd->stats & STATS_R)
+ if (cmd->stats & STATS_R)
reg_stats_r (c, var);
- if (cmd->stats & STATS_ANOVA)
+ if (cmd->stats & STATS_ANOVA)
reg_stats_anova (c, var);
if (cmd->stats & STATS_COEFF)
static void
-run_regression (const struct regression *cmd,
+run_regression (const struct regression *cmd,
struct regression_workspace *ws,
struct casereader *input)
{
size_t i;
- linreg **models;
+ struct linreg **models;
int k;
struct ccase *c;
fill_all_vars (all_vars, cmd);
cov = covariance_1pass_create (n_all_vars, all_vars,
dict_get_weight (dataset_dict (cmd->ds)),
- MV_ANY);
+ MV_ANY, cmd->origin == false);
reader = casereader_clone (input);
reader = casereader_create_filter_missing (reader, all_vars, n_all_vars,
gsl_matrix *this_cm = gsl_matrix_alloc (n_indep + 1, n_indep + 1);
double n_data = fill_covariance (this_cm, cov, vars, n_indep,
dep_var, all_vars, n_all_vars, means);
- models[k] = linreg_alloc (dep_var, vars, n_data, n_indep);
- models[k]->depvar = dep_var;
+ models[k] = linreg_alloc (dep_var, vars, n_data, n_indep, cmd->origin);
for (i = 0; i < n_indep; i++)
{
linreg_set_indep_variable_mean (models[k], i, means[i]);
}
linreg_set_depvar_mean (models[k], means[i]);
- /*
- For large data sets, use QR decomposition.
- */
- if (n_data > sqrt (n_indep) && n_data > REG_LARGE_DATA)
- {
- models[k]->method = LINREG_QR;
- }
-
if (n_data > 0)
{
/*
if (ws->extras > 0)
{
struct casereader *r = casereader_clone (reader);
-
+
for (; (c = casereader_read (r)) != NULL; case_unref (c))
{
- struct ccase *outc = case_clone (c);
+ struct ccase *outc = case_create (casewriter_get_proto (ws->writer));
for (k = 0; k < cmd->n_dep_vars; k++)
{
const struct variable **vars = xnmalloc (cmd->n_vars, sizeof (*vars));
if (cmd->resid)
{
- double obs = case_data (c, models[k]->depvar)->f;
+ double obs = case_data (c, linreg_dep_var (models[k]))->f;
double res = linreg_residual (models[k], obs, vals, n_indep);
case_data_rw_idx (outc, k * ws->extras + ws->res_idx)->f = res;
}
free (vals);
free (vars);
- }
+ }
casewriter_write (ws->writer, outc);
}
casereader_destroy (r);
static void
-reg_stats_r (const linreg * c, const struct variable *var)
+reg_stats_r (const struct linreg * c, const struct variable *var)
{
struct tab_table *t;
int n_rows = 2;
Table showing estimated regression coefficients.
*/
static void
-reg_stats_coeff (const linreg * c, const gsl_matrix *cov, const struct variable *var, const struct regression *cmd)
+reg_stats_coeff (const struct linreg * c, const gsl_matrix *cov, const struct variable *var, const struct regression *cmd)
{
size_t j;
int n_cols = 7;
const int heading_rows = 2;
int n_rows;
- int this_row;
- double t_stat;
+ int this_row = heading_rows;
double pval;
double std_err;
double beta;
tab_vline (t, TAL_0, 1, 0, 0);
- tab_hline (t, TAL_1, 2, 4, 1);
+ tab_hline (t, TAL_1, 2, 4, 1);
tab_joint_text (t, 2, 0, 3, 0, TAB_CENTER | TAT_TITLE, _("Unstandardized Coefficients"));
tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("B"));
tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Std. Error"));
tab_text (t, 4, 1, TAB_CENTER | TAT_TITLE, _("Beta"));
tab_text (t, 5, 1, TAB_CENTER | TAT_TITLE, _("t"));
tab_text (t, 6, 1, TAB_CENTER | TAT_TITLE, _("Sig."));
- tab_text (t, 1, heading_rows, TAB_LEFT | TAT_TITLE, _("(Constant)"));
- tab_double (t, 2, heading_rows, 0, linreg_intercept (c), NULL, RC_OTHER);
+
std_err = sqrt (gsl_matrix_get (linreg_cov (c), 0, 0));
if (cmd->stats & STATS_CI)
tab_double (t, 8, heading_rows, 0, upper, NULL, RC_OTHER);
tab_joint_text_format (t, 7, 0, 8, 0, TAB_CENTER | TAT_TITLE, _("%g%% Confidence Interval for B"), cmd->ci * 100);
- tab_hline (t, TAL_1, 7, 8, 1);
+ tab_hline (t, TAL_1, 7, 8, 1);
tab_text (t, 7, 1, TAB_CENTER | TAT_TITLE, _("Lower Bound"));
tab_text (t, 8, 1, TAB_CENTER | TAT_TITLE, _("Upper Bound"));
}
- tab_double (t, 3, heading_rows, 0, std_err, NULL, RC_OTHER);
- tab_double (t, 4, heading_rows, 0, 0.0, NULL, RC_OTHER);
- t_stat = linreg_intercept (c) / std_err;
- tab_double (t, 5, heading_rows, 0, t_stat, NULL, RC_OTHER);
- pval =
- 2 * gsl_cdf_tdist_Q (fabs (t_stat),
- (double) (linreg_n_obs (c) - linreg_n_coeffs (c)));
- tab_double (t, 6, heading_rows, 0, pval, NULL, RC_PVALUE);
-
- for (j = 0; j < linreg_n_coeffs (c); j++)
+
+ if (!cmd->origin)
+ {
+ tab_text (t, 1, this_row, TAB_LEFT | TAT_TITLE, _("(Constant)"));
+ tab_double (t, 2, this_row, 0, linreg_intercept (c), NULL, RC_OTHER);
+ tab_double (t, 3, this_row, 0, std_err, NULL, RC_OTHER);
+ tab_double (t, 4, this_row, 0, 0.0, NULL, RC_OTHER);
+ double t_stat = linreg_intercept (c) / std_err;
+ tab_double (t, 5, this_row, 0, t_stat, NULL, RC_OTHER);
+
+ double pval =
+ 2 * gsl_cdf_tdist_Q (fabs (t_stat),
+ (double) (linreg_n_obs (c) - linreg_n_coeffs (c)));
+ tab_double (t, 6, this_row, 0, pval, NULL, RC_PVALUE);
+ this_row++;
+ }
+
+ for (j = 0; j < linreg_n_coeffs (c); j++, this_row++)
{
struct string tstr;
ds_init_empty (&tstr);
- this_row = j + heading_rows + 1;
v = linreg_indep_var (c, j);
label = var_to_string (v);
/*
Test statistic for H0: coefficient is 0.
*/
- t_stat = linreg_coeff (c, j) / std_err;
+ double t_stat = linreg_coeff (c, j) / std_err;
tab_double (t, 5, this_row, 0, t_stat, NULL, RC_OTHER);
/*
P values for the test statistic above.
{
double lower = linreg_coeff (c, j) - tval * std_err ;
double upper = linreg_coeff (c, j) + tval * std_err ;
-
+
tab_double (t, 7, this_row, 0, lower, NULL, RC_OTHER);
tab_double (t, 8, this_row, 0, upper, NULL, RC_OTHER);
}
Display the ANOVA table.
*/
static void
-reg_stats_anova (const linreg * c, const struct variable *var)
+reg_stats_anova (const struct linreg * c, const struct variable *var)
{
int n_cols = 7;
int n_rows = 4;
const double msm = linreg_ssreg (c) / linreg_dfmodel (c);
const double mse = linreg_mse (c);
const double F = msm / mse;
- const double pval = gsl_cdf_fdist_Q (F, c->dfm, c->dfe);
+ const double pval = gsl_cdf_fdist_Q (F, linreg_dfmodel (c),
+ linreg_dferror (c));
struct tab_table *t;
/* Degrees of freedom */
- tab_text_format (t, 3, 1, TAB_RIGHT, "%.*g", DBL_DIG + 1, c->dfm);
- tab_text_format (t, 3, 2, TAB_RIGHT, "%.*g", DBL_DIG + 1, c->dfe);
- tab_text_format (t, 3, 3, TAB_RIGHT, "%.*g", DBL_DIG + 1, c->dft);
+ tab_text_format (t, 3, 1, TAB_RIGHT, "%.*g", DBL_DIG + 1, linreg_dfmodel (c));
+ tab_text_format (t, 3, 2, TAB_RIGHT, "%.*g", DBL_DIG + 1, linreg_dferror (c));
+ tab_text_format (t, 3, 3, TAB_RIGHT, "%.*g", DBL_DIG + 1, linreg_dftotal (c));
/* Mean Squares */
tab_double (t, 4, 1, TAB_RIGHT, msm, NULL, RC_OTHER);
static void
-reg_stats_bcov (const linreg * c, const struct variable *var)
+reg_stats_bcov (const struct linreg * c, const struct variable *var)
{
int n_cols;
int n_rows;
struct tab_table *t;
assert (c != NULL);
- n_cols = c->n_indeps + 1 + 2;
- n_rows = 2 * (c->n_indeps + 1);
+ n_cols = linreg_n_indeps (c) + 1 + 2;
+ n_rows = 2 * (linreg_n_indeps (c) + 1);
t = tab_create (n_cols, n_rows);
tab_headers (t, 2, 0, 1, 0);
tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
col = (i <= k) ? k : i;
row = (i <= k) ? i : k;
tab_double (t, k + 2, i, TAB_CENTER,
- gsl_matrix_get (c->cov, row, col), NULL, RC_OTHER);
+ gsl_matrix_get (linreg_cov (c), row, col), NULL, RC_OTHER);
}
}
tab_title (t, _("Coefficient Correlations (%s)"), var_to_string (var));