#include <math/covariance.h>
#include <math/categoricals.h>
+#include <math/moments.h>
#include <gsl/gsl_matrix.h>
#include <linreg/sweep.h>
#include <language/command.h>
#include <data/procedure.h>
+#include <data/value.h>
#include <data/dictionary.h>
-
#include <language/dictionary/split-file.h>
#include <libpspp/hash.h>
#include <libpspp/taint.h>
/* The weight variable */
const struct variable *wv;
+
};
+/* Per category data */
+struct descriptive_data
+{
+ const struct variable *var;
+ struct moments1 *mom;
+
+ double minimum;
+ double maximum;
+};
/* Workspace variable for each dependent variable */
struct per_var_ws
int n_groups;
- double cc;
+ double mse;
};
struct oneway_workspace
struct hsh_table *group_hash;
struct per_var_ws *vws;
+
+ /* An array of descriptive data. One for each dependent variable */
+ struct descriptive_data **dd_total;
};
/* Routines to show the output tables */
static void show_anova_table (const struct oneway_spec *, const struct oneway_workspace *);
-static void show_descriptives (const struct oneway_spec *);
+static void show_descriptives (const struct oneway_spec *, const struct oneway_workspace *);
static void show_homogeneity (const struct oneway_spec *, const struct oneway_workspace *);
static void output_oneway (const struct oneway_spec *, struct oneway_workspace *ws);
struct casereader *group;
bool ok;
-
-
grouper = casegrouper_create_splits (proc_open (ds), dict);
while (casegrouper_get_next_group (grouper, &group))
run_oneway (&oneway, group, ds);
free (value);
}
+
+
static void postcalc (const struct oneway_spec *cmd);
static void precalc (const struct oneway_spec *cmd);
+static struct descriptive_data *
+dd_create (const struct variable *var)
+{
+ struct descriptive_data *dd = xmalloc (sizeof *dd);
+
+ dd->mom = moments1_create (MOMENT_VARIANCE);
+ dd->minimum = DBL_MAX;
+ dd->maximum = -DBL_MAX;
+ dd->var = var;
+
+ return dd;
+}
+
+
+static void *
+makeit (void *aux1, void *aux2 UNUSED)
+{
+ const struct variable *var = aux1;
+
+ struct descriptive_data *dd = dd_create (var);
+
+ return dd;
+}
+
+static void
+updateit (void *user_data,
+ enum mv_class exclude,
+ const struct variable *wv,
+ const struct variable *catvar UNUSED,
+ const struct ccase *c,
+ void *aux1, void *aux2)
+{
+ struct descriptive_data *dd = user_data;
+
+ const struct variable *varp = aux1;
+
+ const union value *valx = case_data (c, varp);
+
+ if ( var_is_value_missing (varp, valx, exclude))
+ return;
+
+ struct descriptive_data *dd_total = aux2;
+
+ double weight = 1.0;
+ if (wv)
+ weight = case_data (c, wv)->f;
+
+ moments1_add (dd->mom, valx->f, weight);
+ if (valx->f * weight < dd->minimum)
+ dd->minimum = valx->f * weight;
+
+ if (valx->f * weight > dd->maximum)
+ dd->maximum = valx->f * weight;
+
+ {
+ const struct variable *var = dd_total->var;
+ const union value *val = case_data (c, var);
+
+ moments1_add (dd_total->mom,
+ val->f,
+ weight);
+
+ if (val->f * weight < dd_total->minimum)
+ dd_total->minimum = val->f * weight;
+
+ if (val->f * weight > dd_total->maximum)
+ dd_total->maximum = val->f * weight;
+ }
+}
static void
run_oneway (const struct oneway_spec *cmd,
struct casereader *reader;
struct ccase *c;
-
struct oneway_workspace ws;
+ ws.actual_number_of_groups = 0;
ws.vws = xmalloc (cmd->n_vars * sizeof (*ws.vws));
+ ws.dd_total = xmalloc (sizeof (struct descriptive_data) * cmd->n_vars);
+ for (v = 0 ; v < cmd->n_vars; ++v)
+ ws.dd_total[v] = dd_create (cmd->vars[v]);
for (v = 0; v < cmd->n_vars; ++v)
{
struct categoricals *cats = categoricals_create (&cmd->indep_var, 1,
- cmd->wv, cmd->exclude);
+ cmd->wv, cmd->exclude,
+ makeit,
+ updateit,
+ cmd->vars[v], ws.dd_total[v]);
ws.vws[v].cov = covariance_2pass_create (1, &cmd->vars[v],
cats,
cmd->wv, cmd->exclude);
- ws.vws[v].cc = 0;
}
c = casereader_peek (input, 0);
{
struct per_var_ws *pvw = &ws.vws[i];
- pvw->cc += weight;
covariance_accumulate_pass1 (pvw->cov, c);
}
gsl_matrix *cm = covariance_calculate_unnormalized (pvw->cov);
const struct categoricals *cats = covariance_get_categoricals (pvw->cov);
+ double n;
+ moments1_calculate (ws.dd_total[v]->mom, &n, NULL, NULL, NULL, NULL);
+
pvw->sst = gsl_matrix_get (cm, 0, 0);
+ // gsl_matrix_fprintf (stdout, cm, "%g ");
+
reg_sweep (cm, 0);
pvw->sse = gsl_matrix_get (cm, 0, 0);
pvw->ssa = pvw->sst - pvw->sse;
pvw->n_groups = categoricals_total (cats);
+
+ pvw->mse = (pvw->sst - pvw->ssa) / (n - pvw->n_groups);
}
postcalc (cmd);
+
+ for (v = 0; v < cmd->n_vars; ++v)
+ {
+ struct categoricals *cats = covariance_get_categoricals (ws.vws[v].cov);
+
+ categoricals_done (cats);
+
+ if (categoricals_total (cats) > ws.actual_number_of_groups)
+ ws.actual_number_of_groups = categoricals_total (cats);
+ }
+
if ( cmd->stats & STATS_HOMOGENEITY )
levene (dict, casereader_clone (input), cmd->indep_var,
cmd->n_vars, cmd->vars, cmd->exclude);
casereader_destroy (input);
- ws.actual_number_of_groups = hsh_count (ws.group_hash);
-
if (!taint_has_tainted_successor (taint))
output_oneway (cmd, &ws);
}
}
-static void show_contrast_coeffs (const struct oneway_spec *cmd, struct oneway_workspace *ws);
-static void show_contrast_tests (const struct oneway_spec *cmd, struct oneway_workspace *ws);
+static void show_contrast_coeffs (const struct oneway_spec *cmd, const struct oneway_workspace *ws);
+static void show_contrast_tests (const struct oneway_spec *cmd, const struct oneway_workspace *ws);
static void
output_oneway (const struct oneway_spec *cmd, struct oneway_workspace *ws)
}
if (cmd->stats & STATS_DESCRIPTIVES)
- show_descriptives (cmd);
+ show_descriptives (cmd, ws);
if (cmd->stats & STATS_HOMOGENEITY)
show_homogeneity (cmd, ws);
show_anova_table (cmd, ws);
-
if (ll_count (&cmd->contrast_list) > 0)
{
show_contrast_coeffs (cmd, ws);
show_contrast_tests (cmd, ws);
}
-
/* Clean up */
for (i = 0; i < cmd->n_vars; ++i )
{
for (i = 0; i < cmd->n_vars; ++i)
{
+ double n;
+ moments1_calculate (ws->dd_total[i]->mom, &n, NULL, NULL, NULL, NULL);
+
const struct per_var_ws *pvw = &ws->vws[i];
- struct group_proc *gp = group_proc_get (cmd->vars[i]);
const double df1 = pvw->n_groups - 1;
- const double df2 = pvw->cc - pvw->n_groups;
+ const double df2 = n - pvw->n_groups;
const double msa = pvw->ssa / df1;
const char *s = var_to_string (cmd->vars[i]);
tab_hline (t, TAL_1, 0, n_cols - 1, i * 3 + 1);
- gp->mse = (pvw->sst - pvw->ssa) / df2;
-
/* Sums of Squares */
tab_double (t, 2, i * 3 + 1, 0, pvw->ssa, NULL);
tab_double (t, 2, i * 3 + 3, 0, pvw->sst, NULL);
/* Degrees of freedom */
tab_fixed (t, 3, i * 3 + 1, 0, df1, 4, 0);
tab_fixed (t, 3, i * 3 + 2, 0, df2, 4, 0);
- tab_fixed (t, 3, i * 3 + 3, 0, pvw->cc - 1, 4, 0);
+ tab_fixed (t, 3, i * 3 + 3, 0, n - 1, 4, 0);
/* Mean Squares */
tab_double (t, 4, i * 3 + 1, TAB_RIGHT, msa, NULL);
- tab_double (t, 4, i * 3 + 2, TAB_RIGHT, gp->mse, NULL);
+ tab_double (t, 4, i * 3 + 2, TAB_RIGHT, pvw->mse, NULL);
{
- const double F = msa / gp->mse ;
+ const double F = msa / pvw->mse ;
/* The F value */
tab_double (t, 5, i * 3 + 1, 0, F, NULL);
/* Show the descriptives table */
static void
-show_descriptives (const struct oneway_spec *cmd)
+show_descriptives (const struct oneway_spec *cmd, const struct oneway_workspace *ws)
{
size_t v;
int n_cols = 10;
const double confidence = 0.95;
const double q = (1.0 - confidence) / 2.0;
- const struct fmt_spec *wfmt = cmd->wv ? var_get_print_format (cmd->wv) : & F_8_0;
+ const struct fmt_spec *wfmt = cmd->wv ? var_get_print_format (cmd->wv) : &F_8_0;
int n_rows = 2;
- for ( v = 0; v < cmd->n_vars; ++v )
- n_rows += group_proc_get (cmd->vars[v])->n_groups + 1;
+ for (v = 0; v < cmd->n_vars; ++v)
+ n_rows += ws->actual_number_of_groups + 1;
t = tab_create (n_cols, n_rows);
tab_headers (t, 2, 0, 2, 0);
-
/* Put a frame around the entire box, and vertical lines inside */
tab_box (t,
TAL_2, TAL_2,
tab_text (t, 8, 1, TAB_CENTER | TAT_TITLE, _("Minimum"));
tab_text (t, 9, 1, TAB_CENTER | TAT_TITLE, _("Maximum"));
-
tab_title (t, _("Descriptives"));
-
row = 2;
for (v = 0; v < cmd->n_vars; ++v)
{
- double T;
- double std_error;
-
- struct group_proc *gp = group_proc_get (cmd->vars[v]);
-
- struct group_statistics *gs;
- struct group_statistics *totals = &gp->ugs;
-
const char *s = var_to_string (cmd->vars[v]);
const struct fmt_spec *fmt = var_get_print_format (cmd->vars[v]);
- struct group_statistics *const *gs_array =
- (struct group_statistics *const *) hsh_sort (gp->group_hash);
int count = 0;
+ struct per_var_ws *pvw = &ws->vws[v];
+ const struct categoricals *cats = covariance_get_categoricals (pvw->cov);
+
tab_text (t, 0, row, TAB_LEFT | TAT_TITLE, s);
if ( v > 0)
tab_hline (t, TAL_1, 0, n_cols - 1, row);
- for (count = 0; count < hsh_count (gp->group_hash); ++count)
+ for (count = 0; count < categoricals_total (cats); ++count)
{
+ double T;
+ double n, mean, variance;
+ double std_dev, std_error ;
+
struct string vstr;
+
+ const union value *gval = categoricals_get_value_by_subscript (cats, count);
+ const struct descriptive_data *dd = categoricals_get_user_data_by_subscript (cats, count);
+
+ moments1_calculate (dd->mom, &n, &mean, &variance, NULL, NULL);
+
+ std_dev = sqrt (variance);
+ std_error = std_dev / sqrt (n) ;
+
ds_init_empty (&vstr);
- gs = gs_array[count];
- var_append_value_name (cmd->indep_var, &gs->id, &vstr);
+ var_append_value_name (cmd->indep_var, gval, &vstr);
tab_text (t, 1, row + count,
TAB_LEFT | TAT_TITLE,
/* Now fill in the numbers ... */
- tab_fixed (t, 2, row + count, 0, gs->n, 8, 0);
+ tab_fixed (t, 2, row + count, 0, n, 8, 0);
- tab_double (t, 3, row + count, 0, gs->mean, NULL);
+ tab_double (t, 3, row + count, 0, mean, NULL);
- tab_double (t, 4, row + count, 0, gs->std_dev, NULL);
+ tab_double (t, 4, row + count, 0, std_dev, NULL);
- std_error = gs->std_dev / sqrt (gs->n) ;
- tab_double (t, 5, row + count, 0,
- std_error, NULL);
+
+ tab_double (t, 5, row + count, 0, std_error, NULL);
/* Now the confidence interval */
- T = gsl_cdf_tdist_Qinv (q, gs->n - 1);
+ T = gsl_cdf_tdist_Qinv (q, n - 1);
tab_double (t, 6, row + count, 0,
- gs->mean - T * std_error, NULL);
+ mean - T * std_error, NULL);
tab_double (t, 7, row + count, 0,
- gs->mean + T * std_error, NULL);
+ mean + T * std_error, NULL);
/* Min and Max */
- tab_double (t, 8, row + count, 0, gs->minimum, fmt);
- tab_double (t, 9, row + count, 0, gs->maximum, fmt);
+ tab_double (t, 8, row + count, 0, dd->minimum, fmt);
+ tab_double (t, 9, row + count, 0, dd->maximum, fmt);
}
- tab_text (t, 1, row + count,
- TAB_LEFT | TAT_TITLE, _("Total"));
+ {
+ double T;
+ double n, mean, variance;
+ double std_dev;
+ double std_error;
- tab_double (t, 2, row + count, 0, totals->n, wfmt);
+ moments1_calculate (ws->dd_total[v]->mom, &n, &mean, &variance, NULL, NULL);
- tab_double (t, 3, row + count, 0, totals->mean, NULL);
+ std_dev = sqrt (variance);
+ std_error = std_dev / sqrt (n) ;
- tab_double (t, 4, row + count, 0, totals->std_dev, NULL);
+ tab_text (t, 1, row + count,
+ TAB_LEFT | TAT_TITLE, _("Total"));
- std_error = totals->std_dev / sqrt (totals->n) ;
+ tab_double (t, 2, row + count, 0, n, wfmt);
- tab_double (t, 5, row + count, 0, std_error, NULL);
+ tab_double (t, 3, row + count, 0, mean, NULL);
- /* Now the confidence interval */
+ tab_double (t, 4, row + count, 0, std_dev, NULL);
- T = gsl_cdf_tdist_Qinv (q, totals->n - 1);
+ tab_double (t, 5, row + count, 0, std_error, NULL);
- tab_double (t, 6, row + count, 0,
- totals->mean - T * std_error, NULL);
+ /* Now the confidence interval */
+ T = gsl_cdf_tdist_Qinv (q, n - 1);
- tab_double (t, 7, row + count, 0,
- totals->mean + T * std_error, NULL);
+ tab_double (t, 6, row + count, 0,
+ mean - T * std_error, NULL);
- /* Min and Max */
+ tab_double (t, 7, row + count, 0,
+ mean + T * std_error, NULL);
- tab_double (t, 8, row + count, 0, totals->minimum, fmt);
- tab_double (t, 9, row + count, 0, totals->maximum, fmt);
+ /* Min and Max */
+ tab_double (t, 8, row + count, 0, ws->dd_total[v]->minimum, fmt);
+ tab_double (t, 9, row + count, 0, ws->dd_total[v]->maximum, fmt);
+ }
- row += gp->n_groups + 1;
+ row += categoricals_total (cats) + 1;
}
tab_submit (t);
int n_cols = 5;
size_t n_rows = cmd->n_vars + 1;
- struct tab_table *t;
-
-
- t = tab_create (n_cols, n_rows);
+ struct tab_table *t = tab_create (n_cols, n_rows);
tab_headers (t, 1, 0, 1, 0);
-
/* Put a frame around the entire box, and vertical lines inside */
tab_box (t,
TAL_2, TAL_2,
for (v = 0; v < cmd->n_vars; ++v)
{
- struct per_var_ws *pvw = &ws->vws[v];
+ double n;
+ moments1_calculate (ws->dd_total[v]->mom, &n, NULL, NULL, NULL, NULL);
+
+ const struct per_var_ws *pvw = &ws->vws[v];
const struct categoricals *cats = covariance_get_categoricals (pvw->cov);
const struct variable *var = cmd->vars[v];
const char *s = var_to_string (var);
const double df1 = pvw->n_groups - 1;
- const double df2 = pvw->cc - pvw->n_groups;
+ const double df2 = n - pvw->n_groups;
double F = gp->levene;
tab_text (t, 0, v + 1, TAB_LEFT | TAT_TITLE, s);
/* Show the contrast coefficients table */
static void
-show_contrast_coeffs (const struct oneway_spec *cmd, struct oneway_workspace *ws)
+show_contrast_coeffs (const struct oneway_spec *cmd, const struct oneway_workspace *ws)
{
int c_num = 0;
struct ll *cli;
/* Show the results of the contrast tests */
static void
-show_contrast_tests (const struct oneway_spec *cmd, struct oneway_workspace *ws)
+show_contrast_tests (const struct oneway_spec *cmd, const struct oneway_workspace *ws)
{
int n_contrasts = ll_count (&cmd->contrast_list);
size_t v;
tab_hline (t, TAL_2, 0, n_cols - 1, 1);
tab_vline (t, TAL_2, 3, 0, n_rows - 1);
-
tab_title (t, _("Contrast Tests"));
tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Contrast"));
for (v = 0; v < cmd->n_vars; ++v)
{
+ const struct per_var_ws *pvw = &ws->vws[v];
+ const struct categoricals *cats = covariance_get_categoricals (pvw->cov);
struct ll *cli;
int i = 0;
int lines_per_variable = 2 * n_contrasts;
++i, cli = ll_next (cli))
{
struct contrasts_node *cn = ll_data (cli, struct contrasts_node, ll);
- struct ll *coeffi = ll_head (&cn->coefficient_list);
- int ci;
+ struct ll *coeffi ;
+ int ci = 0;
double contrast_value = 0.0;
double coef_msq = 0.0;
- struct group_proc *grp_data = group_proc_get (cmd->vars[v]);
- struct hsh_table *group_hash = grp_data->group_hash;
-
- void *const *group_stat_array;
double T;
double std_error_contrast;
double df_denominator = 0.0;
double df_numerator = 0.0;
+
+ double grand_n;
+ moments1_calculate (ws->dd_total[v]->mom, &grand_n, NULL, NULL, NULL, NULL);
+ df = grand_n - pvw->n_groups;
+
if ( i == 0 )
{
tab_text (t, 1, (v * lines_per_variable) + i + 1,
if (cn->bad_count)
continue;
- group_stat_array = hsh_sort (group_hash);
-
- for (ci = 0;
- coeffi != ll_null (&cn->coefficient_list) &&
- ci < hsh_count (group_hash);
+ for (coeffi = ll_head (&cn->coefficient_list);
+ coeffi != ll_null (&cn->coefficient_list);
++ci, coeffi = ll_next (coeffi))
{
+ double n, mean, variance;
+ const struct descriptive_data *dd = categoricals_get_user_data_by_subscript (cats, ci);
+
+ moments1_calculate (dd->mom, &n, &mean, &variance, NULL, NULL);
+
struct coeff_node *cn = ll_data (coeffi, struct coeff_node, ll);
const double coef = cn->coeff;
- struct group_statistics *gs = group_stat_array[ci];
- const double winv = pow2 (gs->std_dev) / gs->n;
+ const double winv = variance / n;
- contrast_value += coef * gs->mean;
+ contrast_value += coef * mean;
- coef_msq += (coef * coef) / gs->n;
+ coef_msq += (pow2 (coef)) / n;
- sec_vneq += (coef * coef) * pow2 (gs->std_dev) /gs->n;
+ sec_vneq += (pow2 (coef)) * variance / n;
- df_numerator += (coef * coef) * winv;
- df_denominator += pow2((coef * coef) * winv) / (gs->n - 1);
+ df_numerator += (pow2 (coef)) * winv;
+ df_denominator += pow2((pow2 (coef)) * winv) / (n - 1);
}
sec_vneq = sqrt (sec_vneq);
n_contrasts,
TAB_RIGHT, contrast_value, NULL);
- std_error_contrast = sqrt (grp_data->mse * coef_msq);
+ std_error_contrast = sqrt (pvw->mse * coef_msq);
/* Std. Error */
tab_double (t, 4, (v * lines_per_variable) + i + 1,
TAB_RIGHT, T,
NULL);
- df = grp_data->ugs.n - grp_data->n_groups;
/* Degrees of Freedom */
tab_fixed (t, 6, (v * lines_per_variable) + i + 1,
projects / pspp / blobdiff