#include <data/casegrouper.h>
#include <data/casereader.h>
+#include <math/covariance.h>
+#include <math/categoricals.h>
+#include <gsl/gsl_matrix.h>
+#include <linreg/sweep.h>
+
#include <libpspp/ll.h>
#include <language/lexer/lexer.h>
size_t n_vars;
const struct variable **vars;
- const struct dictionary *dict;
-
const struct variable *indep_var;
enum statistics stats;
/* List of contrasts */
struct ll_list contrast_list;
+
+ /* The weight variable */
+ const struct variable *wv;
+};
+
+
+/* Workspace variable for each dependent variable */
+struct per_var_ws
+{
+ struct covariance *cov;
+
+ double sst;
+ double sse;
+ double ssa;
+
+ int n_groups;
+
+ double cc;
};
struct oneway_workspace
/* A hash table containing all the distinct values of the independent
variable */
struct hsh_table *group_hash;
+
+ struct per_var_ws *vws;
};
/* Routines to show the output tables */
-static void show_anova_table (const struct oneway_spec *);
-static void show_descriptives (const struct oneway_spec *, const struct dictionary *dict);
-static void show_homogeneity (const struct oneway_spec *);
+static void show_anova_table (const struct oneway_spec *, const struct oneway_workspace *);
+static void show_descriptives (const struct oneway_spec *);
+static void show_homogeneity (const struct oneway_spec *, const struct oneway_workspace *);
static void output_oneway (const struct oneway_spec *, struct oneway_workspace *ws);
static void run_oneway (const struct oneway_spec *cmd, struct casereader *input, const struct dataset *ds);
int
cmd_oneway (struct lexer *lexer, struct dataset *ds)
{
+ const struct dictionary *dict = dataset_dict (ds);
struct oneway_spec oneway ;
oneway.n_vars = 0;
oneway.vars = NULL;
oneway.stats = 0;
oneway.missing_type = MISS_ANALYSIS;
oneway.exclude = MV_ANY;
- oneway.dict = dataset_dict (ds);
+ oneway.wv = dict_get_weight (dict);
ll_init (&oneway.contrast_list);
lex_match (lexer, '=');
}
- if (!parse_variables_const (lexer, oneway.dict,
+ if (!parse_variables_const (lexer, dict,
&oneway.vars, &oneway.n_vars,
PV_NO_DUPLICATE | PV_NUMERIC))
goto error;
lex_force_match (lexer, T_BY);
- oneway.indep_var = parse_variable_const (lexer, oneway.dict);
+ oneway.indep_var = parse_variable_const (lexer, dict);
while (lex_token (lexer) != '.')
{
while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
{
- union value val;
- if ( parse_value (lexer, &val, 0))
+ if ( lex_is_number (lexer))
{
struct coeff_node *cc = xmalloc (sizeof *cc);
- cc->coeff = val.f;
+ cc->coeff = lex_number (lexer);
ll_push_tail (coefficient_list, &cc->ll);
+ lex_get (lexer);
}
else
{
struct casereader *group;
bool ok;
- grouper = casegrouper_create_splits (proc_open (ds), oneway.dict);
+
+
+ grouper = casegrouper_create_splits (proc_open (ds), dict);
while (casegrouper_get_next_group (grouper, &group))
run_oneway (&oneway, group, ds);
ok = casegrouper_destroy (grouper);
struct casereader *input,
const struct dataset *ds)
{
+ int v;
struct taint *taint;
struct dictionary *dict = dataset_dict (ds);
struct casereader *reader;
struct ccase *c;
+
struct oneway_workspace ws;
+ ws.vws = xmalloc (cmd->n_vars * sizeof (*ws.vws));
+
+
+ for (v = 0; v < cmd->n_vars; ++v)
+ {
+ struct categoricals *cats = categoricals_create (&cmd->indep_var, 1,
+ cmd->wv, cmd->exclude);
+
+ 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);
if (c == NULL)
{
input = casereader_create_filter_weight (input, dict, NULL, NULL);
reader = casereader_clone (input);
+
for (; (c = casereader_read (reader)) != NULL; case_unref (c))
{
size_t i;
for (i = 0; i < cmd->n_vars; ++i)
{
+ {
+ struct per_var_ws *pvw = &ws.vws[i];
+
+ pvw->cc += weight;
+ covariance_accumulate_pass1 (pvw->cov, c);
+ }
+
const struct variable *v = cmd->vars[i];
const union value *val = case_data (c, v);
}
casereader_destroy (reader);
+ reader = casereader_clone (input);
+ for ( ; (c = casereader_read (reader) ); case_unref (c))
+ {
+ int i;
+ for (i = 0; i < cmd->n_vars; ++i)
+ {
+ struct per_var_ws *pvw = &ws.vws[i];
+ covariance_accumulate_pass2 (pvw->cov, c);
+ }
+ }
+ casereader_destroy (reader);
+
+ for (v = 0; v < cmd->n_vars; ++v)
+ {
+ struct per_var_ws *pvw = &ws.vws[v];
+ gsl_matrix *cm = covariance_calculate_unnormalized (pvw->cov);
+ const struct categoricals *cats = covariance_get_categoricals (pvw->cov);
+
+ pvw->sst = gsl_matrix_get (cm, 0, 0);
+
+ reg_sweep (cm, 0);
+
+ pvw->sse = gsl_matrix_get (cm, 0, 0);
+
+ pvw->ssa = pvw->sst - pvw->sse;
+
+ pvw->n_groups = categoricals_total (cats);
+ }
postcalc (cmd);
}
if (cmd->stats & STATS_DESCRIPTIVES)
- show_descriptives (cmd, cmd->dict);
+ show_descriptives (cmd);
if (cmd->stats & STATS_HOMOGENEITY)
- show_homogeneity (cmd);
+ show_homogeneity (cmd, ws);
- show_anova_table (cmd);
+ show_anova_table (cmd, ws);
if (ll_count (&cmd->contrast_list) > 0)
/* Show the ANOVA table */
static void
-show_anova_table (const struct oneway_spec *cmd)
+show_anova_table (const struct oneway_spec *cmd, const struct oneway_workspace *ws)
{
size_t i;
int n_cols =7;
for (i = 0; i < cmd->n_vars; ++i)
{
- struct group_statistics *totals = &group_proc_get (cmd->vars[i])->ugs;
- struct hsh_table *group_hash = group_proc_get (cmd->vars[i])->group_hash;
- struct hsh_iterator g;
- struct group_statistics *gs;
- double ssa = 0;
- const char *s = var_to_string (cmd->vars[i]);
-
- for (gs = hsh_first (group_hash, &g);
- gs != 0;
- gs = hsh_next (group_hash, &g))
- {
- ssa += pow2 (gs->sum) / gs->n;
- }
+ 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 msa = pvw->ssa / df1;
- ssa -= pow2 (totals->sum) / totals->n;
+ const char *s = var_to_string (cmd->vars[i]);
tab_text (t, 0, i * 3 + 1, TAB_LEFT | TAT_TITLE, s);
tab_text (t, 1, i * 3 + 1, TAB_LEFT | TAT_TITLE, _("Between Groups"));
if (i > 0)
tab_hline (t, TAL_1, 0, n_cols - 1, i * 3 + 1);
- {
- struct group_proc *gp = group_proc_get (cmd->vars[i]);
- const double sst = totals->ssq - pow2 (totals->sum) / totals->n;
- const double df1 = gp->n_groups - 1;
- const double df2 = totals->n - gp->n_groups;
- const double msa = ssa / df1;
-
- gp->mse = (sst - ssa) / df2;
+ gp->mse = (pvw->sst - pvw->ssa) / df2;
- /* Sums of Squares */
- tab_double (t, 2, i * 3 + 1, 0, ssa, NULL);
- tab_double (t, 2, i * 3 + 3, 0, sst, NULL);
- tab_double (t, 2, i * 3 + 2, 0, sst - ssa, NULL);
+ /* 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);
+ tab_double (t, 2, i * 3 + 2, 0, pvw->sse, 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, totals->n - 1, 4, 0);
+ /* 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);
- /* 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);
+ /* 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);
- {
- const double F = msa / gp->mse ;
+ {
+ const double F = msa / gp->mse ;
- /* The F value */
- tab_double (t, 5, i * 3 + 1, 0, F, NULL);
+ /* The F value */
+ tab_double (t, 5, i * 3 + 1, 0, F, NULL);
- /* The significance */
- tab_double (t, 6, i * 3 + 1, 0, gsl_cdf_fdist_Q (F, df1, df2), NULL);
- }
+ /* The significance */
+ tab_double (t, 6, i * 3 + 1, 0, gsl_cdf_fdist_Q (F, df1, df2), NULL);
}
}
-
tab_title (t, _("ANOVA"));
tab_submit (t);
}
/* Show the descriptives table */
static void
-show_descriptives (const struct oneway_spec *cmd, const struct dictionary *dict)
+show_descriptives (const struct oneway_spec *cmd)
{
size_t v;
int n_cols = 10;
const double confidence = 0.95;
const double q = (1.0 - confidence) / 2.0;
- const struct variable *wv = dict_get_weight (dict);
- const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : & F_8_0;
+ const struct fmt_spec *wfmt = cmd->wv ? var_get_print_format (cmd->wv) : & F_8_0;
int n_rows = 2;
/* Show the homogeneity table */
static void
-show_homogeneity (const struct oneway_spec *cmd)
+show_homogeneity (const struct oneway_spec *cmd, const struct oneway_workspace *ws)
{
size_t v;
int n_cols = 5;
tab_hline (t, TAL_2, 0, n_cols - 1, 1);
tab_vline (t, TAL_2, 1, 0, n_rows - 1);
-
tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("Levene Statistic"));
tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("df1"));
tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("df2"));
for (v = 0; v < cmd->n_vars; ++v)
{
- double F;
+ 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 struct group_proc *gp = group_proc_get (cmd->vars[v]);
const char *s = var_to_string (var);
- const struct group_statistics *totals = &gp->ugs;
- const double df1 = gp->n_groups - 1;
- const double df2 = totals->n - gp->n_groups;
+ const double df1 = pvw->n_groups - 1;
+ const double df2 = pvw->cc - pvw->n_groups;
+ double F = gp->levene;
tab_text (t, 0, v + 1, TAB_LEFT | TAT_TITLE, s);
- F = gp->levene;
+
tab_double (t, 1, v + 1, TAB_RIGHT, F, NULL);
tab_fixed (t, 2, v + 1, TAB_RIGHT, df1, 8, 0);
tab_fixed (t, 3, v + 1, TAB_RIGHT, df2, 8, 0);
/* Now the significance */
- tab_double (t, 4, v + 1, TAB_RIGHT,gsl_cdf_fdist_Q (F, df1, df2), NULL);
+ tab_double (t, 4, v + 1, TAB_RIGHT, gsl_cdf_fdist_Q (F, df1, df2), NULL);
}
tab_submit (t);
int n_cols = 2 + ws->actual_number_of_groups;
int n_rows = 2 + n_contrasts;
- void *const *group_values;
-
struct tab_table *t;
+ const struct covariance *cov = ws->vws[0].cov ;
+
t = tab_create (n_cols, n_rows);
tab_headers (t, 2, 0, 2, 0);
tab_joint_text (t, 2, 0, n_cols - 1, 0, TAB_CENTER | TAT_TITLE,
var_to_string (cmd->indep_var));
- group_values = hsh_sort (ws->group_hash);
-
for ( cli = ll_head (&cmd->contrast_list);
cli != ll_null (&cmd->contrast_list);
cli = ll_next (cli))
{
int count = 0;
struct contrasts_node *cn = ll_data (cli, struct contrasts_node, ll);
- struct ll *coeffi = ll_head (&cn->coefficient_list);
+ struct ll *coeffi ;
tab_text_format (t, 1, c_num + 2, TAB_CENTER, "%d", c_num + 1);
- for (count = 0;
- count < hsh_count (ws->group_hash) && coeffi != ll_null (&cn->coefficient_list);
- ++count)
+ for (coeffi = ll_head (&cn->coefficient_list);
+ coeffi != ll_null (&cn->coefficient_list);
+ ++count, coeffi = ll_next (coeffi))
{
- double *group_value_p;
- union value group_value;
+ const struct categoricals *cats = covariance_get_categoricals (cov);
+ const union value *val = categoricals_get_value_by_subscript (cats, count);
struct string vstr;
ds_init_empty (&vstr);
- group_value_p = group_values[count];
- group_value.f = *group_value_p;
- var_append_value_name (cmd->indep_var, &group_value, &vstr);
+ var_append_value_name (cmd->indep_var, val, &vstr);
- tab_text (t, count + 2, 1, TAB_CENTER | TAT_TITLE,
- ds_cstr (&vstr));
+ tab_text (t, count + 2, 1, TAB_CENTER | TAT_TITLE, ds_cstr (&vstr));
ds_destroy (&vstr);
tab_text_format (t, count + 2, c_num + 2, TAB_RIGHT, "%g", coeffn->coeff);
}
-
- coeffi = ll_next (coeffi);
}
++c_num;
}