#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>
bool bad_count; /* True if the number of coefficients does not equal the number of groups */
};
-struct oneway
+struct oneway_spec
{
size_t n_vars;
const struct variable **vars;
+ const struct dictionary *dict;
+
const struct variable *indep_var;
enum statistics stats;
enum missing_type missing_type;
enum mv_class exclude;
+ /* List of contrasts */
+ struct ll_list contrast_list;
+};
+
+
+/* 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
+{
/* The number of distinct values of the independent variable, when all
missing values are disregarded */
int actual_number_of_groups;
variable */
struct hsh_table *group_hash;
- /* List of contrasts */
- struct ll_list contrast_list;
+ struct per_var_ws *vws;
};
/* Routines to show the output tables */
-static void show_anova_table (const struct oneway *);
-static void show_descriptives (const struct oneway *, const struct dictionary *dict);
-static void show_homogeneity (const struct oneway *);
+static void show_anova_table (const struct oneway_spec *, const struct oneway_workspace *);
+static void show_descriptives (const struct oneway_spec *, const struct dictionary *dict);
+static void show_homogeneity (const struct oneway_spec *);
-static void output_oneway (const struct oneway *, const struct dictionary *dict);
-static void run_oneway (struct oneway *cmd, struct casereader *input, const struct dataset *ds);
+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 oneway ;
+ struct oneway_spec oneway ;
oneway.n_vars = 0;
oneway.vars = NULL;
oneway.indep_var = NULL;
oneway.stats = 0;
oneway.missing_type = MISS_ANALYSIS;
oneway.exclude = MV_ANY;
- oneway.actual_number_of_groups = 0;
- oneway.group_hash = NULL;
+ oneway.dict = dataset_dict (ds);
ll_init (&oneway.contrast_list);
lex_match (lexer, '=');
}
- if (!parse_variables_const (lexer, dict,
+ if (!parse_variables_const (lexer, oneway.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, dict);
+ oneway.indep_var = parse_variable_const (lexer, oneway.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), dict);
+ grouper = casegrouper_create_splits (proc_open (ds), oneway.dict);
while (casegrouper_get_next_group (grouper, &group))
run_oneway (&oneway, group, ds);
ok = casegrouper_destroy (grouper);
free (value);
}
-static void postcalc (const struct oneway *cmd);
-static void precalc (const struct oneway *cmd);
+static void postcalc (const struct oneway_spec *cmd);
+static void precalc (const struct oneway_spec *cmd);
static void
-run_oneway (struct oneway *cmd,
+run_oneway (const struct oneway_spec *cmd,
struct casereader *input,
const struct dataset *ds)
{
+ int v;
struct taint *taint;
struct dictionary *dict = dataset_dict (ds);
struct casereader *reader;
struct ccase *c;
+ const struct variable *wv = dict_get_weight (dict);
+
+ struct oneway_workspace ws;
+
+ ws.vws = xmalloc (cmd->n_vars * sizeof (*ws.vws));
+
+ for (v = 0; v < cmd->n_vars; ++v)
+ {
+ ws.vws[v].cov = covariance_2pass_create (1, &cmd->vars[v],
+ 1, &cmd->indep_var,
+ wv, cmd->exclude);
+ ws.vws[v].cc = 0;
+ }
c = casereader_peek (input, 0);
if (c == NULL)
taint = taint_clone (casereader_get_taint (input));
- cmd->group_hash = hsh_create (4,
+ ws.group_hash = hsh_create (4,
compare_double_3way,
do_hash_double,
free_double,
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;
const double weight = dict_get_case_weight (dict, c, NULL);
const union value *indep_val = case_data (c, cmd->indep_var);
- void **p = hsh_probe (cmd->group_hash, &indep_val->f);
+ void **p = hsh_probe (ws.group_hash, &indep_val->f);
if (*p == NULL)
{
double *value = *p = xmalloc (sizeof *value);
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);
casereader_destroy (input);
- cmd->actual_number_of_groups = hsh_count (cmd->group_hash);
+ ws.actual_number_of_groups = hsh_count (ws.group_hash);
if (!taint_has_tainted_successor (taint))
- output_oneway (cmd, dict);
+ output_oneway (cmd, &ws);
taint_destroy (taint);
}
/* Pre calculations */
static void
-precalc (const struct oneway *cmd)
+precalc (const struct oneway_spec *cmd)
{
size_t i = 0;
/* Post calculations for the ONEWAY command */
static void
-postcalc (const struct oneway *cmd)
+postcalc (const struct oneway_spec *cmd)
{
size_t i = 0;
}
}
-static void show_contrast_coeffs (const struct oneway *cmd);
-static void show_contrast_tests (const struct oneway *cmd);
+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
-output_oneway (const struct oneway *cmd, const struct dictionary *dict)
+output_oneway (const struct oneway_spec *cmd, struct oneway_workspace *ws)
{
size_t i = 0;
struct ll_list *cl = &coeff_list->coefficient_list;
++i;
- if (ll_count (cl) != cmd->actual_number_of_groups)
+ if (ll_count (cl) != ws->actual_number_of_groups)
{
msg (SW,
_("Number of contrast coefficients must equal the number of groups"));
}
if (cmd->stats & STATS_DESCRIPTIVES)
- show_descriptives (cmd, dict);
+ show_descriptives (cmd, cmd->dict);
if (cmd->stats & STATS_HOMOGENEITY)
show_homogeneity (cmd);
- show_anova_table (cmd);
+ show_anova_table (cmd, ws);
if (ll_count (&cmd->contrast_list) > 0)
{
- show_contrast_coeffs (cmd);
- show_contrast_tests (cmd);
+ show_contrast_coeffs (cmd, ws);
+ show_contrast_tests (cmd, ws);
}
hsh_destroy (group_hash);
}
- hsh_destroy (cmd->group_hash);
+ hsh_destroy (ws->group_hash);
}
/* Show the ANOVA table */
static void
-show_anova_table (const struct oneway *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 *cmd, const struct dictionary *dict)
+show_descriptives (const struct oneway_spec *cmd, const struct dictionary *dict)
{
size_t v;
int n_cols = 10;
/* Show the homogeneity table */
static void
-show_homogeneity (const struct oneway *cmd)
+show_homogeneity (const struct oneway_spec *cmd)
{
size_t v;
int n_cols = 5;
/* Show the contrast coefficients table */
static void
-show_contrast_coeffs (const struct oneway *cmd)
+show_contrast_coeffs (const struct oneway_spec *cmd, struct oneway_workspace *ws)
{
int c_num = 0;
struct ll *cli;
int n_contrasts = ll_count (&cmd->contrast_list);
- int n_cols = 2 + cmd->actual_number_of_groups;
+ int n_cols = 2 + ws->actual_number_of_groups;
int n_rows = 2 + n_contrasts;
void *const *group_values;
tab_joint_text (t, 2, 0, n_cols - 1, 0, TAB_CENTER | TAT_TITLE,
var_to_string (cmd->indep_var));
- group_values = hsh_sort (cmd->group_hash);
+ group_values = hsh_sort (ws->group_hash);
for ( cli = ll_head (&cmd->contrast_list);
cli != ll_null (&cmd->contrast_list);
tab_text_format (t, 1, c_num + 2, TAB_CENTER, "%d", c_num + 1);
for (count = 0;
- count < hsh_count (cmd->group_hash) && coeffi != ll_null (&cn->coefficient_list);
+ count < hsh_count (ws->group_hash) && coeffi != ll_null (&cn->coefficient_list);
++count)
{
double *group_value_p;
group_value.f = *group_value_p;
var_append_value_name (cmd->indep_var, &group_value, &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);
/* Show the results of the contrast tests */
static void
-show_contrast_tests (const struct oneway *cmd)
+show_contrast_tests (const struct oneway_spec *cmd, struct oneway_workspace *ws)
{
int n_contrasts = ll_count (&cmd->contrast_list);
size_t v;