#include <data/casegrouper.h>
#include <data/casereader.h>
+#include <math/covariance.h>
+#include <math/categoricals.h>
+#include <math/moments.h>
+#include <gsl/gsl_matrix.h>
+#include <linreg/sweep.h>
+
#include <libpspp/ll.h>
#include <language/lexer/lexer.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>
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;
+
+ struct moments1 *totals;
+ double minimum;
+ double maximum;
};
/* 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 *, 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);
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) != '.')
{
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);
free (value);
}
+
+
static void postcalc (const struct oneway_spec *cmd);
static void precalc (const struct oneway_spec *cmd);
+struct descriptive_data
+{
+ struct moments1 *mom;
+ double minimum;
+ double maximum;
+};
+
+static void *
+makeit (void)
+{
+ struct descriptive_data *dd = xmalloc (sizeof *dd);
+ dd->mom = moments1_create (MOMENT_VARIANCE);
+ dd->minimum = DBL_MAX;
+ dd->maximum = -DBL_MAX;
+
+ return dd;
+}
+
+static void
+updateit (void *user_data, const struct variable *wv,
+ const struct variable *catvar, const struct ccase *c, void *aux)
+{
+ const union value *val = case_data_idx (c, 0);
+ struct descriptive_data *dd = user_data;
+ struct oneway_workspace *ws = aux;
+
+ double weight = 1.0;
+ if (wv)
+ weight = case_data (c, wv)->f;
+
+ moments1_add (dd->mom, val->f, weight);
+ moments1_add (ws->totals, val->f, weight);
+
+ if (val->f * weight < dd->minimum)
+ dd->minimum = val->f * weight;
+
+ if (val->f * weight > dd->maximum)
+ dd->maximum = val->f * weight;
+
+
+ if (val->f * weight < ws->minimum)
+ ws->minimum = val->f * weight;
+
+ if (val->f * weight > ws->maximum)
+ ws->maximum = val->f * weight;
+}
static void
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 oneway_workspace ws;
+ ws.vws = xmalloc (cmd->n_vars * sizeof (*ws.vws));
+
+ ws.totals = moments1_create (MOMENT_VARIANCE);
+ ws.minimum = DBL_MAX;
+ ws.maximum = -DBL_MAX;
+
+
+ for (v = 0; v < cmd->n_vars; ++v)
+ {
+ struct categoricals *cats = categoricals_create (&cmd->indep_var, 1,
+ cmd->wv, cmd->exclude,
+ makeit,
+ updateit, &ws);
+
+ 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);
+ for (v = 0; v < cmd->n_vars; ++v)
+ {
+ struct categoricals *cats = covariance_get_categoricals (ws.vws[v].cov);
+
+ categoricals_done (cats);
+ }
+
if ( cmd->stats & STATS_HOMOGENEITY )
levene (dict, casereader_clone (input), cmd->indep_var,
cmd->n_vars, cmd->vars, cmd->exclude);
}
}
-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, cmd->dict);
+ show_descriptives (cmd, ws);
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, 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 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;
- 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;
+
+ 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);
+
+ double std_dev = sqrt (variance);
+ double std_error = std_dev / sqrt (n) ;
+
struct string vstr;
+
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;
- tab_double (t, 2, row + count, 0, totals->n, wfmt);
+ moments1_calculate (ws->totals, &n, &mean, &variance, NULL, NULL);
- tab_double (t, 3, row + count, 0, totals->mean, NULL);
+ double std_dev = sqrt (variance);
+ double 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 */
- tab_double (t, 7, row + count, 0,
- totals->mean + T * std_error, NULL);
+ T = gsl_cdf_tdist_Qinv (q, n - 1);
- /* Min and Max */
+ tab_double (t, 6, 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);
+ tab_double (t, 7, row + count, 0,
+ mean + T * std_error, NULL);
+
+ /* Min and Max */
+
+ tab_double (t, 8, row + count, 0, ws->minimum, fmt);
+ tab_double (t, 9, row + count, 0, ws->maximum, fmt);
+ }
- row += gp->n_groups + 1;
+ row += categoricals_total (cats) + 1;
}
tab_submit (t);
/* 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;
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,
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;
+ 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 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);
/* 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;
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_format (t, count + 2, c_num + 2, TAB_RIGHT, "%g", coeffn->coeff);
}
-
- coeffi = ll_next (coeffi);
}
++c_num;
}
/* 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"));
projects / pspp / blobdiff