1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429
|
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <math.h>
#include "autoclass.h"
#include "minmax.h"
#include "globals.h"
/* SUPRESS CODECENTER WARNING MESSSAGES */
/* empty body for 'while' statement */
/*SUPPRESS 570*/
/* formal parameter '<---->' was not used */
/*SUPPRESS 761*/
/* automatic variable '<---->' was not used */
/*SUPPRESS 762*/
/* automatic variable '<---->' was set but not used */
/*SUPPRESS 765*/
void sn_cm_params_influence_fn( model_DS model, tparm_DS tparm, int term_index,int n_att,
float *v, float *class_mean, float *class_sigma, float *class_known_prob,
float *global_mean, float *global_sigma, float *global_known_prob)
{
struct sn_cm_param *param;
tparm_DS *p;
float class_variance, global_variance, v1, v2, v3;
param = &(tparm->ptype.sn_cm);
*class_mean = param->mean;
*class_sigma = param->sigma;
*class_known_prob = param->known_prob;
class_variance = param->variance;
p = model_global_tparms(model);
*global_mean = p[term_index]->ptype.sn_cm.mean;
*global_sigma = p[term_index]->ptype.sn_cm.sigma;
*global_known_prob = p[term_index]->ptype.sn_cm.known_prob;
global_variance = p[term_index]->ptype.sn_cm.variance;
v1 = *class_known_prob *
(param->known_log_prob - p[term_index]->ptype.sn_cm.known_log_prob);
v2 = (1.0 - *class_known_prob) *
(param->unknown_log_prob - p[term_index]->ptype.sn_cm.unknown_log_prob);
v3 = *class_known_prob *
((float) log ((double) (*global_sigma / *class_sigma)) +
(((square(*class_mean - *global_mean) +
(class_variance - global_variance)) / 2.0) / global_variance));
*v = v1 + v2 + v3;
}
/* BUILD_SN_CM_PRIORS
30jul95 wmt: change log calls to safe_log to prevent "log: SING error"
error messages.
Builds an SN-CM prior from the information in a fully instantiated att
structure of the real type.
*/
static priors_DS build_sn_cm_priors( database_DS data_base, att_DS att)
{
int n;
float range, sigma_min, sigma_max;
priors_DS priors;
real_stats_DS statistics = att->r_statistics;/* Attribute range information */
range = statistics->mx - statistics->mn;
sigma_min = SN_CM_SIGMA_SAFETY_FACTOR * att->error;
/* The max sigma for values in range. */
sigma_max = max(sigma_min, range / 2.0);
if (sigma_min == sigma_max) {
if ( (n = att->warnings_and_errors->num_expander_errors ++) == 0)
att->warnings_and_errors->model_expander_errors =
(fxlstr *) malloc(sizeof(fxlstr));
else
att->warnings_and_errors->model_expander_errors =
(fxlstr *) realloc(att->warnings_and_errors->model_expander_errors,
(n+1) * sizeof(fxlstr));
strcpy(att->warnings_and_errors->model_expander_errors[n],
" single_normal_cm is faulty due to large error-to-range\n"
" ratio on sigma priors.\n");
}
priors = (priors_DS) malloc(sizeof(struct priors));
priors->known_prior = (0.5 + statistics->count) / (1.0 + data_base->n_data);
priors->sigma_min = sigma_min;
priors->sigma_max = sigma_max;
priors->mean_mean = statistics->mean;
priors->mean_sigma =
max((range / 2.0),
(SN_CM_SIGMA_SAFETY_FACTOR /
(float) sqrt( ABSOLUTE_MIN_CLASS_WT)) * sigma_min);
priors->mean_var = square(priors->mean_sigma);
priors->minus_log_log_sigmas_ratio =
- (float) log( max( log( 1.0 / (1.0 - SINGLE_FLOAT_EPSILON)),
max( safe_log( (double) (priors->sigma_max / priors->sigma_min)),
LEAST_POSITIVE_SHORT_FLOAT)));
priors->minus_log_mean_sigma = - (float) safe_log( (double) priors->mean_sigma);
return(priors);
}
/* SINGLE_NORMAL_CM_MODEL_TERM_BUILDER
21nov94 wmt: initialize all slots in tparm
30jul95 wmt: change log calls to safe_log to prevent "log: SING error"
error messages.
Funcalled from Expand-Model-Terms. This constructs parameter, prior, and
intermediate results structures appropriate to a single-normal likelihood
term, and places them in the model. Constructs corresponding log-likelihood
and parameter update function elements and saves them on the model for later
compilation.
*/
void single_normal_cm_model_term_builder( model_DS model, term_DS term, int n_term)
{
void ***att_trans_data;
int n, n_att, n_att_trans_data;
float log_att_delta, log_delta_div_root_2pi, error;
att_DS att;
database_DS data_base;
priors_DS prior_set;
tparm_DS tparm;
struct sn_cm_param *sn;
n_att = term->att_list[0];
data_base = model->database;
att = data_base->att_info[n_att];
error = att->error;
att_trans_data = (void ***) get("single_normal_cm", "att_trans_data");
n_att_trans_data = ((int *) get("single_normal_cm", "n_att_trans_data"))[0];
if (getf(att_trans_data, att->type, n_att_trans_data) == NULL)
fprintf( stderr,
"Attribute %d: \"%s\" not one of those allowed for single_normal_cm terms.\n",
n_att, att->dscrp);
if ( att->missing == FALSE) {
if( (n = att->warnings_and_errors->num_expander_warnings ++) == 0)
att->warnings_and_errors->model_expander_warnings =
(fxlstr *) malloc(sizeof(fxlstr));
else
att->warnings_and_errors->model_expander_warnings =
(fxlstr *) realloc(att->warnings_and_errors->model_expander_warnings,
(n+1) * sizeof(fxlstr));
strcpy(att->warnings_and_errors->model_expander_warnings[n],
" using single_normal_cm model on att which has NO missing values\n");
}
if (term->n_atts != 1)
fprintf( stderr,
"Attribute %d: \"%s\": attempting to use single_normal_cm model in a \n"
" non-singleton attribute set\n",
n_att, att->dscrp);
if (error <= 0.0)
fprintf( stderr, "Attribute %d: \"%s\", attempting to use single_normal_cm model "
"with non-positive error value %f.\n",
n_att, att->dscrp, error);
log_att_delta = (float) safe_log((double) error);
log_delta_div_root_2pi = LN_1_DIV_ROOT_2PI + log_att_delta;
/* Allocate parameters struct. */
term->tparm = tparm = (tparm_DS) malloc(sizeof(struct new_term_params));
tparm->tppt=SN_CM;
tparm->n_atts = term->n_atts;
tparm->n_term = n_term;
tparm->n_att = n_att;
tparm->n_att_indices = tparm->n_datum = tparm->n_data = 0;
tparm->wts = tparm->datum = tparm->att_indices = NULL;
tparm->data = NULL;
tparm->w_j = tparm->ranges = tparm->class_wt = 0.0;
tparm->disc_scale = 0.0;
tparm->wt_m = tparm->log_marginal = 0.0;
tparm->log_delta = log_delta_div_root_2pi;
tparm->log_att_delta = log_att_delta;
tparm->log_pi = (float) (-1.0 * log( M_PI));
sn = &(tparm->ptype.sn_cm);
sn->known_wt = sn->known_prob = sn->known_log_prob = sn->unknown_log_prob = 0.0;
sn->weighted_mean = sn->weighted_var = sn->mean = sn->sigma = 0.0;
sn->log_sigma = sn->variance = sn->log_variance = sn->inv_variance = 0.0;
sn->ll_min_diff = sn->skewness = sn->kurtosis = 0.0;
/* Allocate & SET priors. */
prior_set = model->priors[n_term] = build_sn_cm_priors(data_base, att);
sn->prior_mean_mean = prior_set->mean_mean;
sn->prior_mean_var = prior_set->mean_var;
sn->prior_mean_sigma = prior_set->mean_sigma;
sn->prior_sigmas_term = (-1.5 * (float) log( 2.0)) +
prior_set->minus_log_log_sigmas_ratio +
(-1.0 * (float) safe_log((double) prior_set->mean_sigma));
sn->prior_sigma_min_2 = square(prior_set->sigma_min);
sn->prior_sigma_max_2 = square(prior_set->sigma_max);
sn->prior_known_prior = prior_set->known_prior;
/*done above temp->parms->prior_sigma_min_2 = square(prior_set->sigma_min);*/
}
/* SINGLE_NORMAL_CM_LOG_LIKELIHOOD
27nov94 wmt: use percent_equal for float tests
20dec94 wmt: return type to double
23dec94 wmt: check unknown values with FLOAT_UNKNOWN, rather than INT_UNKNOWN
*/
double single_normal_cm_log_likelihood( tparm_DS tparm)
{
int n_att = tparm->n_att;
struct sn_cm_param *sn = &(tparm->ptype.sn_cm);
float log_delta = tparm->log_delta, *datum = tparm->datum, value, diff, temp;
value = datum[n_att];
if( percent_equal( (double) value, FLOAT_UNKNOWN, REL_ERROR))
return(sn->unknown_log_prob);
diff = sn->mean - value;
if ((float) fabs((double) diff) <= sn->ll_min_diff)
temp = 0.0;
else
temp = square(diff) * sn->inv_variance;
return (sn->known_log_prob + log_delta + (-0.5 * (temp + sn->log_variance)));
}
/* SINGLE_NORMAL_CM_UPDATE_L_APPROX
20dec94 wmt: return type to double
When called within the environment of Update-L-Approx-fn, this calculates
the approximate log likelihood log-a<w_j.S_j/H_j.theta_j>_k of observing
the weighted statistics given the class hypothesis and current parameters.
*/
double single_normal_cm_update_l_approx( tparm_DS tparm)
{
struct sn_cm_param *sn=&(tparm->ptype.sn_cm);
float w_j_known, log_delta = tparm->log_delta, diff, t1, t2;
float w_j = tparm->w_j;
w_j_known = sn->known_wt;
diff = sn->weighted_mean - sn->mean;
if (fabs((double) diff) <= sqrt( LEAST_POSITIVE_SINGLE_FLOAT))
t1 = 0.0;
else
t1 = square( diff);
t1 = -0.5 * w_j_known * ((sn->weighted_var + t1) / sn->variance);
t2 = ((w_j - w_j_known) * sn->unknown_log_prob) +
(w_j_known * sn->known_log_prob) +
(w_j_known * (log_delta - (float) log((double) sn->sigma))) + t1;
return (t2);
}
/* SINGLE_NORMAL_CM_UPDATE_M_APPROX
20dec94 wmt: return type to double
29mar95 wmt: calculation to double
When called within the environment of Update-M-Approx-fn, this calculates the
approximate log marginal likelihood log-a<w_j.S_j/H_j>_k of observing the
weighted statistics given the class hypothesis alone. See
Single-Normal-cn-Update-M-approx-term-caller.
A LOG-LINEAR APPROXIMATION IS USED IN THE REGION WHERE
0 <= w_j-known <= (* .75 *absolute-min-class-wt*)
*/
double single_normal_cm_update_m_approx( tparm_DS tparm)
{
struct sn_cm_param *sn=&(tparm->ptype.sn_cm);
float log_att_delta = tparm->log_att_delta;
float prior_mean_mean = sn->prior_mean_mean;
float prior_sigmas_term = sn->prior_sigmas_term, diff, t1, t2;
float prior_mean_sigma = sn->prior_mean_sigma;
float log_pi = tparm->log_pi, w_j = tparm->w_j, w_j_known, t_w_j_known;
double temp;
w_j_known = sn->known_wt;
t_w_j_known = max(w_j_known, 0.75 * ABSOLUTE_MIN_CLASS_WT);
diff = sn->weighted_mean - prior_mean_mean;
if ((float) fabs((double) diff) <=
(prior_mean_sigma *
(float) sqrt( 2.0 * LEAST_POSITIVE_SINGLE_FLOAT)))
t1 = 0.0;
else t1 = -0.5 * square(diff / prior_mean_sigma);
if (w_j_known == t_w_j_known)
t2 = 1.0;
else
t2 = w_j_known / t_w_j_known;
temp = (double) log_pi + (-1.0 * (log_gamma( (double) (w_j + 1.0), FALSE))) +
log_gamma( (double) (0.5 + (w_j - t_w_j_known)), FALSE) +
log_gamma( (double) (0.5 + t_w_j_known), FALSE) +
log_gamma( (double) (0.5 * (t_w_j_known - 1.0)), FALSE) + (double) t1 +
(double) (t_w_j_known * log_att_delta) +
(-0.5 * (double) t_w_j_known *
log( M_PI * (double) t_w_j_known)) +
(-0.5 * (double) (t_w_j_known - 1.0) *
log((double) max( LEAST_POSITIVE_SINGLE_FLOAT, sn->weighted_var))) +
prior_sigmas_term;
return( ((double) t2) * temp);
}
/* SINGLE_NORMAL_CM_UPDATE_PARAMS
27nov94 wmt: use percent_equal for float tests
20dec94 wmt: return type to void
When called within the environment of Update-Params-fn, this updates the
param-set of a Single-Normal-cn term. See
Single-Normal-cn-Update-Params-term-caller.
Revised 12Feb90 JCS
Use of double precision for weighted calculations will triple the runtime.
*/
void single_normal_cm_update_params( tparm_DS tparm, int known_parms_p)
{
int n_att = tparm->n_att, n_data = tparm->n_data;
struct sn_cm_param *sn_cm=&(tparm->ptype.sn_cm);
float prior_sigma_min_2 = sn_cm->prior_sigma_min_2;
float prior_sigma_max_2 = sn_cm->prior_sigma_max_2;
float prior_mean_mean = sn_cm->prior_mean_mean;
float prior_mean_var = sn_cm->prior_mean_var;
float prior_known_prior = sn_cm->prior_known_prior;
float **data = tparm->data, *wts = tparm->wts, prob_known, var_ratio;
float class_wt = tparm->class_wt, class_wt_1;
float ignore1, known, mean, variance, skewness, kurtosis;
class_wt_1 = class_wt + 1;
if (class_wt > 0.0) { /* Zero class-wt implies null class */
/* Update the class statistics from class-DS-wts & database */
/* If not collect?, we proceed with the previous values. */
if (tparm->collect == TRUE) {
central_measures_x(data, n_data, n_att, wts,
percent_equal( (double) sn_cm->mean, FLOAT_UNKNOWN,
REL_ERROR) ?
(double) prior_mean_mean : (double) sn_cm->mean,
&ignore1, &known, &mean, &variance, &skewness, &kurtosis);
if (known == 0.0) {
sn_cm->known_wt = 0.0;
sn_cm->weighted_mean = prior_mean_mean;
sn_cm->weighted_var = prior_mean_var;
}
else {
sn_cm->known_wt = known;
/* commented fprintf(stderr, "Setting known_wt to known %f\n", known);*/
sn_cm->weighted_mean = mean;
sn_cm->weighted_var =
max(prior_sigma_min_2, min(prior_sigma_max_2, variance));
sn_cm->skewness = skewness;
sn_cm->kurtosis = kurtosis;
}
}
}
else {
sn_cm->known_wt = 0.0;
sn_cm->weighted_mean = prior_mean_mean;
sn_cm->weighted_var = prior_mean_var;
}
if (known_parms_p != TRUE) {
prob_known = max(LEAST_POSITIVE_SINGLE_FLOAT,
(sn_cm->known_wt + prior_known_prior) / class_wt_1);
sn_cm->known_prob = prob_known;
sn_cm->known_log_prob = (float) log((double) prob_known);
sn_cm->unknown_log_prob =
(float) log((double) max( LEAST_POSITIVE_SINGLE_FLOAT,
1.0 - prob_known));
var_ratio = sn_cm->weighted_var / (class_wt_1 * prior_mean_var);
sn_cm->mean = (sn_cm->weighted_mean * (1.0 - var_ratio)) +
(prior_mean_mean * var_ratio);
sn_cm->variance = max(sn_cm->weighted_var * (class_wt / class_wt_1),
(float) sqrt( LEAST_POSITIVE_SINGLE_FLOAT));
sn_cm->sigma = (float) sqrt((double) sn_cm->variance);
sn_cm->log_variance = (float) log((double) sn_cm->variance);
sn_cm->log_sigma = 0.5 * sn_cm->log_variance;
sn_cm->inv_variance = 1.0 / sn_cm->variance;
sn_cm->ll_min_diff =
max((float) sqrt( LEAST_POSITIVE_SINGLE_FLOAT),
(sn_cm->variance *
(float) sqrt( LEAST_POSITIVE_SINGLE_FLOAT)));
}
/* return(class_wt); */
}
/* When called within the environment of Class-Equivalence-fn, this tests for a
difference of means less than sigma=ratio times MIN of sigmas. */
int single_normal_cm_class_equivalence( tparm_DS tparm1,tparm_DS tparm2, double sigma_ratio)
{
struct sn_cm_param *sn1 = &(tparm1->ptype.sn_cm);
struct sn_cm_param *sn2 = &(tparm2->ptype.sn_cm);
if (fabs((double) (sn1->mean - sn2->mean)) <
(sigma_ratio * (double) min(sn1->sigma, sn2->sigma)))
return(TRUE);
else
return(FALSE);
}
/* SINGLE_NORMAL_CM_CLASS_MERGED_MARGINAL
20dec94 wmt: return type to void
When called within the environment of Class-Merged-Marginal-fn, this
generates the sufficient statistics of Single-Normal-cn term equivalent
to the weighted merging of params-0 and params-1, storing same in
params-m.
*/
void single_normal_cm_class_merged_marginal( tparm_DS tparm0,tparm_DS tparm1,tparm_DS tparmm)
{
struct sn_cm_param *sn0=&(tparm0->ptype.sn_cm);
struct sn_cm_param *sn1=&(tparm1->ptype.sn_cm);
struct sn_cm_param *snm=&(tparmm->ptype.sn_cm);
float prior_sigma_min_2 = sn0->prior_sigma_min_2;
float kwt0, kwt1, kwtm;
kwt0 = sn0->known_wt;
kwt1 = sn1->known_wt;
kwtm = kwt0 + kwt1;
snm->known_wt = kwtm;
fprintf(stderr, "Set known_wt to %f\n", kwtm);
if (kwtm != 0.0) {
snm->weighted_mean =
((kwt0 * sn0->weighted_mean) + (kwt1 * sn1->weighted_mean)) / kwtm;
snm->weighted_var =
max(prior_sigma_min_2,
(((kwt0 * (square(sn0->weighted_mean) + sn0->weighted_var)) +
(kwt1 * (square(sn1->weighted_mean) + sn1->weighted_var))) /
kwtm) -
square(snm->weighted_mean));
/* return (snm->weighted_var); **************/
}
/* else return (FLOAT_UNKNOWN); ****************** nil? */
}
|