BOOL Update_def_Settings( def_Settings * defsetting )
{
char buf[LENGTH] ;
char msg[20] ;
BOOL batchgoon = TRUE ;
char * pstr ;
unsigned int sz = 0 ;
unsigned int serialno = 0 ;
if (NULL == defsetting)
return FALSE ;
defsetting->time = 0 ;
if ( FALSE == Is_Data_Empty(&defsetting->pairs) )
{
// create next input file name
pstr = strrchr( defsetting->inputfile, '.') ; // 46
if ( NULL != pstr )
sz = abs( pstr - defsetting->inputfile ) ;
else
{
sz = strlen(defsetting->inputfile) ;
return FALSE ;
}
strcpy (buf, pstr+1) ;
serialno = atoi(buf) ;
serialno += 1 ;
strncpy( buf, defsetting->inputfile , sz ) ;
buf[sz] = '\0' ;
strcat( buf, "." ) ;
// add appendix
sprintf(msg,"%d",serialno) ;
strcat( buf, msg ) ;
// save inputfile name in def_Settings
if ( NULL != defsetting->inputfile )
{
free( (void*) defsetting->inputfile ) ;
defsetting->inputfile = NULL ;
}
if ( NULL == ( defsetting->inputfile = strdup(buf) ) )
batchgoon = FALSE ;
// create next test file name
pstr = strrchr( defsetting->testfile, '.') ; // 46
if ( NULL != pstr )
sz = abs( pstr - defsetting->testfile ) ;
else
sz = strlen(defsetting->testfile) ;
strcpy (buf, pstr+1) ;
serialno = atoi(buf) ;
serialno += 1 ;
strncpy( buf, defsetting->testfile , sz ) ;
buf[sz] = '\0' ;
strcat( buf, "." ) ;
// add appendix
sprintf(msg,"%d",serialno) ;
strcat( buf, msg ) ;
// save testfile name in def_Settings
if ( NULL != defsetting->testfile )
{
free( (void*) defsetting->testfile ) ;
defsetting->testfile = NULL ;
}
if ( NULL == ( defsetting->testfile = strdup(buf) ) )
batchgoon = FALSE ;
// load data into pairs
if ( FALSE == smo_Loadfile(&(defsetting->pairs), defsetting->inputfile, 0) )
batchgoon = FALSE ;
}
else
{
// load data into pairs
if ( FALSE == smo_Loadfile(&(defsetting->pairs), defsetting->inputfile, 0) )
{
batchgoon = FALSE ;
printf("Failed to load training data from the file %s\n", defsetting->inputfile) ;
}
#ifdef _PROSTATE_VIVO
else if (defsetting->pairs.count!=100||defsetting->pairs.dimen!=6117||CLASSIFICATION!=defsetting->pairs.datatype)
{
batchgoon = FALSE ;
printf("The file %s is not ProstateABreast.VIVO.\n", defsetting->inputfile) ;
}
#endif
}
defsetting->cache_size = DEF_CACHE ;
defsetting->time = 0 ;
defsetting->lnC_step = defsetting->def_lnC_step ;
defsetting->lnK_step = defsetting->def_lnK_step;
defsetting->lnC_start = defsetting->def_lnC_start ;
defsetting->lnK_start = defsetting->def_lnK_start ;
defsetting->lnC_end = defsetting->def_lnC_end ;
defsetting->lnK_end = defsetting->def_lnK_end ;
// load data into pairs
// load testing data into pairs
//if (batchgoon == TRUE)
// if ( FALSE == smo_Loadfile(&(defsetting->testdata), defsetting->testfile, defsetting->pairs.dimen) )
// printf ("Failed to load testing data from the file %s", defsetting->testfile ) ;
return batchgoon ;
}
int main( int argc, char * argv[])
{
def_Settings * defsetting = NULL ;
smo_Settings * smosetting = NULL ;
kcv_Settings * kcvsetting ;
//Data_Node * node ;
char buf[LENGTH] ;
unsigned int sz = 0;
unsigned int index = 0 ;
double parameter = 0 ;
//double * guess ;
FILE * log ;
printf("\nSupport Vector Ordinal Regression Using K-fold Cross Validation v2.%d \n--- Chu Wei Copyright(C) 2003-2004\n\n", VERSION) ;
if ( 1 == argc || NULL == (defsetting = Create_def_Settings(argv[--argc])) )
{
// display help
printf("\nUsage: svor [-v] [...] [-R r] file \n\n") ;
printf(" file specifies the file containing training samples.\n") ;
printf(" -v activates the verbose mode to display message.\n") ;
printf(" -F k k-fold is used in cross validation (default 5 folds).\n") ;
printf(" -Cs s set searching step of C at s in log10 scale (default 0.5).\n") ;
printf(" -Cc c start searching C at c (default 0.1).\n") ;
printf(" -Ce e stop searching C at e (default 100).\n") ;
printf(" -Ks s set searching step of K at s in log10 scale (default 0.5).\n") ;
printf(" -Kc c start searching K at c (default 0.01).\n") ;
printf(" -Ke e stop searching K at e (default 10).\n") ;
//printf(" -L use imbalanced Linear kernel (default Gaussian kernel).\n") ;
printf(" -P p use Polynomial kernel with order p (default Gaussian kernel).\n") ;
//printf(" -B force regularizers Balanced for classification only.\n") ;
//printf(" -E e set Epsilon at e for regression only (default 0.1).\n") ;
printf(" -Z z set the times of Zooming at z (default 2).\n") ;
printf(" -i normalize the training inputs.\n") ;
//printf(" -o normalize the training targets.\n") ;
printf(" -a activates loading weighted kernels.\n") ;
printf(" -S s set the Seed of random number generator at s (default random).\n") ;
printf(" -R r set the Rehearsal times of fold generation at r (default 1).\n") ;
printf(" -T t set Tolerance at t (default 0.001).\n") ;
printf("new option:\n") ;
printf(" -Z 0 set the zooming at 0 to skip cross-validation.\n") ;
printf(" -Ko o fix K at o manually (default 1).\n") ;
printf(" -Co o fix C at o manually (default 1).\n") ;
printf("\n") ;
if (NULL !=defsetting)
Clear_def_Settings( defsetting ) ;
return 0;
}
else
{
if (argc>1)
printf("Options:\n") ;
do
{
strcpy(buf, argv[--argc]) ;
sz = strlen(buf) ;
//printf ("%s %d\n", buf, sz) ;
if ( '-' == buf[0] )
{
for (index = 1 ; index < sz ; index++)
{
switch (buf[index])
{
case 'v' :
printf(" - Verbose mode in display.\n") ;
defsetting->smo_display = TRUE ;
break ;
case 'a' :
printf(" - kernels with Ard parameters.\n") ;
defsetting->ardon = TRUE ;
break ;
case 'o' :
printf(" - normalize the Outputs in training data.\n") ;
defsetting->normalized_output = TRUE ;
defsetting->pairs.normalized_output = TRUE ;
break ;
case 'i' :
printf(" - normalize the Inputs in training data.\n") ;
defsetting->normalized_input = TRUE ;
defsetting->pairs.normalized_input = TRUE ;
break ;
case 'S' :
if (parameter>0)
{
printf(" - specify %.0f as the seed of random number.\n", parameter) ;
defsetting->seeds = (unsigned int)parameter ;
}
break ;
case 'F' :
if (parameter>1)
{
printf(" - %.0f-fold cross validation.\n", parameter) ;
defsetting->kfold = (unsigned int)parameter ;
}
break ;
case 'Z' :
if (parameter>=0)
{
printf(" - Zoom in scale %.0f.\n", parameter) ;
defsetting->loops = (unsigned int)parameter ;
}
break ;
case 'R' :
if (parameter>=1)
{
printf(" - Rehearsal %.0f times.\n", parameter) ;
defsetting->repeat = (unsigned int)parameter ;
}
break ;
case 'L' :
printf(" - choose Linear kernel.\n") ;
defsetting->kernel = LINEAR ;
break ;
case 'B' :
printf(" - set regularization factors balanced.\n") ;
defsetting->smo_balance = TRUE ;
break ;
case 'T' :
if (parameter>0)
{
printf(" - set Tol as %.6f.\n", parameter) ;
defsetting->tol = parameter ;
}
break ;
case 'C' :
if (parameter > 0)
{
if (index + 1 == sz && parameter != DEF_COARSESTEP )
{
printf(" - C is invalid.\n") ;
parameter = 0;
}
else if (index+1==sz-1)
{
index+=1;
switch (buf[index])
{
case 's' :
defsetting->def_lnC_step = parameter ;
printf(" - initialize lnC_step at the value %f.\n", parameter) ;
parameter = 0 ;
break ;
case 'c' :
defsetting->def_lnC_start = log10(parameter) ;
printf(" - C start at %f.\n", parameter) ;
parameter = 0 ;
break ;
case 'e' :
defsetting->def_lnC_end = log10(parameter) ;
printf(" - C end at %f.\n", parameter) ;
parameter = 0 ;
break ;
case 'o' :
defsetting->vc = (parameter) ;
printf(" - C at %f.\n", parameter) ;
parameter = 0 ;
break ;
default :
printf(" - C%c is invalid.\n", buf[index]) ;
break ;
}
}
}
break ;
case 'K' :
if (parameter > 0)
{
if (index + 1 == sz && parameter != DEF_COARSESTEP )
{
printf(" - K is invalid.\n") ;
parameter = 0;
}
else if (index+1==sz-1)
{
index+=1;
switch (buf[index])
{
case 's' :
defsetting->def_lnK_step = parameter ;
printf(" - initialize lnK_step at the value %f.\n", parameter) ;
parameter = 0 ;
break ;
case 'c' :
defsetting->def_lnK_start = log10(parameter) ;
printf(" - K start at %f.\n", parameter) ;
parameter = 0 ;
break ;
case 'e' :
defsetting->def_lnK_end = log10(parameter) ;
printf(" - K end at %f.\n", parameter) ;
parameter = 0 ;
break ;
case 'o' :
defsetting->kappa = (parameter) ;
printf(" - K at %f.\n", parameter) ;
parameter = 0 ;
break ;
default :
printf(" - K%c is invalid.\n", buf[index]) ;
break ;
}
}
}
break ;
case 'P' :
if (parameter >= 1)
{
defsetting->kernel = POLYNOMIAL ;
defsetting->p = (unsigned int) parameter ;
printf(" - choose Polynomial kernel with order %d.\n", defsetting->p) ;
parameter = 0 ;
defsetting->def_lnK_start = 0 ;
defsetting->def_lnK_end = 0 ;
}
break ;
default :
if ('-' != buf[index])
printf(" -%c is invalid.\n", buf[index]) ;
break ;
}
}
}
else
parameter = atof(buf) ;
}
while ( argc > 1 ) ;
printf("\n") ;
}
log = fopen ("validation_implicit.log", "w+t") ;
if (NULL != log)
fclose(log) ; // clear the old file.
while ( TRUE == Update_def_Settings(defsetting) )
{
/* sz = defsetting->loops ;
while (sz > 0)
{
// coarse search
kcvsetting = Create_Kcv ( defsetting ) ;
while (kcvsetting->index<defsetting->repeat)
{
Rehearsal_Kcv ( kcvsetting, defsetting ) ;
Init_Kcv ( kcvsetting, defsetting ) ;
kcvsetting->index += 1 ;
}
// update defsetting
defsetting->lnC_start = kcvsetting->best_lnC - defsetting->lnC_step*(1.0-1/defsetting->zoomin) ;
defsetting->lnC_end = kcvsetting->best_lnC + defsetting->lnC_step*(1.0-1/defsetting->zoomin) ;
defsetting->lnK_start = kcvsetting->best_lnK - defsetting->lnK_step*(1.0-1/defsetting->zoomin) ;
defsetting->lnK_end = kcvsetting->best_lnK + defsetting->lnK_step*(1.0-1/defsetting->zoomin) ;
defsetting->lnC_step = defsetting->lnC_step/defsetting->zoomin ;
defsetting->lnK_step = defsetting->lnK_step/defsetting->zoomin ;
defsetting->time += kcvsetting->time ;
defsetting->best_rate = kcvsetting->best_rate ;
defsetting->vc = pow(10.0, kcvsetting->best_lnC) ;
defsetting->kappa = pow(10.0, kcvsetting->best_lnK) ;
if (1==sz)
{
log = fopen ("validation_implicit.log", "a+t") ;
if (NULL != log)
{
fprintf(log,"%lu %u %u %f %f %f %.6f\n", defsetting->pairs.dimen, defsetting->kfold, defsetting->loops, log10(defsetting->vc), log10(defsetting->kappa), defsetting->best_rate, defsetting->time) ;
fclose(log) ;
}
}
Clear_Kcv ( kcvsetting ) ;
sz -= 1 ;
}
*/
// save validation output
defsetting->lnC_start = log10(defsetting->vc) ;
defsetting->lnC_end = log10(defsetting->vc) ;
defsetting->lnK_start = log10(defsetting->kappa) ;
defsetting->lnK_end = log10(defsetting->kappa) ;
defsetting->lnC_step = defsetting->lnC_step ;
defsetting->lnK_step = defsetting->lnK_step ;
/*
guess = (double *) calloc(defsetting->pairs.count,sizeof(double)) ;
kcvsetting = Create_Kcv ( defsetting ) ;
while (kcvsetting->index<defsetting->repeat)
{
Rehearsal_Kcv ( kcvsetting, defsetting ) ;
Init_Kcv ( kcvsetting, defsetting ) ;
kcvsetting->index += 1 ;
sz = 0 ;
node = defsetting->pairs.front ;
while(NULL != node)
{
guess[sz] += node->fx ;
node = node->next ;
sz += 1 ;
}
}
sprintf(buf, "%s.validation", defsetting->inputfile) ;
parameter = 0 ;
log = fopen (buf, "w+t") ;
if (NULL != log)
{
printf("save validation output in %s.\n", buf) ;
node = defsetting->pairs.front ;
sz = 0 ;
while(NULL != node)
{
fprintf(log,"%f ", guess[sz]/(double)defsetting->repeat) ;
fprintf(log," %u\n", node->target) ;
node = node->next ;
sz += 1 ;
}
fclose(log) ;
}
free (guess) ;
Clear_Kcv ( kcvsetting ) ;//
*/
defsetting->training.count = defsetting->pairs.count ;
defsetting->training.front = defsetting->pairs.front ;
defsetting->training.rear = defsetting->pairs.rear ;
defsetting->training.classes = defsetting->pairs.classes ;
defsetting->training.dimen = defsetting->pairs.dimen ;
defsetting->training.featuretype = defsetting->pairs.featuretype ;
// create smosettings
printf ("\n\n TESTING on %s...\n", defsetting->testfile ) ;
smosetting = Create_smo_Settings(defsetting) ;
smosetting->pairs = &defsetting->pairs ;
defsetting->training.count = 0 ;
defsetting->training.front = NULL ;
defsetting->training.rear = NULL ;
defsetting->training.featuretype = NULL ;
// load test data
if ( FALSE == smo_Loadfile(&(defsetting->testdata), defsetting->testfile, defsetting->pairs.dimen) )
{
printf ("No testing data found in the file %s.\n", defsetting->testfile ) ;
svm_saveresults (&defsetting->pairs, smosetting) ;
}
// calculate the test output
else
{
smo_routine (smosetting) ;
svm_predict (&defsetting->testdata, smosetting) ;
svm_saveresults (&defsetting->testdata, smosetting) ;
if (ORDINAL == smosetting->pairs->datatype)
printf ("\r\nTEST ERROR NUMBER %.0f, AAE %.0f and SVs %.0f, at C=%.3f Kappa=%.3f with %.3f seconds.\r\n",
smosetting->testerror*defsetting->testdata.count,smosetting->testrate*defsetting->testdata.count, smosetting->svs, smosetting->vc, smosetting->kappa, smosetting->smo_timing) ;
if (NULL != (log = fopen ("kfoldsvc.log", "a+t")) )
{
if (REGRESSION == smosetting->pairs->datatype)
fprintf(log,"%d-fold: TEST ASE %f, AAE %f and SVs %.0f at C=%f and Kappa=%f with %.3f seconds.\r\n", defsetting->kfold, smosetting->testrate, smosetting->testerror, smosetting->svs, smosetting->vc, smosetting->kappa, smosetting->smo_timing) ;
else
fprintf(log,"%d-fold: TEST ERROR %f, RATE %f and SVs %.0f at C=%f and Kappa=%f with %.3f seconds.\r\n", defsetting->kfold, smosetting->testerror, smosetting->testrate, smosetting->svs, smosetting->vc, smosetting->kappa, smosetting->smo_timing) ;
fclose(log) ;
}
// write another log
if (ORDINAL != smosetting->pairs->datatype)
{
if (NULL != (log = fopen ("ordinal.log", "a+t")) )
{
fprintf(log,"%.0f %.0f %f %.3f\n", smosetting->testerror*defsetting->testdata.count, smosetting->testrate*defsetting->testdata.count, smosetting->testrate, smosetting->smo_timing) ;
fclose(log) ;
}
}
if (ORDINAL == smosetting->pairs->datatype)
{
if (NULL != (log = fopen ("ordinal_implicit.log", "a+t")) )
{
fprintf(log,"%.0f %.0f %f %f\n", smosetting->testerror*defsetting->testdata.count, smosetting->testrate*defsetting->testdata.count, smosetting->testrate, smosetting->smo_timing) ;
fclose(log) ;
}
}
}
Clear_smo_Settings( smosetting ) ;
}
// free memory then exit
Clear_def_Settings( defsetting ) ;
return 0;
}