BOOL smo_Loadfile ( Data_List * pairs, char * inputfilename, int inputdim )
{
FILE * smo_stream ;
FILE * smo_target = NULL ;
char * pstr = NULL ;
char buf[LENGTH] ;
char * temp ;
int dim = -2 ;
unsigned long index = 1 ;
unsigned int result, sz ;
int var = 0, chg = 0 ;
double * point = NULL ;
unsigned int y ;
int i = 0, j = 0 ;
double mean = 0 ;
double ymax = LONG_MIN ;
double ymin = LONG_MAX ;
double * xmean = NULL;
Data_Node * node = NULL ;
int t0=0, tr=0 ;
FILE * fid ;
Data_List label ;
if ( NULL == pairs || NULL == inputfilename )
return FALSE ;
Clear_Data_List( pairs ) ;
Create_Data_List( &label ) ;
if( (smo_stream = fopen( inputfilename, "r+t" )) == NULL )
{
//printf( "can not open the file %s.\n", inputfilename );
return FALSE ;
}
// save file name
var = strlen( inputfilename ) ;
if (NULL != pairs->filename)
free(pairs->filename) ;
pairs->filename = (char*)malloc((var+1)*sizeof(char)) ;
if (NULL == pairs->filename)
{
printf("fail to malloc for pairs->filename.\n") ;
exit(0) ;
}
strncpy(pairs->filename,inputfilename,var) ;
pairs->filename[var]='\0' ;
// check the input dimension here
if ( NULL == fgets( buf, LENGTH, smo_stream ))
{
printf( "fgets error in reading the first line.\n" );
fclose( smo_stream );
return FALSE ;
}
var = strlen( buf ) ;
if (var >= LENGTH-1)
{
printf( "the line is too long in the file %s.\n", inputfilename );
fclose( smo_stream );
return FALSE ;
}
if (0 < var)
{
do
{
dim = dim + 1 ;
strtod( buf, &temp ) ;
strcpy( buf, temp ) ;
chg = var - strlen(buf) ;
var = var - chg ;
}
while ( 0 != chg ) ;
}
else
{
fclose( smo_stream );
printf("the first line in the file is empty.\n") ;
return FALSE ;
}
if ( 0 > dim || (0 == dim && 0 == inputdim) )
{
fclose( smo_stream );
#ifdef SMO_DEBUG
printf( "input dimension is less than one.\n") ;
#endif
return FALSE ;
}
if (inputdim > 0)
{
if (inputdim == dim + 1 ) // test file without target
{
// try to open "*target*.*" as target
// create target file name
pstr = strstr( inputfilename , "test" ) ;
if (NULL != pstr)
{
result = abs( inputfilename - pstr ) ;
strncpy (buf, inputfilename, result ) ;
buf[result] = '\0' ;
strcat(buf, "targets") ;
strcat (buf, pstr+4) ;
smo_target = fopen( buf, "r+t" ) ;
}
dim = inputdim ;
pairs->dimen = dim ;
}
else if ( inputdim != dim )
{
printf("Dimensionality in testdata is inconsistent with traindata.\n") ;
return FALSE ;
}
else
pairs->dimen = dim ;
}
else
pairs->dimen = dim ;
//initialize the x_mean and x_devi in Data_List pairs
if ( NULL == (pairs->x_mean = (double *)(malloc(dim*sizeof(double))) )
|| NULL == (pairs->x_devi = (double *)(malloc(dim*sizeof(double))) )
|| NULL == (xmean = (double *)(malloc(dim*sizeof(double))) ) )
{
if (NULL != pairs->x_mean)
free(pairs->x_mean) ;
if (NULL != pairs->x_devi)
free(pairs->x_devi) ;
if (NULL != xmean)
free(xmean) ;
if (NULL != smo_target)
fclose( smo_target ) ;
if (NULL != smo_stream)
fclose( smo_stream );
return FALSE ;
}
for ( j = 0; j < dim; j ++ )
pairs->x_mean[j] = 0 ;
for ( j = 0; j < dim; j ++ )
pairs->x_devi[j] = 0 ;
for ( j = 0; j < dim; j ++ )
xmean[j] = 0 ;
// begin to initialize data_list for digital input only
printf("\nLoading %s ... \n", inputfilename) ;
pairs->datatype = CLASSIFICATION ;
rewind( smo_stream ) ;
fgets( buf, LENGTH, smo_stream ) ;
do
{
#ifdef SMO_DEBUG
printf("%d\n", index) ;
printf("%s\n\n\n", buf) ;
#endif
point = (double *) malloc( (dim+1) * sizeof(double) ) ; // Pairs to free them
if ( NULL == point )
{
printf("not enough memory.\n") ;
if (NULL != smo_target)
fclose( smo_target ) ;
if (NULL != smo_stream)
fclose( smo_stream );
if (NULL != pairs->x_mean)
free(pairs->x_mean) ;
if (NULL != pairs->x_devi)
free(pairs->x_devi) ;
if (NULL != xmean)
free(xmean) ;
Clear_Data_List( pairs ) ;
return FALSE ;
}
var = strlen( buf ) ;
i = 0 ;
chg = dim ;
while ( chg>0 && i<dim)
{
point[i] = strtod( buf, &temp ) ;
i++ ;
strcpy( buf, temp ) ;
chg = var - strlen(buf) ;
var = var - chg ;
}
point[dim]=0 ;
if (i==dim && chg>0 && var>0)
y = (unsigned int)strtod( buf, &temp ) ;
else
{
free(point) ;
y = 0 ;
printf("Warning: the input file %s contains a blank or defective line.\n",inputfilename) ;
exit(1) ;
}
// load y as target from other file when dim+1
if (NULL != smo_target)
{
if ( NULL != fgets( buf, LENGTH, smo_target ) )
{
var = strlen( buf ) ;
y = (int)strtod( buf, &temp ) ;
strcpy( buf, temp ) ;
chg = var - strlen(buf) ;
if (0==chg)
printf("Warning: the target file contains a blank line.\n") ;
}
else
printf("Warning: the target file is shorter than the input file.\n") ;
}
/*
for ( i = 0; i < dim; i ++ )
{
point[i] = strtod( buf, &temp ) ;
strcpy( buf, temp ) ;
}
y = strtod( buf, &temp ) ;
// load y as target from other file when dim+1
if (NULL != smo_target)
{
fgets( buf, LENGTH, smo_target ) ;
y = strtod( buf, &temp ) ;
}
*/
if (chg>0)
{
if ( TRUE == Add_Data_List( pairs, Create_Data_Node(index, point, y) ) )
{
// update statistics
pairs->mean = (mean * (((double)(pairs->count)) - 1) + y )/ ((double)(pairs->count)) ;
pairs->deviation = pairs->deviation + (y-mean)*(y-mean) * ((double)(pairs->count)-1)/((double)(pairs->count));
mean = pairs->mean ;
for ( j=0; j<dim; j++ )
{
pairs->x_mean[j] = (xmean[j] * (((double)(pairs->count)) - 1) + point[j] )/ ((double)(pairs->count)) ;
pairs->x_devi[j] = pairs->x_devi[j] + (point[j]-xmean[j])*(point[j]-xmean[j]) * ((double)(pairs->count)-1)/((double)(pairs->count));
xmean[j] = pairs->x_mean[j] ;
}
if (y>ymax)
{ ymax = y ; pairs->i_ymax = index ;}
if (y<ymin)
{ ymin = y ; pairs->i_ymin = index ;}
// check data type
Add_Label_Data_List( &label, Create_Data_Node(index, point, y) ) ;
index ++ ;
}
else
{
#ifdef SMO_DEBUG
printf("%d\n", index) ;
printf("duplicate data \n") ;
#endif
}
}
}
while( !feof( smo_stream ) && NULL != fgets( buf, LENGTH, smo_stream ) ) ;
if (label.count>=2||inputdim>0)
pairs->datatype = ORDINAL ;
else
printf("Warning : not a ordinal regression.\n") ;
if (pairs->count < MINNUM || (pairs->datatype == UNKNOWN && inputdim == 0 ) )
{
printf("too few input pairs\n") ;
Clear_Data_List( pairs ) ;
if (NULL != pairs->x_mean)
free(pairs->x_mean) ;
if (NULL != pairs->x_devi)
free(pairs->x_devi) ;
if (NULL != xmean)
free(xmean) ;
if (NULL != smo_target)
fclose( smo_target ) ;
if (NULL != smo_stream)
fclose( smo_stream );
return FALSE ;
}
// load index file for feature types strstr
pairs->featuretype = (int *) malloc(pairs->dimen*sizeof(int)) ;
if (NULL != pairs->featuretype)
{
//default 0
for (sz=0;sz<pairs->dimen;sz++)
pairs->featuretype[sz] = 0 ;
if (0==inputdim)
pstr = strstr( inputfilename, "train") ; // 46
else
pstr = strstr( inputfilename, "test") ; // 46
if (NULL != pstr)
{
sz = abs( pstr - inputfilename ) ;
pstr = strrchr( inputfilename, '.') ; // 46
strncpy( buf, inputfilename, sz ) ;
buf[sz]='\0' ;
strcat( buf, "feature" ) ;
strcat( buf, pstr ) ;
fid = fopen(buf,"r+t") ;
if (NULL != fid)
{
printf("Loading the specifications of feature type in %s ...",buf) ;
sz = 0 ;
while (!feof(fid) && NULL!=fgets(buf,LENGTH,fid) )
{
i=strlen(buf) ;
if (i>1)
{
if (sz>=pairs->dimen)
{
printf("Warning : feature type file is too long.\n") ;
sz = pairs->dimen-1 ;
}
pairs->featuretype[sz] = atoi(buf) ;
sz += 1 ;
}
else
printf("Warning : blank line in feature type file.\n") ;
}
if (sz!=pairs->dimen)
{
//default 0
for (sz=0;sz<pairs->dimen;sz++)
pairs->featuretype[sz] = 0 ;
printf(" RESET as default.\n") ;
}
else
printf(" done.\n") ;
fclose(fid) ;
}
}
}
pairs->deviation = sqrt( pairs->deviation / ((double)(pairs->count - 1.0)) ) ;
for ( j=0; j<dim; j++ )
pairs->x_devi[j] = sqrt( pairs->x_devi[j] / ((double)(pairs->count - 1.0)) ) ;
// set target value as +1 or -1, if data type is CLASSIFICATION
if ( UNKNOWN != pairs->datatype && 0 == inputdim )
{
pairs->deviation = 1.0 ;
pairs->mean = 0 ;
pairs->normalized_output = FALSE ;
}
for ( j=0; j<dim; j++ )
{
if (pairs->featuretype[j] != 0)
{
pairs->x_devi[j] = 1 ;
pairs->x_mean[j] = 0 ;
}
}
if (inputdim>0) // do not normailize data for TESTING
{
pairs->normalized_output = FALSE ;
pairs->normalized_input = FALSE ;
}
// normalize the target if needed
node = pairs->front ;
while ( node != NULL )
{
if ( TRUE == pairs->normalized_input )
{
for ( j=0; j<dim; j++ )
{
if (pairs->x_devi[j]>0)
node->point[j] = (node->point[j]-pairs->x_mean[j])/(pairs->x_devi[j]) ;
else
node->point[j] = 0 ;
}
}
node = node->next ;
}
printf("Total %lu samples with %u dimensions for ", pairs->count, pairs->dimen) ;
if (inputdim > 0)
printf("TESTING.\r\n") ;
else
{
if( CLASSIFICATION == pairs->datatype )
printf("CLASSIFICATION.\r\n") ;
else if ( ORDINAL == pairs->datatype )
{
printf("ORDINAL %lu REGRESSION.\r\n",label.count) ;
pairs->classes = label.count ;
if (NULL != pairs->labels)
free( pairs->labels ) ;
i=0;
pairs->labels = (unsigned int*)malloc(pairs->classes*sizeof(unsigned int)) ;
pairs->labelnum = (unsigned int*)malloc(pairs->classes*sizeof(unsigned int)) ;
if (NULL != pairs->labels&&NULL != pairs->labelnum)
{
node = label.front ;
printf("ordinal varibles : ") ;
while (NULL!=node)
{
if (node->target<1 || node->target>pairs->classes)
{
printf("Error : targets should be from 1 to %d.\n",(int)pairs->classes) ;
exit(1) ;
}
pairs->labels[node->target-1] = node->target ;
if (node->target-1==0)
t0 = node->target ;
if (node->target==(int)pairs->classes)
tr = node->target ;
pairs->labelnum[node->target-1] = node->fold ;
i += node->fold ;
printf("%d(%d) ", node->target, node->fold) ;
node = node->next ;
}
printf("\n") ;
if (i!=(int)pairs->count||t0!=1||tr!=(int)pairs->classes)
{
printf("Error in data list.\n") ;
exit(1) ;
}
}
else
{
printf("fail to malloc for pairs->labels.\n") ;
exit(1) ;
}
}
else
printf("UNKNOWN.\r\n") ;
}
if (1 == pairs->normalized_input)
printf("Inputs are normalized.\r\n") ;
if (1 == pairs->normalized_output && pairs->deviation > 0)
printf("Outputs are normalized.\r\n") ;
if ( inputdim > 0 && pairs->deviation <= 0 )
printf("Tragets are not at hand.\r\n") ;
Clear_Label_Data_List (&label) ;
if (NULL != smo_target)
fclose( smo_target ) ;
if (NULL != smo_stream)
fclose( smo_stream );
if ( NULL != xmean )
free( xmean ) ;
return TRUE ;
}
def_Settings * Create_def_Settings ( char * filename )
{
def_Settings * settings = NULL ;
char * pstr = NULL ;
unsigned int result = 0 ;
char buf[LENGTH] = "" ;
if (NULL == filename)
{
printf("\r\nFATAL ERROR : the input pointer is NULL.\r\n") ;
return NULL ;
}
if (NULL == (settings = (def_Settings *)(malloc(sizeof(def_Settings)))))
{
printf("\r\nFATAL ERROR : fail to malloc def_settings.\r\n") ;
return NULL ;
}
// set all elements as default values
INPUTFILE = NULL ;
TESTFILE = NULL ;
VC = DEF_VC ;
TOL = DEF_TOL ;
KAPPA = DEF_KAPPA ;
EPS = DEF_EPS ;
P = DEF_P ;
METHOD = DEF_METHOD ;
KERNEL = DEF_KERNEL ;
SMO_DISPLAY = DEF_DISPLAY ;
KFOLD = DEF_KFOLD ;
INDEX = 1 ;
settings->cache_size = DEF_CACHE ;
settings->lnC_step = DEF_COARSESTEP ;
settings->lnK_step = DEF_COARSESTEP ;
settings->lnC_start = DEF_INFERLNC ;
settings->lnK_start = DEF_INFERLNK ;
settings->lnC_end = DEF_SUPERLNC ;
settings->lnK_end = DEF_SUPERLNK ;
settings->def_lnC_step = DEF_COARSESTEP ;
settings->def_lnK_step = DEF_COARSESTEP ;
settings->def_lnC_start = DEF_INFERLNC ;
settings->def_lnK_start = DEF_INFERLNK ;
settings->def_lnC_end = DEF_SUPERLNC ;
settings->def_lnK_end = DEF_SUPERLNK ;
settings->zoomin = DEF_ZOOMIN ;
settings->loops = DEF_LOOP ;
settings->ardon = DEF_ARDON ;
settings->repeat = DEF_REPEAT ;
settings->seeds = 0 ;
settings->smo_balance = DEF_BALANCE ;
settings->best_rate = 0 ;
settings->time = 0 ;
settings->normalized_input = DEF_NORMALIZEINPUT ;
settings->normalized_output = DEF_NORMALIZETARGET ;
settings->trainmethod = DEF_TRAINING ;
// save data file name in INPUTFILE
if ( 0!=strlen(filename) && '-'!=filename[0] )
{
if ( NULL != INPUTFILE )
{
free( (void*) INPUTFILE ) ;
INPUTFILE = NULL ;
}
if ( NULL == ( INPUTFILE = strdup(filename) ) )
{
// clear the structure before exit
free (settings) ;
printf("\r\nFATAL ERROR : fail to save the name of input file.\r\n") ;
return NULL ;
}
}
else
{
free (settings) ;
return NULL ;
}
// if there is "train" in the file name of training data, such as "*train*.*",
// test data set should be named as "*test*.*".
// if we fail to find "train" in the training data file,
// we just use the train data as test data.
// create testing file name
pstr = strstr( INPUTFILE, "train" ) ;
if (NULL == pstr)
TESTFILE = strdup(filename) ;
else
{
result = abs( INPUTFILE - pstr ) ;
strncpy (buf, INPUTFILE, result ) ;
buf[result] = '\0' ;
strcat(buf, "test") ;
strcat (buf, pstr+5) ;
TESTFILE = strdup(buf) ;
}
Create_Data_List( &(settings->pairs) ) ;
Create_Data_List( &(settings->testdata) ) ;
Create_Data_List( &(settings->training) ) ;
Create_Data_List( &(settings->validation) ) ;
return settings ;
}
BOOL smo_LoadMatrix ( Data_List * pairs, char * inputfilename, int inputdim, int nFil, int nCol, double ** matrix)
{
int fila=0,columna=0;
FILE * smo_stream ;
FILE * smo_target = NULL ;
char * pstr = NULL ;
char buf[LENGTH] ;
char * temp ;
int dim = -2 ;
unsigned long index = 1 ;
unsigned int result, sz ;
int var = 0, chg = 0 ;
double * point = NULL ;
unsigned int y ;
int i = 0, j = 0 ;
double mean = 0 ;
double ymax = LONG_MIN ;
double ymin = LONG_MAX ;
double * xmean = NULL;
Data_Node * node = NULL ;
int t0=0, tr=0 ;
FILE * fid ;
Data_List label ;
if ( NULL == pairs || NULL == inputfilename )
return FALSE ;
Clear_Data_List( pairs ) ;
Create_Data_List( &label ) ;
dim = nCol-1;
if (NULL != pairs->filename)
free(pairs->filename) ;
pairs->dimen = dim ;
/*/initialize the x_mean and x_devi in Data_List pairs*/
if ( NULL == (pairs->x_mean = (double *)(malloc(dim*sizeof(double))) )
|| NULL == (pairs->x_devi = (double *)(malloc(dim*sizeof(double))) )
|| NULL == (xmean = (double *)(malloc(dim*sizeof(double))) ) )
{
if (NULL != pairs->x_mean)
free(pairs->x_mean) ;
if (NULL != pairs->x_devi)
free(pairs->x_devi) ;
if (NULL != xmean)
free(xmean) ;
if (NULL != smo_target)
fclose( smo_target ) ;
if (NULL != smo_stream)
fclose( smo_stream );
return FALSE ;
}
for ( j = 0; j < dim; j ++ )
pairs->x_mean[j] = 0 ;
for ( j = 0; j < dim; j ++ )
pairs->x_devi[j] = 0 ;
for ( j = 0; j < dim; j ++ )
xmean[j] = 0 ;
/*/ begin to initialize data_list for digital input only*/
//printf("\nLOADING.... \n") ;
pairs->datatype = CLASSIFICATION ;
do
{
#ifdef SMO_DEBUG
printf("%d\n", index) ;
printf("%s\n\n\n", buf) ;
#endif
point = (double *) malloc( (dim+1) * sizeof(double) ) ; /*/ Pairs to free them*/
if ( NULL == point )
{
printf("not enough memory.\n") ;
if (NULL != smo_target)
fclose( smo_target ) ;
if (NULL != smo_stream)
fclose( smo_stream );
if (NULL != pairs->x_mean)
free(pairs->x_mean) ;
if (NULL != pairs->x_devi)
free(pairs->x_devi) ;
if (NULL != xmean)
free(xmean) ;
Clear_Data_List( pairs ) ;
return FALSE ;
}
columna = 0 ;
while( columna < dim )
{
point[columna]= matrix[fila][columna];
columna++ ;
}
y = matrix[fila][columna];
fila++;
point[dim]=0 ;
if ( TRUE == Add_Data_List( pairs, Create_Data_Node(index, point, y) ) )
{
pairs->mean = (mean * (((double)(pairs->count)) - 1) + y )/ ((double)(pairs->count)) ;
pairs->deviation = pairs->deviation + (y-mean)*(y-mean) * ((double)(pairs->count)-1)/((double)(pairs->count));
mean = pairs->mean ;
for ( j=0; j<dim; j++ )
{
pairs->x_mean[j] = (xmean[j] * (((double)(pairs->count)) - 1) + point[j] )/ ((double)(pairs->count)) ;
pairs->x_devi[j] = pairs->x_devi[j] + (point[j]-xmean[j])*(point[j]-xmean[j]) * ((double)(pairs->count)-1)/((double)(pairs->count));
xmean[j] = pairs->x_mean[j] ;
}
if (y>ymax)
{ ymax = y ; pairs->i_ymax = index ;}
if (y<ymin)
{ ymin = y ; pairs->i_ymin = index ;}
/*/ check data type */
Add_Label_Data_List( &label, Create_Data_Node(index, point, y) ) ;
index ++ ;
}
else
{
#ifdef SMO_DEBUG
printf("%d\n", index) ;
printf("duplicate data \n") ;
#endif
}
}
while ( fila < nFil);
if (label.count>=2||inputdim>0)
pairs->datatype = ORDINAL ;
else
printf("Warning : not a ordinal regression.\n") ;
if (pairs->count < MINNUM || (pairs->datatype == UNKNOWN && inputdim == 0 ) )
{
printf("too few input pairs\n") ;
Clear_Data_List( pairs ) ;
if (NULL != pairs->x_mean)
free(pairs->x_mean) ;
if (NULL != pairs->x_devi)
free(pairs->x_devi) ;
if (NULL != xmean)
free(xmean) ;
if (NULL != smo_target)
fclose( smo_target ) ;
if (NULL != smo_stream)
fclose( smo_stream );
return FALSE ;
}
/*/ load index file for feature types strstr*/
pairs->featuretype = (int *) malloc(pairs->dimen*sizeof(int)) ;
if (NULL != pairs->featuretype)
{
/*default 0*/
for (sz=0;sz<pairs->dimen;sz++)
pairs->featuretype[sz] = 0 ;
if (0==inputdim)
pstr = strstr( inputfilename, "train") ;
else
pstr = strstr( inputfilename, "test") ;
if (NULL != pstr)
{
sz = abs( pstr - inputfilename ) ;
pstr = strrchr( inputfilename, '.') ;
strncpy( buf, inputfilename, sz ) ;
buf[sz]='\0' ;
strcat( buf, "feature" ) ;
strcat( buf, pstr ) ;
fid = fopen(buf,"r+t") ;
if (NULL != fid)
{
printf("Loading the specifications of feature type in %s ...",buf) ;
sz = 0 ;
while (!feof(fid) && NULL!=fgets(buf,LENGTH,fid) )
{
i=strlen(buf) ;
if (i>1)
{
if (sz>=pairs->dimen)
{
printf("Warning : feature type file is too long.\n") ;
sz = pairs->dimen-1 ;
}
pairs->featuretype[sz] = atoi(buf) ;
sz += 1 ;
}
else
printf("Warning : blank line in feature type file.\n") ;
}
if (sz!=pairs->dimen)
{
/*/default 0*/
for (sz=0;sz<pairs->dimen;sz++)
pairs->featuretype[sz] = 0 ;
printf(" RESET as default.\n") ;
}
else
printf(" done.\n") ;
fclose(fid) ;
}
}
}
pairs->deviation = sqrt( pairs->deviation / ((double)(pairs->count - 1.0)) ) ;
for ( j=0; j<dim; j++ )
pairs->x_devi[j] = sqrt( pairs->x_devi[j] / ((double)(pairs->count - 1.0)) ) ;
/*/ set target value as +1 or -1, if data type is CLASSIFICATION*/
if ( UNKNOWN != pairs->datatype && 0 == inputdim )
{
pairs->deviation = 1.0 ;
pairs->mean = 0 ;
pairs->normalized_output = FALSE ;
}
for ( j=0; j<dim; j++ )
{
if (pairs->featuretype[j] != 0)
{
pairs->x_devi[j] = 1 ;
pairs->x_mean[j] = 0 ;
}
}
if (inputdim>0) /*/ do not normailize data for TESTING*/
{
pairs->normalized_output = FALSE ;
pairs->normalized_input = FALSE ;
}
/*/ normalize the target if needed */
node = pairs->front ;
while ( node != NULL )
{
if ( TRUE == pairs->normalized_input )
{
for ( j=0; j<dim; j++ )
{
if (pairs->x_devi[j]>0)
node->point[j] = (node->point[j]-pairs->x_mean[j])/(pairs->x_devi[j]) ;
else
node->point[j] = 0 ;
}
}
node = node->next ;
}
//printf("Total %lu samples with %u dimensions for ", pairs->count, pairs->dimen) ;
//if (inputdim > 0)
//printf("TESTING.\r\n") ;
//else
if (inputdim <= 0)
{
if( CLASSIFICATION == pairs->datatype )
printf("CLASSIFICATION.\r\n") ;
else if ( ORDINAL == pairs->datatype )
{
//printf("ORDINAL %lu REGRESSION.\r\n",label.count) ;
pairs->classes = label.count ;
if (NULL != pairs->labels)
free( pairs->labels ) ;
i=0;
pairs->labels = (unsigned int*)malloc(pairs->classes*sizeof(unsigned int)) ;
pairs->labelnum = (unsigned int*)malloc(pairs->classes*sizeof(unsigned int)) ;
if (NULL != pairs->labels&&NULL != pairs->labelnum)
{
node = label.front ;
//printf("ordinal varibles : ") ;
while (NULL!=node)
{
if (node->target<1 || node->target>pairs->classes)
{
printf("Error : targets should be from 1 to %d.\n",(int)pairs->classes) ;
exit(1) ;
}
pairs->labels[node->target-1] = node->target ;
if (node->target-1==0)
t0 = node->target ;
if (node->target==(int)pairs->classes)
tr = node->target ;
pairs->labelnum[node->target-1] = node->fold ;
i += node->fold ;
//printf("%d(%d) ", node->target, node->fold) ;
node = node->next ;
}
//printf("\n") ;
if (i!=(int)pairs->count||t0!=1||tr!=(int)pairs->classes)
{
printf("Error in data list.\n") ;
exit(1) ;
}
}
else
{
printf("fail to malloc for pairs->labels.\n") ;
exit(1) ;
}
}
else
printf("UNKNOWN.\r\n") ;
}
if (1 == pairs->normalized_input)
printf("Inputs are normalized.\r\n") ;
if (1 == pairs->normalized_output && pairs->deviation > 0)
printf("Outputs are normalized.\r\n") ;
if ( inputdim > 0 && pairs->deviation <= 0 )
printf("Tragets are not at hand.\r\n") ;
Clear_Label_Data_List (&label) ;
if (NULL != smo_target)
fclose( smo_target ) ;
if (NULL != smo_stream)
fclose( smo_stream );
if ( NULL != xmean )
free( xmean ) ;
return TRUE ;
}
def_Settings * Create_def_Settings_Matlab ( void )
{
/*/printf("%s\n",filename);*/
char filename[30];
strcpy(filename,"mytask_train.0");
/*printf("%s\n",filename);*/
def_Settings * settings = NULL ;
char * pstr = NULL ;
unsigned int result = 0 ;
char buf[LENGTH] = "" ;
if (NULL == filename)
{
printf("\r\nFATAL ERROR : the input pointer is NULL.\r\n") ;
return NULL ;
}
if (NULL == (settings = (def_Settings *)(malloc(sizeof(def_Settings)))))
{
printf("\r\nFATAL ERROR : fail to malloc def_settings.\r\n") ;
return NULL ;
}
/* set all elements as default values*/
INPUTFILE = NULL ;
TESTFILE = NULL ;
VC = DEF_VC ;
TOL = DEF_TOL ;
KAPPA = DEF_KAPPA ;
EPS = DEF_EPS ;
P = DEF_P ;
METHOD = DEF_METHOD ;
KERNEL = DEF_KERNEL ;
SMO_DISPLAY = DEF_DISPLAY ;
KFOLD = DEF_KFOLD ;
INDEX = 1 ;
settings->cache_size = DEF_CACHE ;
settings->lnC_step = DEF_COARSESTEP ;
settings->lnK_step = DEF_COARSESTEP ;
settings->lnC_start = DEF_INFERLNC ;
settings->lnK_start = DEF_INFERLNK ;
settings->lnC_end = DEF_SUPERLNC ;
settings->lnK_end = DEF_SUPERLNK ;
settings->def_lnC_step = DEF_COARSESTEP ;
settings->def_lnK_step = DEF_COARSESTEP ;
settings->def_lnC_start = DEF_INFERLNC ;
settings->def_lnK_start = DEF_INFERLNK ;
settings->def_lnC_end = DEF_SUPERLNC ;
settings->def_lnK_end = DEF_SUPERLNK ;
settings->zoomin = DEF_ZOOMIN ;
settings->loops = DEF_LOOP ;
settings->ardon = DEF_ARDON ;
settings->repeat = DEF_REPEAT ;
settings->seeds = 0 ;
settings->smo_balance = DEF_BALANCE ;
settings->best_rate = 0 ;
settings->time = 0 ;
settings->normalized_input = DEF_NORMALIZEINPUT ;
settings->normalized_output = DEF_NORMALIZETARGET ;
settings->trainmethod = DEF_TRAINING ;
/* save data file name in INPUTFILE*/
if ( 0!=strlen(filename) && '-'!=filename[0] )
{
if ( NULL != INPUTFILE )
{
free( (void*) INPUTFILE ) ;
INPUTFILE = NULL ;
}
if ( NULL == ( INPUTFILE = strdup(filename) ) )
{
free (settings) ;
printf("\r\nFATAL ERROR : fail to save the name of input file.\r\n") ;
return NULL ;
}
}
else
{
free (settings) ;
return NULL ;
}
pstr = strstr( INPUTFILE, "train" ) ;
if (NULL == pstr)
TESTFILE = strdup(filename) ;
else
{
result = abs( INPUTFILE - pstr ) ;
strncpy (buf, INPUTFILE, result ) ;
buf[result] = '\0' ;
strcat(buf, "test") ;
strcat (buf, pstr+5) ;
TESTFILE = strdup(buf) ;
}
Create_Data_List( &(settings->pairs) ) ;
Create_Data_List( &(settings->testdata) ) ;
Create_Data_List( &(settings->training) ) ;
Create_Data_List( &(settings->validation) ) ;
return settings ;
}