void SOM::train( QMap<QString, QVariant> somParameters, QVector<FeaturePtr> features, bool skipInit_debugOnly )
{
if( skipInit_debugOnly == false )
initializeTraining( somParameters, features );
do
{
// Shuffle the list of features to remove bias
std::random_shuffle( features.begin(), features.end() );
// Update the SOM with each feature from the globalFeatures list.
// @todo: consider random sampling for the training if the number of features is high
foreach( FeaturePtr feature, features )
update( feature );
} while( nextEpoch() != false );
}
int main( int argc, char **argv ) {
char *progname = argv[0];
FILE *in , *out;
char opt;
int i;
double startTime;
C = 0;
kernelType = -1;
degree = 0;
sigmaSqr = 0;
binaryFeature = 0;
/* Check command line options */
optarg = NULL;
while (((opt = getopt(argc, argv, OPTSTRING)) != -1)) {
switch(opt) {
case 't':
if ( strcmp(optarg, "0") == 0 ) kernelType = 0;
else if ( strcmp(optarg, "1") == 0 ) kernelType = 1;
else if ( strcmp(optarg, "2") == 0 ) kernelType = 2;
else {
fprintf( stderr, "kernel type is either 0,1 or 2\n");
exit(1);
}
break;
case 'c':
if( sscanf(optarg, "%f", &C) == 0 ) {
fprintf(stderr,"Expect a positive number for C.\n");
exit(1);
} else
C = atof(optarg);
if( C <= 0 ) {
fprintf( stderr, "C has to be > 0\n");
exit(1);
}
break;
case 'd':
if( sscanf(optarg, "%d", °ree) == 0 ) {
fprintf( stderr, "Expect degree to be a positive integer.\n");
exit(1);
} else degree = atoi(optarg);
if ( degree <= 0 ) {
fprintf( stderr, "degree has to be a positive integer.\n");
exit(1);
}
break;
case 'v':
if( sscanf(optarg, "%f", &sigmaSqr) == 0 ) {
fprintf(stderr,"Expect a positive number for variance.\n");
exit(1);
} else sigmaSqr = atof(optarg);
if( sigmaSqr <= 0 ) {
fprintf( stderr, "variance has to be > 0\n");
exit(1);
}
rbfConstant = 1/(2*sigmaSqr);
break;
case 'b':
if( sscanf(optarg, "%d", &binaryFeature) == 0 ) {
fprintf( stderr, "binaryFeature option is either 0 or 1.\n");
exit(1);
} else binaryFeature = atoi(optarg);
if ( binaryFeature != 0 && binaryFeature != 1 ) {
fprintf( stderr, "binaryFeature option is either 0 or 1.\n");
exit(1);
}
break;
case 'h':
useMsg( progname );
exit(1);
break;
default:
useMsg( progname );
exit(1);
break;
}
}
/* Check all necessary parameters are in */
if( kernelType == -1 ) {
fprintf( stderr, "Kernel type has not been specified.\n");
exit(2);
} else if( kernelType == 1 && degree == 0 ) {
fprintf( stderr, "Degree has not been specified.\n");
exit(2);
} else if( kernelType == 2 && sigmaSqr == 0 ){
fprintf( stderr, "Variance has not been specified.\n");
exit(2);
} else if( C == 0 ) C = DEFAULT;
/* Check training file and model file */
printf("INPUT FILE %s\n", argv[argc-2]);
if (( in = fopen( argv[argc-2], "r") ) == NULL ) {
fprintf( stderr, "Can't open %s\n", argv[argc-2] );
exit(2);
}
if (( out = fopen( argv[argc-1], "w") ) == NULL ) {
fprintf( stderr, "Can't open %s\n", argv[argc-1] );
exit(2);
}
printf("smo_learn is preparing to learn. . .\n");
if( ! readFile( in ) ) {
fprintf( stderr, "Error in initializing. Program exits.\n" );
exit (1);
} else fclose( in );
if( !initializeTraining()) {
fprintf( stderr, "Error in initializing data structure. Program exits.\n");
exit(1);
}
printf("Start training . . .\n");
startTime = clock()/CLOCKS_PER_SEC;
startLearn();
printf("Training is completed\n");
/* Print training statistics */
printf("CPU time is %f secs\n", clock()/CLOCKS_PER_SEC-startTime);
printf("Writing training results . . .\n");
writeModel( out );
fclose( out );
printf("Finish writing training results.\n");
printf("no of iteration is %f\n", iteration);
printf("threshold b is %f\n", getb());
if ( kernelType == 0 )
printf("norm of weight vector is %f\n", calculateNorm());
printf("no. of unBound multipliers is %d\n", unBoundSv );
printf("no. of bounded multipliers is %d\n", boundSv );
/* Free memory */
free( target );
free( lambda );
free( nonZeroFeature );
free( error );
free( nonBound );
free( weight );
free( unBoundIndex );
free( nonZeroLambda );
for( i = 0; i <= numExample; i++ ) {
free( example[i] );
}
free( example );
free( errorCache );
return 0;
}
