int main_vcfsom(int argc, char *argv[])
{
int c;
args_t *args = (args_t*) calloc(1,sizeof(args_t));
args->argc = argc; args->argv = argv;
args->nbin = 20;
args->learn = 1.0;
args->bmu_th = 0.9;
args->nfold = 5;
args->rand_seed = 1;
args->ndim = 2;
args->bad_class = 1;
args->good_class = 2;
args->merge = MERGE_AVG;
args->train_bad = 1;
static struct option loptions[] =
{
{"help",0,0,'h'},
{"prefix",1,0,'p'},
{"ntrain-sites",1,0,'n'},
{"random-seed",1,0,'r'},
{"bmu-threshold",1,0,'b'},
{"exclude-bad",0,0,'e'},
{"learning-rate",1,0,'l'},
{"size",1,0,'s'},
{"som-dimension",1,0,'d'},
{"nfold",1,0,'f'},
{"merge",1,0,'m'},
{"train",0,0,'t'},
{"classify",0,0,'c'},
{0,0,0,0}
};
while ((c = getopt_long(argc, argv, "htcp:n:r:b:l:s:f:d:m:e",loptions,NULL)) >= 0) {
switch (c) {
case 'e': args->train_bad = 0; break;
case 'm':
if ( !strcmp(optarg,"min") ) args->merge = MERGE_MIN;
else if ( !strcmp(optarg,"max") ) args->merge = MERGE_MAX;
else if ( !strcmp(optarg,"avg") ) args->merge = MERGE_AVG;
else error("The -m method not recognised: %s\n", optarg);
break;
case 'p': args->prefix = optarg; break;
case 'n': args->ntrain = atoi(optarg); break;
case 'r': args->rand_seed = atoi(optarg); break;
case 'b': args->bmu_th = atof(optarg); break;
case 'l': args->learn = atof(optarg); break;
case 's': args->nbin = atoi(optarg); break;
case 'f': args->nfold = atoi(optarg); break;
case 'd':
args->ndim = atoi(optarg);
if ( args->ndim<2 ) error("Expected -d >=2, got %d\n", args->ndim);
if ( args->ndim>3 ) fprintf(stderr,"Warning: This will take a long time and is not going to make the results better: -d %d\n", args->ndim);
break;
case 't': args->action = SOM_TRAIN; break;
case 'c': args->action = SOM_CLASSIFY; break;
case 'h':
case '?': usage();
default: error("Unknown argument: %s\n", optarg);
}
}
if ( !args->rand_seed ) args->rand_seed = time(NULL);
if ( argc!=optind+1 ) usage();
args->fname = argv[optind];
init_data(args);
if ( args->action == SOM_TRAIN ) do_train(args);
else if ( args->action == SOM_CLASSIFY ) do_classify(args);
destroy_data(args);
free(args);
return 0;
}
/* Gateway function */
void mexFunction(int nlhs, mxArray_t *plhs[],
int nrhs, const mxArray_t *prhs[])
{
double *y;
mxArray_t *f_dimensions, *f_numcategories, *obj, *classifier;
int dimensions, data_len, categories;
/* Check that we have 2 RHS arguments */
if (nrhs != 2) {
mexErrMsgTxt("classify: There must be exactly two arguments");
return;
}
/* Check that we have zero or one LHS arguments */
if ((nlhs < 0) || (nlhs > 1)) {
mexErrMsgTxt("classify: Too many output arguments");
return;
}
if (nlhs == 0) return;
(const mxArray_t *)obj = prhs[0];
/* Turn it into a BOOST object */
call_matlab_1_1(&obj, "as_boost", prhs[0]);
/* NOTE: This is a hack, but seems necessary as MATLAB won't seem to
recognise that obj has an inherited DIMENSIONS method from the
CLASSIFIER ancestor. */
call_matlab_1_1(&classifier, "as_classifier", obj);
/* Get our number of dimensions */
call_matlab_1_1(&f_dimensions, "dimensions", classifier);
dimensions = (int)mxGetScalar(f_dimensions);
mxDestroyArray(f_dimensions);
/* Ask MATLAB to tell us how many categories are in our dataset */
call_matlab_1_1(&f_numcategories, "numcategories", classifier);
categories = (int)mxGetScalar(f_numcategories);
mxDestroyArray(f_numcategories);
mxDestroyArray(classifier);
/* Get our X variable. Make sure its a matrix, and get its number of
rows and columns. */
if (!mxIsClass(prhs[1], "double")) {
mexErrMsgTxt("classify: x must be a double array");
return;
}
data_len = mxGetM(prhs[1]);
/* Check that the number of dimensions matches */
if (dimensions != mxGetN(prhs[1])) {
mexErrMsgTxt("classify: dimensions of x and obj don't match");
return;
}
/* Create storage for our output */
plhs[0] = mxCreateDoubleMatrix(data_len, 1, mxREAL);
if (plhs[0] == NULL) {
mexErrMsgTxt("classify: out of memory allocating Y!");
return;
}
y = mxGetPr(plhs[0]);
/* Call our computational routine */
do_classify(obj, prhs[1], y, data_len, categories);
}
