//main method to train the decision treee
void Dectree_class::train(const cv::Mat& training_data, const cv::Mat& labels, int depth_thresh, unsigned int samples_thresh, int vars_per_node)
{
//---(1)initialize random generator if no external is given---//
if(!is_ext_rng)
{
rng = cv::RNG(time(NULL));
//std::cout << "No ext rng" << std::endl;
}
//---(2)determine the number of classes based on the training data---//
set_classes(labels);
//std::cout << "Classes:\n" << classes << std::endl;
//---(3)make a vector giving an id to each attribute---//
set_attributes(training_data);
//for debbugging
/*
for(std::vector<int>::iterator it = attributes.begin(); it != attributes.end(); ++it)
std::cout << *it << " ";
std::cout << std::endl;
*/
//---(4)verify constraints---//
//maximum depth
if(depth_thresh < 0)
depth_limit = std::numeric_limits<int>::max();
else
depth_limit = depth_thresh;
//minimum samples
min_samples = samples_thresh;
//active variables
if(attributes.size() < (unsigned)vars_per_node)
active_vars = attributes.size();
else
active_vars = vars_per_node;
//maximum number of split fails
max_split_fail = attributes.size();
//---(5)train the tree---//
int depth = 1;
int no_split_fail = 0;
dbst.set_root(learn_dectree(cv::Mat(),labels, training_data, depth, no_split_fail));
find_depth(dbst.get_root()); //to find the real-true depth of the created tree
}
inline char *
finishconf(void)
{
static char buf[256];
char *ret;
if (!new_MeLine || !new_MeLine->servername)
return "Missing global block";
if ((ret = set_classes()))
{
ircsnprintf(buf, sizeof(buf), "Missing class block for referenced "
"class '%s'", ret);
return buf;
}
if (!new_ports)
return "No ports defined";
return NULL;
}
