//prediction for external data
Eigen::VectorXd Tree::predict(DataFrame &testSet, const bool verbose) {
//starting at root level, then call makePredictions recursively
double pi = 0.0;
Eigen::VectorXd p(testSet.nrrows);
for (int obs = 0; obs < testSet.nrrows; obs++) {
//in some rare case we do have a terminal node at the 0th level
if (root->isTerminal) {
pi = root->cm;
} else if (testSet.matrix(obs, root->feature) < root->splitvalue) {
//makePrediction leftDF
pi = makePrediction(testSet, root->left, obs, verbose);
} else {
//makePrediction rightDF
pi = makePrediction(testSet, root->right, obs, verbose);
}
if (!probability && !regression) {
//cout<<"Rounding "<<pi<<" to "<<LUtils::round(pi)<<endl;
pi = LUtils::round(pi);
}
p(obs) = pi;
if (verbose) {
cout << "p(" << obs << "): " << p(obs) << " order:"
<< testSet.order[obs] <<endl;
}
}
//original order for p
Eigen::VectorXd tmp(testSet.nrrows);
int idx_orig = 0;
for (int i = 0; i < testSet.nrrows; i++) {
idx_orig = testSet.order.at(i);
tmp(idx_orig) = p(i);
}
for (int i = 0; i < testSet.nrrows; i++) {
p(i) = tmp(i);
}
//after prediction we have to re-establish the original order
testSet.restoreOrder();
if (verbose)
cout << "###Tree size:" << tree_size + 1 << " nodes." << endl;
return p;
}
//make prediction for a certain row, recursive until terminal node is reached
double Tree::makePrediction(const DataFrame &testset,
const boost::shared_ptr<Node> localNode, const int obs,
const bool verbose) {
double t = 0.0;
if (localNode->isTerminal) {
if (verbose) {
cout << " cm(Node" << localNode->nodeID << "):";
}
return localNode->cm;
}
if (testset.matrix(obs, localNode->feature) < localNode->splitvalue) {
//makePrediction leftDF
t = makePrediction(testset, localNode->left, obs, verbose);
} else {
//makePrediction rightDF
t = makePrediction(testset, localNode->right, obs, verbose);
}
//if we reach this end we go upwards again...
return t;
}
