#include "dfpar.hpp"
using namespace dfpar;

// define the number of cores we want to use for training and classification.
const size_t NUM_CORES = 4;

void demo() {
    // Declare a dataset of float type features and string type labels.
    // You can use most numeric (including discrete like integers) for
    // features and even more data types for labels, as long as they
    // can be reasonably compared. For custom datatypes, you need to
    // to hack the library.
    DataSet<double, std::string> data;

    // load the dataset from file
    {
        std::ifstream fin("data/iris.data");
        fin >> data;
        fin.close();
    }

    {
        // Initiate the decision forest builder with 500 trees
        // and two random features to try at each split.
        DForestBuilder<double, std::string> builder(data);
        builder.setNumTrees(500);
        // I would suggest number of random features from 2 up to sqrt(num_features),
        // test and see which one works best on your data.
        builder.setNumRandomFeatures(2);

        // build the forest
        DForest<double, std::string> dforest = builder.build(NUM_CORES);

        // report the out-of-bag (OOB) error
        std::cout << "Oob error: " << dforest.getOobError() << std::endl;

        // store the tree on the disk
        std::ofstream fout("model");
        fout << dforest;
        fout.close();
    }

    // load the model back from the disk
    DForest<double, std::string> dforest;
    std::ifstream fin("model");
    fin >> dforest;
    fin.close();

    // classify the same dataset with the built model
    // (this is something you do not do, when doing real evaluation)
    std::vector<ResultRow<std::string>> results = dforest.classifyAll(data, NUM_CORES);
    size_t numCorrect = 0;
    for (auto iter=results.begin() ; iter != results.end() ; ++iter) {
        numCorrect += (*iter).getResponse() == data.getY(iter-results.begin());
    }
    // print the number of correctly classified instances
    std::cerr << "Classified " << numCorrect << " correct out from ";
    std::cerr << data.numRows() << std::endl;
}

int main(int argc, char **argv) {
    try {   
        demo();
    } catch (std::exception &e) {
        std::cerr << e.what() << std::endl;
    } catch (...) {
        std::cerr << "Unkown exception!" << std::endl;
    }
    return EXIT_SUCCESS;
}