void kmedians::process(const dataset & data, cluster_data & output_result) {
m_ptr_data = &data;
m_ptr_result = (kmedians_data *) &output_result;
if (data[0].size() != m_initial_medians[0].size()) {
throw std::invalid_argument("kmedians: dimension of the input data and dimension of the initial medians must be equal.");
}
m_ptr_result->medians() = m_initial_medians;
double changes = std::numeric_limits<double>::max();
double prev_changes = 0.0;
std::size_t counter_repeaters = 0;
for (std::size_t iteration = 0; (iteration < m_max_iter) && (changes > m_tolerance) && (counter_repeaters < 10); iteration++)
{
update_clusters(m_ptr_result->medians(), m_ptr_result->clusters());
changes = update_medians(m_ptr_result->clusters(), m_ptr_result->medians());
double change_difference = std::abs(changes - prev_changes);
if (change_difference < THRESHOLD_CHANGE) {
counter_repeaters++;
}
else {
counter_repeaters = 0;
}
prev_changes = changes;
}
m_ptr_data = nullptr;
m_ptr_result = nullptr;
}
void kmedians::process(const std::vector<point> & data) {
m_ptr_data = (std::vector<point> *) &data;
if (data[0].size() != m_medians[0].size()) {
throw std::runtime_error("CCORE [kmedians]: dimension of the input data and dimension of the initial cluster medians must be equal.");
}
m_clusters.clear();
double stop_condition = m_tolerance * m_tolerance;
double changes = 0.0;
double prev_changes = 0.0;
size_t counter_repeaters = 0;
do {
update_clusters();
changes = update_medians();
double change_difference = abs(changes - prev_changes);
if (change_difference < 0.000001) {
counter_repeaters++;
}
else {
counter_repeaters = 0;
}
prev_changes = changes;
}
while ((changes > stop_condition) && (counter_repeaters < 10));
m_ptr_data = nullptr;
}