void oskmeans::clusterize()
{
// initialize.
initialize();
if (_snippets.empty())
return;
// clustering.
while (!stopping_criterion())
{
#ifdef DEBUG
std::cerr << "[Debug]:clusterize, iteration #" << _iterations << std::endl;
#endif
for (short c=0; c<_K; c++)
{
// clear the cluster.
_clusters[c].clear();
}
// clear the garbage cluster.
_garbage_cluster.clear();
// iterates points and associate each of them with a cluster.
hash_map<uint32_t,hash_map<uint32_t,float,id_hash_uint>*,id_hash_uint>::const_iterator hit
= _points.begin();
while (hit!=_points.end())
{
float learning_rate = oskmeans::_nu0*pow((oskmeans::_nuf/oskmeans::_nu0),_t/static_cast<float>(_points.size()*oskmeans::_niterations));
#ifdef DEBUG
std::cerr << "learning rate: " << learning_rate << std::endl;
#endif
// find closest cluster to this point.
short cl = assign_cluster((*hit).first,(*hit).second);
// recomputation of centroids/medoids.
if (cl != -1)
{
float cl_norm = 0.0;
recompute_centroid(learning_rate,&_clusters[cl]._c,(*hit).second,cl_norm);
normalize_centroid(&_clusters[cl]._c,cl_norm);
}
++hit;
_t++;
}
// count iteration.
_iterations++;
}
}
void kmeans(int n, int k){
int i, count, cluster;
double percent_change;
int* cluster_membership;
double **new_clusters;
new_clusters = (double **)malloc(k*sizeof(double *));
cluster_membership = (int *)malloc(n*sizeof(int));
count = 0;
int threshold = 10;
percent_change = 100;
//choose first set of cluster centers
for (i = 0; i < k; i++){
clusters[i][0] = data[i][0];
clusters[i][1] = data[i][1];
}
//initialize new clusters variable
//it has three coords: [x][y][number of members]
for (i = 0; i < k; i++){
new_clusters[i] = (double *)malloc(3*sizeof(double));
new_clusters[i][0] = 0;
new_clusters[i][1] = 0;
new_clusters[i][2] = 0;
}
for (i = 0; i < n; i++){
cluster_membership[i] = -1;
}
while((percent_change > threshold) && (count < 100)){
percent_change = 0;
for(i = 0; i < n; i++){
//find the cluster it belongs to
cluster = assign_cluster(data[i][0], data[i][1], k);
//check to see if this data point changed clusters
if (cluster != cluster_membership[i]) {
percent_change++;
cluster_membership[i] = cluster;
}
//track the new cluster sum
new_clusters[cluster][0] += data[i][0];
new_clusters[cluster][1] += data[i][1];
new_clusters[cluster][2]++;
}
//calculate the new cluster centers and update
for (i = 0; i < k; i++){
if (new_clusters[i][2] > 0){
new_clusters[i][0] /= new_clusters[i][2];
new_clusters[i][1] /= new_clusters[i][2];
}
clusters[i][0] = new_clusters[i][0];
clusters[i][1] = new_clusters[i][1];
new_clusters[i][0] = new_clusters[i][1] = new_clusters[i][2] = 0;
}
percent_change /= n;
count++;
}
for (i = 0; i < n; i++){
printf("%d %d %d\n", data[i][0], data[i][1], cluster_membership[i]);
}
}
