Algorithm

Initialize $k$ prototypes $$w_j=x_{p}, \ j\in\{1,\ldots,k\}, \ p\in\{1,\ldots,P\}$$where we pick random points in the dataset (i.e. $k\le P$) to initialize each weight. Each cluster $C_{j}$ is associated with prototype $w_{j}$ while for each $x_p$ in input set: put $x_p$ in cluster with nearest prototype $w_j$ for $j$ in range($k$): $w_j=\frac{1}{|c_{J}|}\sum\limits_{x_{l}\in c_{j}}x_{l}$ where $c_{j}$ is the cluster size $E=\sum\limits^{k}_{j=1}\sum\limits_{x_{l}\in c_{j}}|x_{l}-w_{j}|^2$ if($E$ not increasing): break