In search of deterministic methods for initializing K-means and Gaussian mixture clustering