Data Science 11 - Clustering algorithms

Data Science 11 - Clustering algorithms k-Means and variants; Initialization: • Randomly chooses k points from X used as the initial means • k-Means++: Pick initial means, such that they are uniformly distributed in the space. This leads to faster convergence k-Means and variants; Representatives: • k-Medoids or Partitioning Around Medoids (PAM): The cluster representatives are medoids (objects from X). Only the distance between objects is needed Problems with k-Means: • Clustering model with Gaussian distribution does not always fit CURE algorithm • Assumes a Euclidean distance • Allows clusters to have any shape • Uses a collection of representative points to represent clusters CURE algorithm; Pass 1: Pick a random sample of points that fit in main memory • Initial clusters: - Cluster these points hierarchically to create initial clusters • Pick representative points: - For each cluster, p