无监督聚类ML10_1Unsupervised_Clustering.ppt

Phase I When “μ” is held fixed, Eqn for J is linear in r So it can be solved in closed form The terms for various values of n are independent, so can be optimized separately for each n by choosing rnk = 1 for whichever value of k the norm becomes a minimum. That is, we assign the nth data point to the nearest cluster center. That is * Phase 2 When r is held fixed, the expression for J is a quadratic in μ. So it can be minimized by setting its derivative with respect to μ to zero Or The denominator is the number of points assigned to cluster k. That is, μk is the mean of the k points assigned to cluster k * Convergence Why should the K-means algorithm ever reach a fixed point? A state in which clusters don’t change. As K-means is a special case of a general procedure known as the Expectation Maximization (EM) algorithm. EM is known to converge. Number of iterations could be large. But in practice usually isn’t * Home Work (Due Nov 4th) Problem. Consider the K-means algorithm discussed in slides 17-19. Show that, as a consequence of there being a finite number of assignments for the set of discrete indicator variables rnk, and that for each such assignment there is a unique assignment of {μk}, the K-means algorithm must converge after a finite number of iterations. * Time Complexity Computing distance between two docs is O(m) where m is the dimensionality of the vectors. Reassigning clusters: O(Kn) distance computations, or O(Knm). Computing centroids: Each doc gets added once to some centroid: O(nm). Assume these two steps are each done once for I iterations: O(IKnm). * Seed Choice Results can vary based on random seed selection. Some seeds can result in poor convergence rate, or convergence to sub-optimal clusterings. Select good seeds using a heuristic (e.g., doc least similar to any existing mean) Try out multiple starting points Initialize with the results of another method. In the above, if you start with B and E as centroids you converge to {A