A summary of the principle of spectral clustering (spectral clustering)

Original address

Spectral clustering (spectral clustering) is a widely used clustering algorithm, compared to the traditional K-means algorithm, spectral clustering of data distribution is more adaptable, clustering effect is also very good, at the same time, the calculation of clustering is much smaller, more commendable is to achieve it is not complex. In dealing with the actual clustering problem, the individual thinks that the spectral clustering is one of several algorithms that should be considered first. Here we summarize the algorithm principle of spectral clustering. 1. Overview of spectral Clustering

Spectral clustering is an evolutionary algorithm from graph theory, which has been widely used in clustering. The main idea is to think of all the data as points in space, which can be connected by edges. The distance between the two points of the edge weight value is lower, and the distance between the two points of the higher edge weight value, through the graph of all data points to be cut, so that the transduction after the different sub-graph edge weights and as low as possible, while the sub-graph edge weights and as high as possible, so as to achieve the purpose of clustering.

At first glance, the principle of this algorithm is indeed simple, but to fully understand the algorithm, it is necessary to the graph theory of the non-direction of the graph, linear algebra and matrix analysis have a certain understanding. Let's start with the basics of what we need, and learn the spectrum clustering step by step. 2. One of the spectral Clustering basics: The graph of non-weighted weights

Since spectral clustering is based on graph theory, we first review the concept of the graph below. For a graph G, we generally describe the point set V and the set E of the edge. That is G (v,e). where V is all the points in our data set