Machine Learning Algorithms

Spectral clustering

Spectral clustering is a more sophisticated approach based on a symmetric affinity matrix:

Here, each element a_ij represents a measure of affinity between two samples. The most diffused measures (also supported by scikit-learn) are radial basis function and nearest neighbors. However, any kernel can be used if it produces measures that have the same features of a distance (non-negative, symmetric, and increasing).  

The Laplacian matrix is computed and a standard clustering algorithm is applied to a subset of eigenvectors (this element is strictly related to each single strategy). scikit-learn implements the Shi-Malik algorithm ...