Spectral Clustering
A graph-theory-based clustering algorithm that computes eigenvectors of the Laplacian matrix of a data similarity graph to project data into a low-dimensional spectral space before applying clustering. The approach is effective at identifying complex, non-convex, and non-spherical data structures.
Timeline (last 90 days)
Spectral clustering can discover non-convex, non-spherical data structures that traditional methods such as k-means tend to miss.
Spectral clustering maps high-dimensional data to a low-dimensional spectral space by computing the eigenvectors of the graph Laplacian matrix, and then completes clustering with k-means.
The space complexity of building an n×n similarity matrix in spectral clustering is O(n²).
The time complexity of performing eigendecomposition on the Laplacian matrix in spectral clustering is typically O(n³), making it the most time-consuming step in the entire process.
In the original demonstration, the author used Scikit-Learn spectral clustering to segment the volatility surfaces of 50,000 options contracts into different risk intervals.
All Facts (5)
Spectral clustering can discover non-convex, non-spherical data structures that traditional methods such as k-means tend to miss.
50%UnverifiedSpectral clustering maps high-dimensional data to a low-dimensional spectral space by computing the eigenvectors of the graph Laplacian matrix, and then completes clustering with k-means.
50%UnverifiedThe space complexity of building an n×n similarity matrix in spectral clustering is O(n²).
50%UnverifiedThe time complexity of performing eigendecomposition on the Laplacian matrix in spectral clustering is typically O(n³), making it the most time-consuming step in the entire process.
50%UnverifiedIn the original demonstration, the author used Scikit-Learn spectral clustering to segment the volatility surfaces of 50,000 options contracts into different risk intervals.
50%