Spectral Embedding
Non-linear dimensionality reduction using eigenvectors of graph Laplacian to preserve local structure
Spectral Embedding
Non-linear dimensionality reduction using eigenvectors of graph Laplacian to preserve local structure.
When to use:
- Data has graph structure
- Want to preserve local neighborhoods
- Have non-convex structure
- Need to reveal clustering structure
Strengths: Handles non-convex structures, reveals clusters, graph-based, preserves local structure Weaknesses: No inference on new data, sensitive to parameters, slow on large datasets
Model Parameters
N Components (default: 2, required) Embedding dimensions.
Affinity (default: "nearest_neighbors") How to construct similarity graph:
- nearest_neighbors: K-nearest neighbors (default)
- rbf: Radial basis function (Gaussian)
- precomputed: Use your own affinity matrix
N Neighbors (default: 5) Number of neighbors for nearest_neighbors affinity.
- Small (3-5): Local structure
- Large (10-20): More global structure
Random State (default: 42) Seed for reproducibility.