Truncated SVD
Singular Value Decomposition truncated to keep only top components, works with sparse matrices
Truncated SVD
Singular Value Decomposition truncated to keep only top components. Works with sparse matrices.
When to use:
- Have sparse matrices (e.g., TF-IDF text data)
- Need PCA-like method without centering
- Latent Semantic Analysis (LSA)
- Large-scale recommendation systems
Strengths: Works with sparse data, fast, no centering needed, good for text Weaknesses: Less interpretable than PCA, sensitive to scaling
Model Parameters
N Components (default: 2, required) Number of components to keep. Must be less than min(n_samples, n_features).
Algorithm (default: "randomized") SVD solver:
- randomized: Fast approximation (default, recommended)
- arpack: Exact but slower, for small n_components
N Iterations (default: 5) Number of power iterations for randomized solver.
- 5: Fast, usually sufficient
- 7-10: Better accuracy for difficult matrices
Random State (default: 42) Seed for reproducibility.