Kernel PCA
Non-linear extension of PCA using kernel trick to implicitly map data to higher dimensions
Kernel PCA
Non-linear extension of PCA using kernel trick to implicitly map data to higher dimensions.
When to use:
- Data has non-linear structure
- Want PCA-like properties with non-linearity
- Have prior knowledge of good kernel
- Need to transform new data
Strengths: Non-linear, flexible (many kernels), can transform new data, PCA-like properties Weaknesses: Sensitive to kernel choice, less interpretable, slower than PCA
Model Parameters
N Components (default: 2, required) Number of components.
Kernel (default: "linear") Kernel function:
- linear: Standard PCA (default)
- rbf: Radial basis function (most common for non-linear)
- poly: Polynomial kernel (flexible)
- sigmoid: Sigmoid kernel (neural-network-like)
- cosine: Cosine similarity
Gamma (optional) Kernel coefficient for rbf, poly, sigmoid.
- null: 1 / n_features (default)
- Low values: Broad, smooth features
- High values: Sharp, local features
Degree (default: 3) Degree for polynomial kernel.
- 2: Quadratic relationships
- 3: Cubic (default)
- 4+: Higher-order interactions
Random State (default: 42) Seed for reproducibility.