Isomap
Geodesic-distance manifold embedding for nonlinear dimensionality reduction
Isomap learns a low-dimensional embedding that preserves the geodesic (shortest path along the manifold) distances between points, rather than straight-line Euclidean distances. It is effective when data lies on a curved manifold.
When to use:
- Curved or Swiss-roll-like manifolds in the data
- Visualization of complex structure that PCA misses
- When geodesic distance is a more meaningful measure of similarity than Euclidean distance
Input: Tabular data with the feature columns defined during training Output: Manifold-embedded coordinates for each row
Model Settings (set during training, used at inference)
N Components (default: 2) Dimensionality of the embedding.
N Neighbors (default: 5) Neighborhood size for graph construction. Larger values capture longer-range structure.
Metric (default: minkowski) Distance metric for the neighborhood graph.
Inference Settings
No dedicated inference-time settings. New points are projected using the trained graph and embedding.