Why can switching label propagation from an RBF kernel to a KNN kernel reduce memory substantially?
Classic MLLevel: IntermediateSemi-supervised graph scale
Why can switching label propagation from an RBF kernel to a KNN kernel reduce memory substantially?
Hints
- Compare all-pairs edges with local edges.
- Sparse connectivity is also a modeling choice.
Answer
RBF constructs a dense similarity graph over observations, whereas KNN retains a sparse neighborhood graph. This changes both computational scale and the neighborhood assumptions, so validate quality.
Source and licence: scikit-learn 1.7.2 — semi supervised · BSD-3-Clause · answer: Quantoros editorial