The HUB TOOLBOX now supports the following hubness reduction methods:
- Mutual Proximity (Empiric, Gauss, Indep. Gauss, Indep. Gamma)
- Local Scaling (Normal, NICDM)
- Shared Nearest Neighbors
- Centering (Normal, Weighted, Localized)
- DisSim (Global, Local)
Most methods now support (sparse) similarity matrices.
Most methods now support train/test splits.
MP and hubness functions now support parallel processing.
Performance improvements for several methods.
Vast unit test coverage.