nmi
Normalized Mutual information is based on the paper:
- Vinh.N.X Information Theoretic Measures for clustering comparison. Vinh 2010
This crate defines various quality measures of clustering based on information theory.
They are normalized with value in the interval [0,1] and some are metrics.
It is also possible to compare different clustering and to quantify how the affectation to items of 2 algorithms are related. It relies on a contingency table.
The various measures are described in the contingency file.
We preferentially use get_nmi_sqrt()
The trait affectation is used to make the contingency table aware of of data versus label association.
License
Licensed under either of
- Apache License, Version 2.0, LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0
- MIT license LICENSE-MIT or http://opensource.org/licenses/MIT
at your option.