I'd like to add the reduced mutual information and its normalized version, introduced in this preprint to clusim. See Eq. (23) and (25) of the paper.
It is an information theoretic measure along the line of the standard MI, which keeps a few extra terms that matter a lot in most cases.
We already have code to compute it for the standard output of scikit-clustering, so the addition should be pretty straightforward. The only dependency is lngamma imported from scipy.special, since we are using the exact combinatorial formulas.
Hi :wave:
I'd like to add the reduced mutual information and its normalized version, introduced in this preprint to clusim. See Eq. (23) and (25) of the paper.
It is an information theoretic measure along the line of the standard MI, which keeps a few extra terms that matter a lot in most cases.
We already have code to compute it for the standard output of scikit-clustering, so the addition should be pretty straightforward. The only dependency is
lngamma
imported from scipy.special, since we are using the exact combinatorial formulas.Let me know if that would fit within the package.