search

open-connectome-classes / StatConn-Spring-2015-Info

introductory material
18 stars 4 forks source link

Permutations vs. adjusted rand index #90

Open ghost opened 9 years ago

ghost commented 9 years ago

I didn't catch the difference between using permutations vs. using the adjusted rand index. What was the drawback for using permutations?

jtmatterer commented 9 years ago

If you want to calculate the loss in practice (for example if you simulate data for which you know ground truth), you may not be able to find the optimal permutation to align the true labels with the cluster labels since the number of permutations to search over is K!, where K is the total number of clusters. ARI is a simple calculation.

SandyaS72 commented 9 years ago

ARI is just based on the number of pairs of vertices that are in the same or different clusters between the two clusterings you're trying to compare. This way, it ignores what the explicit label assigned to the clusters is.