0

I'm using scikit-learn to compute the Adjusted Rand Index for a binary classifier where the two labels are $\{0, 255\}$ (binary images). According to the ARI's documentation found here, the ARI should take values in $[-1, 1]$. However, I'm getting very weird results as can be seen in the attached screenshot.

How can we interpret such results? Is there some normalization that needs to be applied here?

Many thanks in advance.

enter image description here

Jakinduria
  • 121
  • 5

1 Answers1

0

The upper bound is 1, but the lower bound is negative and dependent on the number of nodes. Any negative value suggests that the two clusterings that you’re comparing have less in common than they would by random chance.

You’re getting impossible results like 176.71, so this suggests a bug in how you’re calling the function.

Arya McCarthy
  • 6,390
  • 1
  • 16
  • 47