vacuity¶
- probly.quantification.vacuity = <flextype.singledispatch.Flexdispatch object>¶
Compute the vacuity of a second-order distribution.
For a Dirichlet distribution Dir(alpha) with K classes and alpha_0 = sum_c alpha_c, the vacuity is defined as K / alpha_0 and lies in (0, 1]. It corresponds to the “I do not know” mass in Dempster-Shafer / subjective-logic theory introduced by [SKK18], and is equivalent (up to the constant K) to the inverse of the predicted evidence in Natural Posterior Networks [CBZugner+22].