Eigenvalues are linearly increasing (Weyl's law). To compensate for that effect (and reduce the influence of large eigenvalues on the distance measures), eigenvalues can be divided by their index: lambda_i / i . Here I check if the first eigenvalue is zero and start dividing at the next one. Area normalization and index division are now switched on by default (not sure what effect this has on downstream processing, for example, WESD should not be combined with index normalization).
Eigenvalues are linearly increasing (Weyl's law). To compensate for that effect (and reduce the influence of large eigenvalues on the distance measures), eigenvalues can be divided by their index: lambda_i / i . Here I check if the first eigenvalue is zero and start dividing at the next one. Area normalization and index division are now switched on by default (not sure what effect this has on downstream processing, for example, WESD should not be combined with index normalization).