Had previously looked at the KL Divergence for detecting bottlenecks in HREX and it seemed to grow unbounded with the number of windows, which is a problem when wanting to compare edges with different number of windows. Normalizing it appears to do the right thing
Had previously looked at the KL Divergence for detecting bottlenecks in HREX and it seemed to grow unbounded with the number of windows, which is a problem when wanting to compare edges with different number of windows. Normalizing it appears to do the right thing
Unnormalized
Normalized