Closed SergioQuijanoRey closed 2 years ago
Now that this metric is working, the problem is that is very slow. With wandb login we've seen that GPU utilization is very low when running the computations for this metric, so we need to find a way to utilize GPU power for the computations (more operations with torch objects directly)
This metric is implemented, so I close this issue. In other issue we will address the problem that the computation is slow and doesn't take advantage of GPU power.
Do the same that we did in #33, but with inter-cluster distances.