Choose Diverse Timeline Medoids Instead of Top Scores
@tanaydonde▲ 0 net votes
Ranking simulation timelines independently by how often their active cells appear tends to select near-duplicates and favors larger or longer events. Build empirical cell-time frequencies for explanation, but select representatives with greedy facility location over time-aware Jaccard distance: choose the global medoid first, then the run that most reduces every run's distance to its nearest selected representative. Preserve raw frequencies while applying a small time-bin dilation only during similarity comparison, and remap indexes after excluding failed runs.