Ordering columns are currently named after the metric (ordering_betweenness_centrality,
ordering_closeness_centrality, ordering_random), so we rename on import by matching any
ordering_* key. A new metric therefore changes our schema without us noticing.
Ask: always ordering. The metric is already in the filename.
Also: The column is 0-based but not contiguous, and 402 of 420 ordered datasets are sparse. Amsterdam
betweenness/grid goes 0, 2, 3, …: 428 features spanning 0…446.
london-random-grid_with-bikenw has 9 872 features with 1 482 missing indices.
Until we caught this, our scrubber compared its frame index directly against ordering, so the tail
of nearly every build-up never rendered. In the worst cases 24 % of segments stayed invisible.
Ask: emit dense 0…n-1
Ordering columns are currently named after the metric (
ordering_betweenness_centrality,ordering_closeness_centrality,ordering_random), so we rename on import by matching anyordering_*key. A new metric therefore changes our schema without us noticing.Ask: always
ordering. The metric is already in the filename.Also: The column is 0-based but not contiguous, and 402 of 420 ordered datasets are sparse. Amsterdam
betweenness/grid goes
0, 2, 3, …: 428 features spanning 0…446.london-random-grid_with-bikenwhas 9 872 features with 1 482 missing indices.Until we caught this, our scrubber compared its frame index directly against
ordering, so the tailof nearly every build-up never rendered. In the worst cases 24 % of segments stayed invisible.
Ask: emit dense
0…n-1