Hi!
When you created the nuPlan Graph Datasets, you applied your own segmentation (consecutive fixed-length scenarios of 2 seconds (?) each), correct? What was the rationale behind this decision versus simply keeping the original scenarios from nuPlan and encoding them into graphs?
Also, tags are lost with the graph dataset, right? Or is there a way to restore the assignment of a set of tags (such as high_lateral_acceleration) to individual frames in your dataset?