> although for some reason Gunnar's baseline is slower. Maybe I'm doing something wrong. I suspect it's because the baseline code is adding each line to a Tree

> although for some reason Gunnar's baseline is slower. Maybe I'm doing something wrong. I suspect it's because the baseline code is adding each line to a TreeMap before processing, effectively sorting 1B rows and then processing them. Your code reduces to the 413 unique stations and then sorts those. I feel like Gunnar purposefully wrote the baseline with lots of room for improvement to encourage participation.