Some bad backtracking routes still sneak through the filtering

blahblahblah- / subwaydle

MIT License

34 stars 12 forks source link

Some bad backtracking routes still sneak through the filtering #9

Open KeetonMartin opened 2 months ago

KeetonMartin commented 2 months ago

I know there's some kind of check to prevent significant backtracking. However, it seems this generated route was selected for today because its backtracking portion is small relative to the distance all the way to Jamaica. I suggest that we should change the backtrack filtering logic somehow to prevent these situations even for super long routes.

KeetonMartin commented 2 months ago

Another rough one today, also probably slipped through due to a long segment at one end (in this case the D)

KeetonMartin commented 2 months ago

I'm not sure if this is the same issue with filtering or something else, but this one just feels wrong

blahblahblah- commented 2 months ago

This won't fix everything, but after the D, 1, N solution, I realized that the Times Sq/Bryant Pk transfer was not included in the data when the solutions were previously generated, so I went ahead and did that in this commit: https://github.com/blahblahblah-/subwaydle/commit/436c5bca18ba257c6ff07b772d426aec4db67a59

KeetonMartin commented 1 month ago

Woof

blahblahblah- commented 1 month ago

Added crude calculation of progress towards destination per segment, and deprioritize solutions where any segment results in progress below -25%. Hope it helps with the backtracking. https://github.com/blahblahblah-/subwaydle/commit/1f27b0dced93acc2885ba73261a633e2df6caec8

KeetonMartin commented 1 month ago

Thank you! Seems like it should be a great fix. I'll post again if I see more crazy routes :)

KeetonMartin commented 3 weeks ago

Ok, unfortunately I've seen a few weird ones in the last few days. N is still too small to say for sure if things got better or worse, but I think we still have some issues on our hands.