Solution for Google Hash Code 2019 Qualification Round. Problem statement can be found here.
Please note: This solution did not actually participate in 2019's qualification round, but was created afterwards. It exceeds the runtime constraints and therefore would not have been suitable as a submission.
1. Naive approach (GreedyNaive.scala) (O(n²) worst case):
- Construct slides from vertical pictures by iteratively combining those with most tags
- For every slide, iteratively find slide with highest interest factor in a greedy-like way
2. Filtered naive approach (GreedyFiltered.scala) (O(n*m) worst case):
- Similar to first approach, but in second step only slides with a similar number of tags as the current slide are taken into account
- Adapt input file path in
Solution.scala sbt run
a_example.txt: 2 points, 0.14 secondsb_lovely_landscapes.txt: 202707 points, 4.68 hoursc_memorable_moments.txt: 1535 points, 2.56 secondsd_pet_pictures.txt: 434189 points, 3.91 hourse_shiny_selfies.txt: 406066 points, 3.69 hours
Total: 1,044,499 points, 12.28 hours ☠️ Est. rank: 83 / 6671 (1.2 %)
a_example.txt: 2 points, 0.13 secondsb_lovely_landscapes.txt: 81306 points, 1.57 hoursc_memorable_moments.txt: 1489 points, 2.65 secondsd_pet_pictures.txt: 406488 points, 1.96 hourse_shiny_selfies.txt: 361380 points, 0.7 hours
Total: 850,665 points, 4.23 hours ☠️ Est. rank: 450 / 6671 (6.7 %)
The results were produced in parallel fashion on a 12-core CPU.
The naive approach is totally unrealistic due to extensive runtime. The filtered naive approach essentially trades off score vs. runtime. By tweaking its parameters, runtime could probably be pushed down to ~ 3 hours while losing total score. All in all, a more sophisticated solution would be appropriate...
MIT