Slide 22
Slide 22 text
© DeNA Co., Ltd. 22
1 Santa 2022 上位解法
● 1 位
● 2 位
● 3 位
● 4 位
… We worked on the former part by solving TSP with additional constraints using “genetic algorithm using edge
assembly crossover (GA-EAX)”, …
… The final push was obtained by using all the (mostly unconstrained) tours I had gathered in the first weeks of the
challenge. I used merging with IPT and some limited genetic algorithm to reach the final score of 74075.706541 …
… For finding a good tour satisfying the constraints, I used GA-EAX-restart, modified with a penalty function that would
enforce the constraints. Depending on the population size and luck, this method gives tours with scores in the 74076.x
range. I was quite surprised, this is the first time that I see a genetic algorithm being useful. …
… My solution is to solve the constrained TSP using Concorde and a genetic algorithm, then convert the resulting tour
to configurations. …
https://www.kaggle.com/competitions/santa-2022/discussion/379080
https://www.kaggle.com/competitions/santa-2022/discussion/379214
https://www.kaggle.com/competitions/santa-2022/discussion/379086
https://www.kaggle.com/competitions/santa-2022/discussion/379167