I discovered the mathematically proven best fleet composition

  • The result seems interesting and finally i knew there would be some guy doing code stuffs.


    Just a small reminder, while look into your code, the shield in attack should be calculated by int%(i believe?) for example when gs shot ds ds would only lose 1000 shield(rather than 1100 as I think). It doesn't that matter but still need to point that out. ( also I'm not sure but might be the edge case the explode happens only when <70% rather than <= 70%? It might matter when LL shoting heavy ships and one more hit on 70% border might matter the ratios a lot).


    Anyway, why gradient Ascent lul( I personally dont think the problem is near-convex but anyway it might be a good way to approx) (but since you have already use the GA on attackers, why not on defenders as well 0 .0)


    Anyway GJ!


  • So much time spent on this, while the answer is so simple. The best fleet composition is the one that can optimally destroy your NEXT target. You adjust your own fleet depending on the target's composition.

    .

  • I don't think the problem is very convex either, but with multistart, it seems to work okay. As you can see in the result, the ascent isn't very big in many cases. As for not doing it on the defenders, that's simply because I wanted to stabilize the output. In particular, if the defenders are co-ascending, the two populations can get stuck in a rotation where they could potentially spiral into a suboptimal pit. (Also -- I was lazy to think about how to actually co-evolve them, as there are many different strategies).

    So much time spent on this, while the answer is so simple. The best fleet composition is the one that can optimally destroy your NEXT target. You adjust your own fleet depending on the target's composition.

    I don't know about you, but I can't adjust my fleet composition on a whim -- doing so in any significant way would require growing the total fleet by at least 10% to be significant, which would be an enormous amount of resources, hardly what you could do on a daily basis. The only thing you can do is select a subset of your fleet, which DonutZ and I discussed on the last page. Plus, hunting a single target is not always profitable, especially when they're on to you. Hence, playing with limited knowledge requires making a choice on how to grow your fleet in the very long run, for which this data can be valuable.