So, I was bored and wrote a python script to do some approximate simulations of Stellaris space combat. I then used this to simulate a series of combats in an attempt to determine which corvette configuration is, statistically, the best.
Why did I do this? I dunno. But I'm gonna be changing my play style a little…
The TL;DR is that fully mixed-equipment corvettes are statistically strong. Further, the best of the best have two armor, one shield component.
Right, so the combat simulator I wrote has some limitations, mostly due to this being the first incarnation of it I've made, and due to time concerns. So, keep the following limitations in mind when thinking about these results.
- Only Interceptor corvettes are considered.
- Only Tier 1 weapons and armor (red laser, mass driver, deflector, and nanocomposite armor) are considered.
- All other modules are set to Tier 1. No auxiliary modules are applied.
- All energy-related concerns are ignored.
- Missiles and point-defense weapons are neglected entirely (as implied by the ship class restrictions).
- All bonuses are ignored – only basic statistics are used.
- Range is ignored.1
- Fire rates are handled in a manner slightly inconsistent with the game.2
- All ships ready to fire will do so simultaneously.
- Withdrawing from combat has not been implemented, although I do not thing it will change the results of this test.
- Shield regeneration is ignored.
- All ships target the same enemy and continue firing at that one target until it dies.
So, all that in mind, I made each of the 16 different possible ships available with these restrictions, paired them off3 in fleets of three, and had them fight. I iterated each battle 100 times, to get a reasonable statistical sample.
Here are some results:
|Ship Name||Win %|
|LCCSAA||<1075, 502, 23>||0.671875|
|LLCSAA||<1061, 506, 33>||0.663125|
|LLCSSA||<993, 588, 19>||0.620625|
|LCCSSA||<990, 588, 22>||0.61875|
|LLCSSS||<815, 774, 11>||0.509375|
|LCCSSS||<811, 775, 14>||0.506875|
|LCCAAA||<806, 782, 12>||0.50375|
|LLCAAA||<803, 784, 13>||0.501875|
|CCCSAA||<686, 903, 11>||0.42875|
|LLLSAA||<683, 904, 13>||0.426875|
|CCCAAA||<670, 910, 20>||0.41875|
|LLLSSA||<661, 917, 22>||0.413125|
|CCCSSA||<660, 929, 11>||0.4125|
|LLLSSS||<658, 920, 22>||0.41125|
|CCCSSS||<656, 936, 8>||0.41|
|LLLAAA||<639, 949, 12>||0.399375|
The ship names are descriptive: L stands for laser, C for cannon, S for shield, A for armor. Draws did occasionally occur, and were tracked; these happened when both fleets destroyed each other simultaneously.
As shown in the table, all corvettes are not made equal! Notably, all four "uniform" type corvettes (i.e. LLLAAA, LLLSSS, etc) did markedly poorly. Shockingly, simply adding one mass driver to a laser-armor build jumped win percentage by over 10%! Meanwhile, LCCSAA won a surprising 2/3 of the matches, with LLCSAA a close contender. The biggest "jump" was between LLCSSS (and all other builds which have either homogeneous weapons or homogeneous defenses) and LCCSAA (the first fully-mixed build) – an ~11% winrate jump.
I'm not sure exactly what causes these differences. There are a lot of things being simulated here, from weapon cooldown times, accuracy, a slight damage difference, as well as the order in which damage is dealt.
Now, while I only simulated tier 1 equipment, I think it's safe to say that these results would hold if I used higher-tier equipment of the same type. However, things such as crystal plating and different weapon classes, not to mention range considerations, ship classes, and the like could and very well likely would have substantial effects on these results.
As for the future: it would be pretty easy to implement larger ship classes and other weapon types, and I might do so at some future time. However, allowing for mixed weapon sizes would really exacerbate range considerations. We'll see. Another possible test would be to allow mixing of ship classes, to see if that makes any difference. Further, an upgrade to the targeting algorithm could be as simple as having each ship pick a random target at the start, then giving them a small chance per time unit of retargeting. This could result in more accurate simulations.
1: I did apply a random delay between the start of the simulation and the first attack for each weapon. In a way, this simulates the ships entering combat at different times.
2: The game divides time into days. Some weapons, including all weapons considered here, have non-integer cooldown times (i.e. the Red Laser fires once ever 4.6 days). Further, previous versions of the game included a "wind-up" time in the weapon fire rates. Since I couldn't find any good descriptions of how either of these are handled in game, I instead chose to divide each day into 20 units At each time unit, the simulation checks every weapon for readiness, fires it if it was ready, and then either advances a cooldown timer by one unit or resets the cooldown timer as appropriate.
3: Due to an oversight, each fleet also fought it's clone. Since scores were tracked on a per-fleet basis, each of these matches either added 1 win and 1 loss, or 2 draws, to the fleet in question. The draws would have the effect of reducing the winrate, while the non-draws would draw winrate towards pairity. Since this ends up being 1/16 of the matches played by each fleet, I've chosen to ignore it for now. The code now accounts for it correctly but I chose not to re-run the tests for time reasons.
Source: Original link
© Post "A brute force statistical study of Interceptor-type corvettes" for game Stellaris.
Top-10 Best Video Games of 2018 So Far
2018 has been a stellar year for video game fans, and there's still more to come. The list for the Best Games of So Far!
Top-10 Most Anticipated Video Games of 2019
With 2018 bringing such incredible titles to gaming, it's no wonder everyone's already looking forward to 2019's offerings. All the best new games slated for a 2019 release, fans all over the world want to dive into these anticipated games!