marjoupir

I tried level 6 Fighter simulations. Results may be not relevant. Some of the duels are simply endless, for Fighter regeneration is sometimes higher than damage inflicted !! For example, considering a duel between 2 Aumaua Fighters with attributes M=21, C=5, D=16, P=15, I=5, R=16 and Hatchet/Large Shield/Plate, regeneration rate is 0.0266 HP/frame (I considered 1 regen each 5s, is that right ??) and mean damage rate is 0.0145 HP/frame (not considering interrupts!). Even Estoc provides 0.026 Hp/frame mean damage rate. In my code, for computational cost, a duel lasting more than 5 minutes (ingame-time) is a draw, which means 1 victory for each duellist. Then, in the last simulation (lvl 6, 20 generations), a tremendous amount of finalists are 1-hand weapon / Large or Medium Shield / Plate, and win almost all their duels with lazy draws. There's almost no convergence for race and quite a big variability for attributes except for Might (very high though, for regen), for these stuffs are not critical to be able to manage draws. Whatever, results are biased and another way is needed to rank builds and manage endless fights. Any intuition, suggestion or remark is highly welcome !!

'Tis obvious, after reading comments, that my stuff is a deeply biased projection of reality! That's not about lacking rigor, that's more about available free time and programming skills : I hardly see a way to code a generic fight simulator, taking into account all the possible skill combinations, and I'd definitely play the game and have social life, rather than coding such painful things . I'd be glad to share the code though. According to the numerous constructive advices (thx for them), I'd rather fix the class (Fighter, the most passive builds) and probably race (except if I can implement interesting native abilities). Higher level simulations are runing (lvl6 @Petrivanzyl, then 12), but still without active skill. As I see no way to deal with simulating all active skill or spells in a generic way, maybe it could be relevant to pick a single, complete, lvl12 melee class build around there (let us call him Brian), fork the fighting simulator to adapt it to Brian's skills and abilities, and use it as a reference character to test and rank the population. If it provides a decent 'nemesis' Fighter build, optimizing the probability of victory just with autoattacks, it may succeed killing Brian with the few active skills it may have ? Python.

Hi there! Here are some of the best close combat, raw builds to optimize character in 1v1 fights. ---Difficulty settings used: not relevant here. ---Race+Class+main role: any. Close combat, duel oriented. ---Attribute distribution used upon character creation: any. ---List of the abilities/talents chosen: none tested. Only raw characters. ---the name of special item: only usual weapons/armors tested. I used a genetic algorithm on a fight engine simulation code: 30 'tournaments' are simulated between an initial population of 1000 randomly build character (random race, culture, class, attributes, weapons/armors). During a tournament, the characters fight each other in 1v1 duels, are ranked, the weakest replaced by new 'children' created by mixing the stats of the bests (of course whatever their sex !! even eugenic, consanguineous love is f*ckin'love! ). After 30 such generations, the population has somehow converged and have similar, optimized attributes. Pros: - the duel simulation code includes action/recovery frame, interrupt, fighter regeneration, and other fight mechanics found in the awesome discussions around there. Many fights are simulated to estimate the probability of victory, and not a single, maybe lucky fight. - not 'tanky', 'dps' oriented builds, but concretely efficient. - possibly a good approximation, despite the following cons... Cons: - only lvl1 character are tested. The relative health per lvl bonus of Barbarian for example, may be underestimated. - sub-races not modelled. Actually, races are taken into account only for attributes bonuses. No special skill tested (Moon Godlike regen, Dwarf poison res). Drawback few races with combat oriented native skills (eg Godlikes, Humans, Hearth Orlans...). - no active skill is modelled. No Rogue sneak attack, no Cipher soul whip, no scroll nor spell etc. Fighter regen and Paladin Faith (deflection bonus) though. Drawback few classes (any magic). - only close combat weapons tested. Drawback few attributes (eg Intellect) and defense types (Will...). - quite selfish stuff: only 1v1, when the fun in game is fighting collective. - the genetic method usually does not provide THE best solution, but still (very) good ones. - probably some fight mechanics are buged/wrong in my code. I'm no Obsidian dev. And maybe theoretical, genetic algo errors: I'm a nerdy smug, not a PhD. - ... Results Amongst the 250 best survivors, resulting from 30 generations: Class : Only Fighters remain ! The convergence is quite fast there. Due to regen ? Barbarians may have been removed for too low deflection, but no Moonk nor Paladin who heroicaly resist during several generations. Armors : Plate armor is your damn best friend in duels, whatever your race, class etc. The convergence is very fast for armor (100% Plate after few generations). Weapons : All the survivors have dual weapons. Most have Sabre&Hatchet (damage & deflection), few have Club & Hatchet (accuracy + crush & deflection). I have some doubts about it though: why Sabre (slash) against Plate (slash DR bonus) ? Is the damage bonus really worth it, compared to mace (crush + DR bypass) for example ? Attributes : Might : mean ~ 18 (std = 0.2). Clearly a must have (damage & regen). Constitution : mean ~ 5.3 (std = 0.. More variability (some builds have not such a pitiful value), but remains quite useless ('cause Plate is your friend). Dexterity : mean ~ 15.9 (std = 0.6). You have a Plate, and the mate in front of you has f*ckin'one too ! You shall hit fast (double weapon + Dex). Perception : mean ~ 15.1 (std = 0.4). You NEED to deflect. Intellect : mean ~ 4.7 (std = 0.5). Useless, according to simulations. Probably biaised by the method (no active skill, no aura...), keep cautious about this one. Resolve : mean ~ 16 (std = 0.4). You still need to deflect. Race : Aumaua (77.7%), Human (14%) and Dwarf(8.3%). See once more the major influence of Might (damage & regen), Resolve (deflection) and Dexterity (that relatively drawbacks Dwarves ?). Culture : Aedyr: 8.3% (Resolve) Deadfire Archipelago: 3.4% (Dexterity) Ixamitl Plains: 6.8% (Resolve) Old Vailia: 0% (Intellect) Rauatai: 1.5% (Constitution) The Living Lands: 41.3% (Might) The White that Wends: 38.6% (Perception) Depending on the race, there are few variations, but the trends are all the same : Might and deflection are prefered. And the winner is... Aumaua, The Living Lands Fighter (level 1) Might: 21 Constitution: 5 Dexterity: 16 Perception: 15 Intellect: 5 Resolve: 16 Sabre|Hatchet|Plate Armor Hope it helps!