[CLASS BUILD] Ultimate Duellist ?

[marjoupir]

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!

[MadDemiurg]

Fighter is pretty obvious because of no actives. This doesn't reflect the real situation though, for instance monk can CC the fighter for extended period of time which would likely tip the scale.

[Climhazzard]

The auto attack class will definitely beat the non auto attack classes in an auto attack duel, lol.

[SilchasRuin]

Neat idea, but seems like the spherical cow in a vacuum of class testing. This test as is ignores far too many variables to get at anything other than whats the best result when 2 classes auto attack each other to death.

Edited April 21, 2015 by SilchasRuin

[knownastherat]

Is this finally the thread "My Fighter/Thief will beat your Sorcerer"? Because s/he will not

[mahe4]

The auto attack class will definitely beat the non auto attack classes in an auto attack duel, lol.

still the simulation proves how worthless constitution is!

@marjoupir thanks for your effort!

i hope you can program in some active abilities.

it would get quite interesting that way

[LadyCrimson]

I don't entirely get it, but then I'm not a very mathematical person... doesn't really seem like a class build per se, just an algorithm simulation testing to see who wins in clvl 1 auto-attack 1x1 fights, with a conclusion being made that the "winner" must be - what? best for playing the game? I think I'd agree with SilchasRuin ... not enough variables being tested to have a true conclusion.

 

Tis sorta interesting tho.

[KDubya]

How did estocs perform?

 

I screwed around with having characters "fight club" at taverns, auto attack only. Mainly for seeing differences in weapon types, armor and stats for fighters looking for good DPS while maintaining adequate tanking abilities. In my results, especially in plate, two handed did better than dual wield due to needing to bypass DR and estocs were runaways. I tested at level eleven,

 

 

I think the issue is testing level one characters. At that level there are no abilities or talents to differentiate the contestants. At level one stats have a massive impact on performance, hence the winning combo of hatchets, high res and per for the largest deflection. At high levels your stat bonuses matter less as you get talents that add 15 to deflection or 20% to damage, plus your accuracy becomes closer to your target's deflection.

 

At level one you have like a 30 accuracy and deflection would be like 46 for your winner. He'd be missed 31%, grazed 35% and hit 34% of the time.

[mahe4]

c'mon guys. instead of theorizing he started to program a simulation! he actually did something!

and all you guys do is complain?

you are all so pathetic!

[Glasny]

Interesting stuff about the constitution stat and dual wielding here

However, the race / class comparisions seem irrelevant because of the conditions of your test.

Still, thanks for this valuable data

 

What language did you use to code your program?

Edited April 22, 2015 by Glasny

[Veynn]

The Deadliest Warrior: Pillars of Eternity Edition

[temek]

Test doesn't really take enough variables into account and testing at lvl 1 isn't the best choice either. Some interesting results that seems to match my findings though. 

 

Race: Auamau is solid for the +2 MIG, but you skipped over the 2 best races (I understand that modeling them would be much harder): Moon Godlike (probable winner at any level as the racial ability effectively nearly doubles your endurance in any encounter) and Fire Godlike (only on high levels though as Battle-Forged is 2 burn dmg/lvl so it doesn't really do anything at low levels)

 

Attributes: Deflection > Endurance so no surprise that PER, RES top CON by quite a margin. MIG also should beat DEX at lvl 1 without the other dmg modifiers, but I'd expect these two to switch in higher levels once you start accumulating more additive dmg modifiers like weapon enchant etc.

 

Armor: No surprise at all that plate armor is the winner. It slows down you about 25-40%, more for dual-wielding less for two-handers. I assume everyone is at lvl 1 with non-enchanted weapons so even non-enchanted plate armor will negate most of the damage done.

 

Weapons: No surprise that you have Sabre at the top as DW sabres is the best DPS choice in the game according to my calculations by a decent margin. Hatchet probably gets in since +5 Deflection is quite significant boost at lower levels, but probably shouldn't be in here at all. 

Edited April 22, 2015 by temek

[gkathellar]

c'mon guys. instead of theorizing he started to program a simulation! he actually did something!

and all you guys do is complain?

you are all so pathetic!

 

Yeah, it's so pathetic that we ask for actual rigor before presenting test results as meaningful.

[Reent]

The test is:

We take all classes, take all abilities away, only the fighter keeps his most OP ability and the Paladin keeps a decent ability... so, from the beginning you could have made the test "whats the best weapon/shield/armor combination for auto attackers hitting each other" and whats a decent attribute combination in that case.

 

If you try some things with weapons and attributes you can get answers what weapons are best, however, thats it...

 

If you want more, you need the active abilities... spellcasters are hard, i would never try them (i wont try anything outside the game)

 

Tried your winner, beat him bloody with a monk... +15 accuracy ability fighter vs Torments Reach Monk, max con/might aumaua monk, thats why i didnt take knockdown with the fighter... 5 times out of 5 tries (platemale monk), naturally not nearly enough to come to a conclusion, however since i was almost able to win without armor (4 endurance for the fighter, 4 out of 5 times, the 5 the fighter hit too good and he won with almost no endurance lost), i dont really care, always think about what your test can and cant do... it cant tell you the ultimate duelist... it can help you choose weapons... maybe your stats... but i am not sure i would use them... like ever.

[mahe4]

 

c'mon guys. instead of theorizing he started to program a simulation! he actually did something!

and all you guys do is complain?

you are all so pathetic!

 

Yeah, it's so pathetic that we ask for actual rigor before presenting test results as meaningful.

 

it is pathetic to complain about an effort someone did, without doing anything yourself.

if the the whole test would be meaningless, because he made a mistake, okay... but we get actual data for assumptions we made.

in this simulation you can see (to a certain point) how important which attribute is.

as we thought, is the constitution attribute pretty meaningless.

it is also interesting, which weapon choices came out on top.

 

but hey! the simulation isn't perfect! so we need to complain about, that someone took his leisure time to program a simulation tool...

this behaviour is just simply pathetic!

Edited April 22, 2015 by mahe4

[petrivanzyl]

Could you run same test at level 6 and 12?

[marjoupir]

'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 ?

 

 

What language did you use to code your program?

Python.

Edited April 22, 2015 by marjoupir

[mahe4]

'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 ?

 

 

What language did you use to code your program?

Python.

you could also randomly pick talents.

and for active usage, you could generate random usage patterns. (1. ability use skill 1, 2. ability use skill 2, 3. ability use skill 1, etc) and also randomly generate the time between every use in these patterns.

but that would be a lot of work to implement, and i think, that i wouldn't have the time for that... so i don't blame anyone who don't want to take their time to code that...

[Nadrac]

I would watch a video of 30 generations punching each other in the face. Got to be said, they are not the brightest.

[MadDemiurg]

Think the main time sink for a better simulator would be just tedious coding of all the class and general talents. Tried that myself some time ago for a different game, didn't have the motivation to finish. Simulating active skills would be far too complicated and would probably kill algorithm convergence even if implemented, since usage patterns are problematic to "mutate" so that they approach the local maximum.

 

Adding levels and at least major passive talents would be a big improvement anyway.

Edited April 22, 2015 by MadDemiurg

[Kanasuke]

Cool idea! You should post on reddit as well if you haven't already

[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 !!

[petrivanzyl]

Maybe just ignore the regeneration? :dunno:

[Ineth]

That's pretty cool marjoupir, thanks for sharing the results.

 

Think the main time sink for a better simulator would be just tedious coding of all the class and general talents.

 

At that point you might as well use Bester's modding framework to run the duels as scripted fights in the actual game.

Edited April 27, 2015 by Ineth

[Luckmann]

Edit: Nevermind. I misunderstood the topic. Derp.

 

 

I'm not sure if this is *ultimate* but I'll just say that I've gotten a lot of mileage out of the following combination:

• Deathlike, for the added damage vs. anything below 25% Endurance.
• Bloody Slaughter Talent, a Passive that adds damage vs. anything below 10% Endurance.
• Vulnerable Attack Talent, a Modal that reduces Attack Speed but adds 5 Penetration.
• Savage Attack Talent, another modal that lowers your Accuracy a bit, but adds *1,20 damage.
• Weapon Focus Noble Talent, for the +6 Accuracy with Rapiers.
• Get the Sword of Daenysis Rapier from The Salty Mast in Ondra's Gift, make it Exceptional and then, when you can, Superb. An amazing all-round rapier that has extra attack speed and 3 Penetration.
• Never wear any armour. Ever. Clothes only.
• Never take the One-Weapon Style Talent. It is utterly useless.

I've been playing a Paladin, and a High-Resolve/High-Perception one to boot, which is something I wouldn't wish on my worst enemy. It's a trash build for the class. But I'm getting a ton of deathblows, which I can take advantage of with Inspiring Triumph and The Black Path. My main lament is the hilariously terrible Flames of Vengeance, which obviously favours heavy, slow weapons.

 

But if you use it on a Rogue, I'm convinced it could be a killed build for duelists. Duelists are still a bum deal compared to others, but the above works reasonably well.

 

Edited April 27, 2015 by Luckmann