Dual-Wielding KC and Ridill(or M.kris)

[Byrthnoth]

Naw, it isn't, though we have no solid proof it works that way.

27/53/20 with 10% DA
27 = .9*S; S=30
20 = .9*T; T =22
53 = D+.1*S+.1*T; D = ~48

Keep in mind that the margins of error make this correction almost not worth doing.

I'm assuming Brutal/Warrior's Cuisses/Fighter's Calligae is +9% DA. Last I heard people thought Fighter's was 1% and Warrior's was 3%. Brutal has been shown to be 5%.
25 = .81*S; S=~31
15 = .81*T; T=~18.5
60 = D + S*.19 + T*.19; D=~50.5

[Masamune]

huh? i always thought AF2legs = 1%, not 3% o.O

btw, updated M/Kris results @ http://www.bluegartr.com/forum/showp...0&postcount=19 with another parse for confirmation, there were apparently a parsing problem with 1st parse. 2nd parse shows so far a nice 30/50/20 ratio, so assuming all weapons having "occasionally attacks 2 to 3 times" share same distrib, Mkris results could be used as proof for Ridill ?

[Retrieving/C.D/Searain]

Dang... was happy to see your results till that last sentence made me like "orz...".

I'm sorry, Warrior was the only 73+ job I had at the time that was capable of equipping the sword, I clearly recall trying to convince a few friends to go out and test it on jobs such as DRK, BST and the likes but people didn't really seem to give a damn D:

That's simply the best I could do alone given the circumstances. Also, I remember doing this more to disprove the 33/33/33 rumor than anything.

[Masamune]

I'm sorry, Warrior was the only 73+ job I had at the time that was capable of equipping the sword, I clearly recall trying to convince a few friends to go out and test it on jobs such as DRK, BST and the likes but people didn't really seem to give a damn D:

That's simply the best I could do alone given the circumstances.

Np, i can understand as i'm having same problem (with any test actually). Poeple so prefer to argue without actual test numbers in forums, that's so easier right...

[CDF]

If you had the inclination, you could derive (maximum likelihood) estimators of the original proportions of single, double, and triple attacks based on Searain's data and on the current idea of how DA adjusts the proportions. Directly "proving" the original distribution is not really important since the whole point was to show that the proportions change in the presence of DA.

Comparing the two data sets given (chi-square test for equality of proportions) shows that Brutal Earring/Warrior's Cuisses/Fighter's Calligae change the distribution of attacks relative to /WAR DA alone. It is more of an inductive leap to assert that DA trait takes priority over Ridill in the fashion described, but it's enough (to me) to call it a reasonable working assumption.

There is some data on Ridill for non-WAR jobs, but the small sample sizes mean merely that 30:50:20 is possible, but in light of the previous results it's a reasonable starting point:

http://www.qcdn.org/ffxi/showthread.php?t=50849

Some results:

I tested the Multi-Hit rate of the Ridill, and compared it to different job combinations. This was to answer the question: "Does Double Attack reduce the amount of hits produced by a ridill." Where two theories arose:

Theory 1: Based on a test a player did(Searain of Garuda) on BG/FFXIclopedia, they killed Robber Crabs in Kuftal tunnel for hours on end and put in 1000 attack rounds fully geared, sans any Double Attack enhancing gear/food, and tested the results. His results showed that if you added double attack gear, your Triple attack rate was reduced diproportionately to the number of Single attacks.

Theory 2: Based on assumptions the Ridill should lose an equal amount of Singles as Triples into doubles with the added presence of the Job trait/enhancements. Based on the premise that it would be easier to code.

Test configuration:
Lebros Cavern Assault: "Excavation Duty"
Jobs:

Drk/Drg.
Thf/Drg.
War/Drg.
War/Drg + Askar Korazin & Brutal Earring.

Merits: Triple Attack: 5/5, Sword 8/8.
(My war is pretty much unmerited... No DA merits, etc.)

Method: Enter assault and hit rocks completely naked, save for the weapon, and Korazin/Earring in the final test.

Results:


Drk/Drg:

Triples: 17.21%
Doubles: 50.93%
Sinlges: 32.09%
Total: 430 Rounds. 795 Swings. (1.85 Swings/Round)


Thf/Drg:

Triples: 21.86%
Doubles: 47.21%
Sinlges: 29.93%
Total: 430 Rounds. 821 Swings. (1.91 Swings/Round)


War/Drg:

Triples: 15.81%
Doubles: 55.35%
Sinlges: 28.84%
Total: 430 Rounds. 804 Swings. (1.87 Swings/Round)


War/Drg + Askar Korazin & Brutal Earring:

Triples: 18.37%
Doubles: 59.77%
Sinlges: 21.86%
Total: 430 Rounds. 845 Swings. (1.97 Swings/Round)

[CDF]

huh? i always thought AF2legs = 1%, not 3% o.O

btw, updated M/Kris results @ http://www.bluegartr.com/forum/showp...0&postcount=19 with another parse for confirmation, there were apparently a parsing problem with 1st parse. 2nd parse shows so far a nice 30/50/20 ratio, so assuming all weapons having "occasionally attacks 2 to 3 times" share same distrib, Mkris results could be used as proof for Ridill ?

Here's some other data on M. Kris (JP). Obviously 10:80:10 is wrong.

[Masamune]

Yes, i did this testing because i suspect total allowed multihits to not go above 70%-80%, this JP test shows literraly 90% multihits i was like "huh? o.O"

Also, I am interested to see some numbers on Merc.GreatSword for "2-4 times" distribution, and either Ridill or BahaZaghnal to prove asumption all "2-3 times" weapons share same distribs.

[Hitoseijuro]

A KC will average a very slight fraction better than a KC/M.Kris

However those numbers assume no dualwield gear, adding in a suppa means the KC/M.Kris becomes a touch better, adding both a suppa and Augmented Nuevo pushes it to about half of a swing. Ridill still loses with those extra.

One thing I would like to add to this that, Im not aware how much of a difference it would make but, if kc/m.kris are slightly better with dual weild gear on, neuvo/suppa, have you factored in what you are losing out on?

Nuevo, means no JSE body, so you would probably have to have it on legs, which would put you -3% haste, -26HP, than comes the earring, another 25-40HP at best. This is with /nin so you would be using the wyvern targe, and losing out 1% haste.

Thats -4% haste and -31~46HP(nuevo HP calculated) I dont know how big of a difference this would make, but it is something to consider esp the haste, unless youre going to sac -2% from Souleater, than thats another calculation you have to make.

[CDF]

For Searain's first data set (n = 1020), I obtained the following MLEs for Ridill's single, double, and triple attack proportions, assuming DA trait procs take priority over ("override") Ridill procs.

1 attack: .3006536
2 attacks: .4782135
3 attacks: .2211329

Out of laziness, I obtained individual 95% confidence intervals for the single and double attack proportions using the basic bootstrap:

1 attack: (.2712418, .3311547)
2 attacks: (.4444444, .5119826)

So, .30 and .50 are within the realm of possibility. You don't necessarily need to obtain data on a non-WAR job ... given all the assumptions are correct.

[Masamune]

For Searain's first data set (n = 1020), I obtained the following MLEs for Ridill's single, double, and triple attack proportions, assuming DA trait procs take priority over ("override") Ridill procs.

1 attack: .3006536
2 attacks: .4782135
3 attacks: .2211329

How did you get those numbers ? Are you sure you counted correctly the "2" and "3"? because if you read my post just after searain's, you'll see i don't find same %s :s

So, .30 and .50 are within the realm of possibility. You don't necessarily need to obtain data on a non-WAR job ... given all the assumptions are correct.

Actually, the problem is not there: it's proving that DAtrait from war could have skewed Searain's data, and by how much? only thing i got to prove that are the model we built together some months ago, and the Mkris results (assuming Ridill share same distrib as Mkris).
Looks like you skipped a good portion of the thread ?

[CDF]

The whole point of my example, which is redundant given Byrthnoth's last post, is to show that you can account for existing DA to obtain an estimate of Ridill's multi-hit distribution without the effect of DA. (I just used the actual proportions instead of the rounded values.)

First, it is easy to show a significant difference between Ridill/DA 17% (Cuisses is +1 per description) and Ridill/DA 10%, so some DA does alter Ridill's multi-hit distribution. Based on this result, it is easy to justify the first sentence, because why wouldn't DA +10% from /WAR alone affect Ridill's distribution?

Sure, the data from Ridill/DA +10% could reflect 30:50:20, but this more likely a result of insufficient sample size than DA +10% not doing anything based on the previous observation (DA does alter the multi-hit distribution).

So, how to explain the result. (It is easy to estimate how much skew, but explaining it is beyond statistics.) There are only a couple of realistic, simple mechanisms that would explain the difference between Ridill/DA 17% and Ridill/DA 10% (invoking Occam's razor), one being that DA trait procs, if they occur, always occur before Ridill's, so there shouldn't be a problem invoking this mechanism to explain the empirical result.

[Xanthu]

One thing I would like to add to this that, Im not aware how much of a difference it would make but, if kc/m.kris are slightly better with dual weild gear on, neuvo/suppa, have you factored in what you are losing out on?

Nuevo, means no JSE body, so you would probably have to have it on legs, which would put you -3% haste, -26HP, than comes the earring, another 25-40HP at best. This is with /nin so you would be using the wyvern targe, and losing out 1% haste.

Thats -4% haste and -31~46HP(nuevo HP calculated) I dont know how big of a difference this would make, but it is something to consider esp the haste, unless youre going to sac -2% from Souleater, than thats another calculation you have to make.

It hasn't been, I honestly don't know what usual gear would be. But that's why I have been stressing "This does not take account any other factors"

[Spider-Dan]

Yeah, supposedly Mercurial Pole attacks one more time than it's supposed to as well.

There was a screenshot on KI a while ago that seemed to indicate that M.Kris is also actually 2-4. As I recall, it was BD+M.Kris combo, and the screenshot looked something like:

70 damage
14 damage
12 damage
69 damage
14 damage
73 damage
10 damage
14 damage
12 damage
10 damage
72 damage
etc

[Byrthnoth]

Mercurial Pole Attacking 5x - http://www.youtube.com/watch?v=30w6bkye_6Q
Mercurial Sword Attacking 4x - http://wiki.ffxiclopedia.org/wiki/Talk:Mercurial_Sword
There's plenty of testimony for this happening with 0% Double Attack. Yet it doesn't seem Mercurial Kris hits 4x when single-wielded, according to Masa' tests. If anything, we'd expect the parser to combine the attack rounds and give 4x procs when there weren't any, and yet none appeared.

[Motenten]

Also, i'm not sure at all if the slow18% i used to parse is enough to avoid KParser from mixing rounds between them, with a delay 196 dagger. Motenten can maybe answer that question tough?

Delay 196 for a weapon that can triple attack is going to be tricky to parse. Multiple attacks from the same round tend to get sent to the chat log at a rate of about 1 or 2 per second, so 3 attacks can take up to ~2 seconds to end up in the log. That leaves just 1 second between the end of the last attack reported from one round and the first attack of the next round, which is the same diff as between the individual hits of a single round. That can easily mess up the overall filtering.

18% slow puts your delay at about 232, almost 4 seconds. The followup test with 31% slow puts your delay at 257. 257 should give a full 2 seconds between end of one round and beginning of the next (worst case), which should be enough to adequately separate attack rounds correctly.

Ideally for any multi-attack rate testing you want at least 2 seconds between maximum end of one round and start of the following round.

[Masamune]

ok thank you Motenten, That might explain the differences observed in 1st parse, compared to the 2nd. (the 2nd one pretty much confirming the 30/50/20 asumption with half less sample size o.O)

According to your explanation, 1st parse with 232delay would let only, in worst case:
4secs round duration - 1sec/hit animation*triple attack = 1sec
..while 2nd parse with 257delay would let 2secs between rounds, ok undertood.

Btw, Merc.GS is 600k on my server, meh...

[Motenten]

Masamune, do you think you could send me those two sample parses (both the 18% slow and the 31% slow)? I've rewritten the Extra Attacks plugin and would like to see if it gives more consistent results.

[Nameless]

put on full hecatomb stuff and parse MK will be the best way to test.

[Xanthu]

I had a sudden idea of possibly helping to parse, it requires some attention but not as much as direct counting. Try to turn away from the fort after every swing, let animations finish, turn back. It probably still wants the slow% to help avoid actual swings, dunno how much a pain that may be.

[Derian_Ifrit]

Ok, now for the important question:

Is the increase in hits/sec enough to make KC/M.kris combo beat KC/30HP or KC/1% haste?

Would it not be either KC/30HP or KC/6% I.E. /drg earring instead of suppa.