cratio for two handers

[Araelus]

1,000 attack rounds in, and this is very boring but I will have data for you guys soon.

If anyone else is interested in helping, please let me or RingThree know. I am looking at cRatio tiers, and he is charting the distribution of "random" output values between pDIF min and max.

Both of us basically use the same data from K-Parser (courtesy of kparser - Google Code) thankyouthankyouthankyou, Motenten!

I would just need to know your STR, ATT, level, and the monsters in question - instructions on exactly how to filter the data before sending it in can be through PMs, and I'm sure RingThree can give more detailed instructions for his project.

Cheers.

[shaddix]

1,000 attack rounds in, and this is very boring but I will have data for you guys soon.

can i have your babies?

[Araelus]

Polynomial fit line looks like a normal distribution, but as you can see the sample size is still too small. Also because I am about to fall asleep this is biased towards the middle, since K-Parser is not grouping as I want it to. Uhhh.... can fix, WILL fix, but not right now.

Will keep working on this, but need to get some sleep. Later.

http://img21.imageshack.us/img21/184...ampletrend.png
"not to scale, or something"

LOL WHAT. Imageshack thinks a polynomial graph is porn. And now it's back THAT SURE WAS WEIRD HUH

[Max™]

Mathporn.

*fap³*

[Araelus]

It does have lots of curves. Really surprised me for imageshack to have it as the "this pic is porn or warez" for about 10 minutes.

Ok, I really need some sleep. At ~1.5 thousand attack rounds of data now.


Some Advice for People Collecting Data
Right now RingThree and I are grabbing info from Lesser Colibri.

Why Lesser Colibri?
Maths

Spoiler: show

1: no Level Correction to take into account
since they are all below L75

2: fSTR can be solved out because
fSTR can be capped for Soboro's weapon rank
regardless of what level the mobs are
Soboro Base Damage: 40
Soboro Weapon Rank: 4.444444~
Soboro lower bound fSTR = -4
Soboro upper bound fSTR = 12

testing STR greater than or equal to 110

L63: dSTR = 58
L64: dSTR = 58
L65: dSTR = 55

fSTR = (dSTR+4)/4

L63: fSTR = 15.5
L64: fSTR = 15.5
L65: fSTR = 14.75

all hit the cap
and will be reduced to fSTR = 12

3: the mob's DEF variation does not take me out of one of Nameless' proposed Ratio tiers
(tier I) 1.8 < cRatio < 2.2
(tier II) 1.2 < cRatio < 1.8
(tier III) ? < cRatio < 1.2

so long as I am careful with my attack value

with Attack = 383
L63: 1.65800 cRatio
L64: 1.62979 cRatio
L65: 1.58921 cRatio

What You Need in Order to Help

Spoiler: show

1. a fast-attacking two-handed weapon

• TWO HANDED WEAPONS ONLY
• even with multihit this gets boring fast

2. Attack: must be in one of the following three ranges

• greater than 440
• between 320 to 390
• less than 254

3. STR:

• Bahamut's Zaghnal? fSTR capped at 104 STR
• Soboro Sukehiro? fSTR capped at 110 STR
• Love Halberd? fSTR capped at 120 STR
• Fortitude Axe? fSTR capped at 122 STR
  • other weapons can contribute as well, but your weapon's rank will determine whether or not you can cap fSTR.
  • Note that I have slightly overestimated Strength required to cap fSTR just to make sure it is capped.
  • Note that capping fSTR is not required for all testers, just some of them.

4. Consistency throughout

• do not change gear or weapon
• do not use Weapon Skills
• no added effect damage
• use the least buffs possible, none at all if you can
• ESPECIALLY NO BUFFS THAT CHANGE ATTACK or STRENGTH!
• NO DEBUFFS ON THE MOB!

The Tricky (Sticky?) Wicket

Spoiler: show

K-Parser records the XP return from each mob so you can tell its level.

With NO Sanction

XP    LEVEL
66	63
70	64
72	65

(I slapped on Signet to avoid XP bonuses, which could change as the Imperial Thingy changes.)

Why is this important? With uncapped fSTR, that value will be different depending on the mob level. And for all testers, Ratio will be slightly different on each mob level! So we need to separate out the data into three sets, one for LV63 Lesser Colibri, one for LV64, one for LV65.

When you look at the Offense Detail, you can "Group Mobs". This will sort them by XP return, and thus level.

The data retrieval from K-Parser and collation into a big Excel file is a stressful part of the process, yet considering how much error can occur just in data collection, this bit is important but isn't my primary concern at the moment. Spending ~50 hours :D fighting mobs is, since Lesser Colibri is only the start of this process.

K-Parser sometimes includes your Seigan Counter swings in with "normal" melee hits - and these can crit but will be marked as a normal swing - so make sure to pay attention and remove these hits. I do not know why it is sometimes included and sometimes it is not.

(Or don't use Seigan, either.)

[Araelus]

2,402 attack rounds later:

average damage = 77.0008
σ = 12.643
R² = 0.7842

I'd like the coefficient of determination to be above 0.9 but I think we can say the output of pDIF between min and max is based on a normal distribution.

Going from R² = 0.5292 to 0.7842 is GREAT!

I'll continue gathering data over the next few days with these settings, to compare with Ring's capped fSTR results, before we try other Ratio values.

If you want to find more mobs to test besides Lesser Colibri, please find out the mob's level range and DEF and VIT.

A monster checks as high defense if your Attack is lower than its Defense.
A monster checks as low defense if your Attack is at least 25% higher than its Defense.

You'll discover the AGI from your crit rate during damage parsing, but VIT? Help!

[Araelus]

I am currently at 9,885 normal attacks out of 12,053 attacks total. Whew! A busy day! :D

Here is what we can discover from this sample:
really tired of imageshack breaking

9885sampletrend.PNG (image)

chart does not show all the frequencies in those little numbers, just some to help you along

Notice the R² value has not increased very much. Alas, nothing is perfect. Note the max and min damage values are within two standard deviations from the mean. That is expected for a normal distribution, where ~95% of values are within ±2*σ.

http://upload.wikimedia.org/wikipedi...iagram.svg.png

From this basis, we can begin to find the new two-handed Ratio → formulas with a high level of confidence in the pDIF(min) and pDIF(max) values.

Once these are found, we can find out what the level correction is for two-handed weapons by fighting mobs and seeing what the Ratio is reduced to (cRatio) since the damage on a >lv75 monster will output pDIF values at a point farther left than expected from the strict ATT/DEF Ratio.

Afterthought: a flat distribution's coefficient of determination on this data is less than 0.0036, so yeah, the green line looks squiggly but it's closer to a bell-curve than you'd think from eyeballing. As our sample size increases, I have not changed the number of independent variables, so we can safely draw a meaningful comparison of models going from 1,000 attacks to 2,402, then to 9,885.

[VZX]

From what I see... it's more like trapezoid than a bell curve.

I wonder if you can produce R square closer to 1.00 with 3 piece-wise linear functions instead of 1 polynomial to fit them all.

So from your last graph, the partition would be on : 50~61, 62~92, 93~104. The range for each partition is 11, 30, and 11..

[Araelus]

From what I see... it's more like trapezoid than a bell curve.

I wonder if you can produce R square closer to 1.00 with 3 piece-wise linear functions instead of 1 polynomial to fit them all.

So from your last graph, the partition would be on : 50~61, 62~92, 93~104. The range for each partition is 11, 30, and 11..

I hadn't thought of that >_<;

9885sampletrendpiece.PNG (image)

Works well for the side ranges, but the middle is below 0.03 with every kind of fit-line I could try.

Also, imageshack is marking all of these graphs as illegal images, and I have no idea why. So blogger URLs for a quick fix.

[VZX]

Oh wow... It's really that bad huh for the middle zone. But I guess I'm not that surprised since you still can see that it has high frequency.

But let's see... Is there a much better quadratic polynomial fitting line for the middle zone?

[Araelus]

That 0.0258 R-squared is for a quadratic polynomial fit.

My null hypothesis was that the pDIF output was equally distributed between (min) and (max) based on RingThree's ideas. That has been disproven.

However, we either use a partial-fit polynomial along the whole range, or the middle is out of sorts.

At least the min/max adhere very closely to two standard deviations from the mean.

The more troubling part of my current finding, to my interests, is the extremely low frequency of the results at the upper and lower bounds. 9 instances ad 13 instances out of 9,885 attacks mean it is very rare to see the actual pDIF(min) or (max).

I want to find the new pDIF function for two-handed weapons. When Studio Gobli (?) came up with the pDIF functions, how many attacks at each Ratio value (our independent variable) did they use?

It should be easy to get the pDIF ~min and ~max values from the middle, high-frequency range. But we'll know these are not the true min and max pDIF.

But that assumes I can even know my Ratio value for enough tests to get a function.

Without an army of characters with attacks ranging from 1 to 400, I'm going to have a difficult time using /check to find out what mobs' Defense is, and thus .

On mobs that don't steal food or use Def+ moves, I'll do with food and without food for multiple Ratios, but I don't have their Defense value to begin.

Since you're the reigning expert on this, VZX, does such a database of DEF and VIT exist? I've run around your site but can't seem to locate what I need.

Thanks.

if it matters, this huge aggregate chart is with an average Ratio of 1.63

If I separate out LV63/64/65 we see identical results in terms of upper and lower bounds' distance from the average, R-squared, frequency shape, and so on, thus I have left them all together for ease of consideration. When I reach 10,000 attacks on each level of Lesser Colibri I will then sort it out into three graphs to be clearly rigorous.

The difference in Ratio across the three levels of mob is only 0.0433%.

[VZX]

I want to find the new pDIF function for two-handed weapons. When Studio Gobli (?) came up with the pDIF functions, how many attacks at each Ratio value (our independent variable) did they use?

I have no idea.

Since you're the reigning expert on this, VZX, does such a database of DEF and VIT exist? I've run around your site but can't seem to locate what I need.

The old page from gobli's calculator have some info from older (Zilart) mobs
http://f4.aaa.livedoor.jp/~rmkta/ws/enemy.html
3rd column is crab, 4th column is beetle.

if it matters, this huge aggregate chart is with an average Ratio of 1.63

If I separate out LV63/64/65 we see identical results in terms of upper and lower bounds' distance from the average, R-squared, frequency shape, and so on, thus I have left them all together for ease of consideration. When I reach 10,000 attacks on each level of Lesser Colibri I will then sort it out into three graphs to be clearly rigorous.

Thanks, I was about to ask the chart.
With BD = 55 and cRatio average = 1.63, you have BD*cRatio = 89.

And from the graph, it looks most of the number generated is less than that number, in this specific region.

I'm interested to see the excel sheet once you done gathering the data.
For now, could you tell me the max-min damage range for each level? in that chart I see min=52 max=103, am I correct?

The difference in Ratio across the three levels of mob is only 0.0433%.

0.0433%? what is this number from?

EDIT: ok now fixed

[Araelus]

The old page from gobli's calculator have some info from older (Zilart) mobs
http://f4.aaa.livedoor.jp/~rmkta/ws/enemy.html
3rd column is crab, 4th column is beetle.

Thanks, that's good stuff. :)

With BD = 55 and cRatio average = 1.63, you have BD*cRatio = 89.

And from the graph, it looks most of the number generated is less than that number, in this specific region.

Assuming

Spoiler: show

Soboro Base Damage: 40
Soboro Weapon Rank: 4.444444~
Soboro lower bound fSTR = -4
Soboro upper bound fSTR = 12

testing STR = 66+18 = 84
L63: 52 VIT
L64: 52 VIT
L65: 55 VIT

L63: dSTR = 32
L64: dSTR = 32
L65: dSTR = 29

fSTR = (dSTR+4)/4

L63: fSTR = 9
L64: fSTR = 9
L65: fSTR = 8

shouldn't it be base damage 49, 49, 48 (40+ fSTR) for L63, L64, L65?

If Attack = 383, and
L63: Defense 231 → Ratio = 1.658
L64: Defense 235 → Ratio = 1.630
L65: Defense 241 → Ratio = 1.589

0.0433%? what is this number from?

Uh, right, whoops. 4.34% from 1.658 to 1.589 >_<;

I'm interested to see the excel sheet once you done gathering the data.
For now, could you tell me the max-min damage range for each level? in that chart I see min=52 max=103, am I correct?

L63: damage 57-103
L64: damage 54-102
L65: damage 52-101

Also, from these samples I find the likelihood of getting pDIF = pDIF(min) or pDIF(max) to be extremely small, at most 9/3732, or 0.2411%, from the L65 birds.

Just going from the L65 birds (3732 attacks, range of 52 to 101 damage, Ratio = 1.589, base damage 48)

If my upper and lower damage bounds are at the actual pDIF(min) and (max)

Level	Ratio	dmg min	dmg max	pDIF min	pDIF max	f(min)	        f(max)
63	1.658	57	103	1.163265306	2.102040816	0.701607543	1.267817139
64	1.630	54	102	1.102040816	2.081632653	0.67609866      1.277075247
65	1.589	52	101	1.083333333	2.104166667	0.681770506	1.324208097

Nameless suggested in this range it might be

pDIF(min) = cRatio x 0.75
pDIF(max) = cRatio x 1.25

Combined with the difficulty of actually hitting for pDIF(min) or (max) damage, it's a bit shaky in terms of "Ratio x ?"
I would suggest

pDIF(min) = cRatio x 0.66
pDIF(max) = cRatio x 1.33

in the range of (1.589 ≤ cRatio ≤ 1.658)

Again, I have the luxury of Ratio = cRatio because of the mobs' level, and that was on purpose.

If Lesser Colibri do not have the DEF and VIT values as otherwiki lists... then my Ratio and fSTR numbers are wrong.

Against L65 Lesser Colibri, my critical hit rate of 12.86% (5 base + 4 merit + 3.86 from dDEX) indicates that my dDEX is 35. However, my DEX was 79 and the listed AGI for L65 is 65, a dDEX of 14.

Then again, against L64 Lesser Colibri my critical hit rate was 13.36%, and for L63 Lesser Colibri, my critical hit rate as 11.90!

L63: 2.9% from dDEX
L64: 4.36% from dDEX
L65: 3.86% from dDEX

So I guess even is not 12,055 attacks total are enough to get an exact Critical Hit rate? :(

Are the otherWiki Lesser Colibri stats right or wrong?

If I understand the /check function correctly, if the mob's Defense is 25% lower than the player's Attack, it will check as "Low Defense"? We should be able to chart out an exact DEF number from this, but I do not know how to get VIT values.

I am also depending on RingThree's tests, which are on these same mobs, at nearly the same Ratio as me, but with much more STR to try and make sure fSTR is capped on Soboro at +12 to base damage.

[VZX]

Ooops, I forgot to impose fSTR cap of soboro.
Then 79 is pretty good standing in the middle.

As for lesser colibri AGI, I doubt it's 65 AGI at that low level. Because we know the level 81-82 have 67 AGI.

f I understand the /check function correctly, if the mob's Defense is 25% lower than the player's Attack, it will check as "Low Defense"? We should be able to chart out an exact DEF number from this, but I do not know how to get VIT values.

To be precise, if your attack = 1.25 * mob defense (truncated) or higher, then /check will tell low defense.
For high defense, if mob defense = your attack, it will show nothing, but if mob defense = your attack+1, it will start showing high defense

[Araelus]

To be precise, if your attack = 1.25 * mob defense (truncated) or higher, then /check will tell low defense.
For high defense, if mob defense = your attack, it will show nothing, but if mob defense = your attack+1, it will start showing high defense

Ok, then AGI aside, I will mess around with my ATT rating until I can be sure of Lesser Colibri's exact defense at all three pop levels. Thanks for clarifying that.

Still no known test for VIT, though, unless we know all the other variables, so I'd have to get some 1-handers in on this, then back out VIT values from Base Damage's fSTR component.

And I'm dissatisfied with the haziness of fSTR as (fSTRtwo)/2 based on

dSTR range 	fSTRtwo approximation
12 or greater 	(dSTR + 4) / 2
6 to 11 	(dSTR + 6) / 2
1 to 5   	(dSTR + 7) / 2
-2 to 0 	(dSTR + 8) / 2
-7 to -3 	(dSTR + 9) / 2
-15 to -8 	(dSTR + 10) / 2
-21 to -16 	(dSTR + 12) / 2
-22 or less 	(dSTR + 13) / 2

http://www.bluegartr.com/forum/attac...2&d=1238573991
All those plateaus irritate me. Why isn't it one linear function?

Anyway, from the testing so far, with the very low rate of actual pDIF(min) and (max) damage returns, is it acceptable to have a slightly hazy pDIF function?

I don't want to transition to other Ratio values to fill out the pDIF lines until this gets settled. On that note, I really appreciate the help and feedback from you, VZX. :D

Imageshack continues to flag these graphs as warez (WTF) so I hope the forum is OK with me uploading here. Blogger won't let me show Blogger-hosted images on other sites at all.

[VZX]

All those plateaus irritate me. Why isn't it one linear function?

Try some regression lines again and discretize the final value to integer?

To be fair, even before the 2-hander adjustment, studio gobli still put notice that the pDIF function they make still "in progress". Meaning that, it's their best model to capture the real function. I guess, as long as the function doesn't skew the average by much, it's fine to go with such approximation.

[Masamune]

any news on this topic Araelus ?

I'm very interested into yur tests, since i encountered kinda same wtf conclusions on Lcolis / lvl1mobsRonfaure / GcoliMeripos, concerning Maxcritdmg (sometimes doing higher dmg than calculated value); and by extension....cRatio Ratio lvlcorrection functions and caps etc.... i bet one of these are either completely wrong nowadays or litterally uncapped (ie not floored at all) ?
I even found one day on WAR forums Allakhazam some dmg calc who were simply multiplying old PDIF function (or cRation i don't remember) by 1.05, for 2Handers, don't know at all where the author got this info from.

[Masamune]

Found another juicy info on KI MNK forum :

Seedo, author of some nice MNK calculator, managed to estimate new caps (still using same old melee functions) :
. cRatio caps @ 2.064 instead o 2;
. Max Crit Dmg caps @ 3.15 instead of 3;
. Crit Dmg is raised by 1.06 instead of 1.

Tried myself those caps into my own calculations, made all my results match my parses o.O

[Starr]

eh um isn't 2handed cRatio/pDIF different to one handers now? H2H comes under 1 handers?

[Miraun]

Try some regression lines again and discretize the final value to integer?

To be fair, even before the 2-hander adjustment, studio gobli still put notice that the pDIF function they make still "in progress". Meaning that, it's their best model to capture the real function. I guess, as long as the function doesn't skew the average by much, it's fine to go with such approximation.

If our underlying assumptions of cratio, ratio, etc. are flawed... how can we properly attempt to determine DEF or VIT to be able to accomodate for any concrete values based on the damage delt?