Everything you need to know about heroes, and more

Range - 85250

[video=youtube;TKOplvgbKnw]http://www.youtube.com/watch?v=TKOplvgbKnw[/video]

Highest
1:22 - 72864? crit
2:56 - 574850 non crit
21:26 - 942096 non crit
can't check highest crit, hits beyond cap

Too lazy to check for lowest damage, maybe someone else can.

Some facts (arguable facts):
48% combat mastery + 15% ignore defense on axe = 63%
PDrate: 0.5 -> 0.35 due to threaten, yes it lowers PDrate, I'll certain that it does I'm almost sure it's multiplicative too and full effect, I tried doing half PDrate reduction like how power guard only reduces 25% touch damage at bosses, but it won't fit the data.
0.35 + 0.65 x 0.63 = 0.8705

214% (IS) x 210% (advanced combo) x 1.3 (boss damage) x 1.08 (total damage) x 1.25 (chance attack, yes, threaten's acc reduction counts as blind) x 0.8705 (PDrate) = 6.8656x multiplier

85250 x 6.8656 = 585293 (compared to 574850)
85250 x 6.8656 x (2.1 + 0.65) / 2.1 = 766455 (compared to 72864x)
85250 x 6.8656 x 1.64 (enrage) = 959880 (compared to 942096)

If I treat boss damage and total damage as additive,

585293 x 1.38 / (1.3 x 1.08) = 575287 (compared to 574850)
766455 x 1.38 / (1.3 x 1.08) = 753353 (compared to 72864x)
959880 x x 1.38 / (1.3 x 1.08) = 943471 (compared to 942096)

Which leads me to believe they are additive. Yes I was the one that said they were multiplicative pre-BB, but it could have just been the different defense formula. Critical formula is still off, I just can't seem to figure it out.

Maybe some of you have better eyes than I do picking out damage. Let me know and see what you think.

Range - 85250

[video=youtube;TKOplvgbKnw]http://www.youtube.com/watch?v=TKOplvgbKnw[/video]

Highest
1:22 - 72864? crit
2:56 - 574850 non crit
21:26 - 942096 non crit
can't check highest crit, hits beyond cap

Too lazy to check for lowest damage, maybe someone else can.

Some facts (arguable facts):
48% combat mastery + 15% ignore defense on axe = 63%
PDrate: 0.5 -> 0.35 due to threaten, yes it lowers PDrate, I'll certain that it does I'm almost sure it's multiplicative too and full effect, I tried doing half PDrate reduction like how power guard only reduces 25% touch damage at bosses, but it won't fit the data.
0.35 + 0.65 x 0.63 = 0.8705

214% (IS) x 210% (advanced combo) x 1.3 (boss damage) x 1.08 (total damage) x 1.25 (chance attack, yes, threaten's acc reduction counts as blind) x 0.8705 (PDrate) = 6.8656x multiplier

85250 x 6.8656 = 585293 (compared to 574850)
85250 x 6.8656 x (2.1 + 0.65) / 2.1 = 766455 (compared to 72864x)
85250 x 6.8656 x 1.64 (enrage) = 959880 (compared to 942096)

If I treat boss damage and total damage as additive,

585293 x 1.38 / (1.3 x 1.08) = 575287 (compared to 574850)
766455 x 1.38 / (1.3 x 1.08) = 753353 (compared to 72864x)
959880 x x 1.38 / (1.3 x 1.08) = 943471 (compared to 942096)

Which leads me to believe they are additive. Yes I was the one that said they were multiplicative pre-BB, but it could have just been the different defense formula. Critical formula is still off, I just can't seem to figure it out.

Maybe some of you have better eyes than I do picking out damage. Let me know and see what you think.

Hi, I believe the additive aspect is somewhat closer to the observed data as, it is not that easy for the randomized function to hit the max dmg, since the damage range is continuous and not discrete ( maybe becomes discrete after rounding off those decimals, or using floor or ceiling functions...)

(note the range is 59675 - 85250, which gives a range of 25 575.)

Now supposed we have a step function with 25 575 steps.
The probability of falling into the last step is 1/25 575 = 0.00003910.
This means that the probability of your dmg showing the maxed dmg as in 943471 is 0.00003910.

Which means that the margin of error from the observed data is very small.

585293 x 1.38 / (1.3 x 1.08) = 575287 (compared to 574850)
766455 x 1.38 / (1.3 x 1.08) = 753353 (compared to 72864x)
959880 x x 1.38 / (1.3 x 1.08) = 943471 (compared to 942096)

However, I am not sure about the pdr formula part, could we verify the pdr part, by trying it out on say zak ,czak and other stuffs with different pdr.

On a side note,

A late gratz on getting a ignore pdr of 15% on your axe coupled with bossing 30.

That was real awesome, by the way, how much cube did you use to achieve that?

That was real awesome, by the way, how much cube did you use to achieve that?

More than 1200 cubes I blif, he mentioned it in the other thread

The 1200 cubes was spent on many different weapons, i think he did the axe 300-400 times trying for 30 + 30, then settled for 30 + 15

Fleyx has a video on basil showing that 30% boss and 6% or 8% total damage are additive, easy to figure it out with the shadow partner with no error at all.

The PDrate is hard to test but I'm almost sure that's what it works with threaten, even the numbers agree with me, everything else regarding non crit is pretty much confirmed now so I would say PDrate is too due to the numbers. Critical is still a mystery to me, I wish there's like a 28% crit weapon I can use to make it easier. =/ Bad mastery + bad crit rate = hella hard time to test.

Axe took like 400 cubes in total, that's including 200 recubing after I decided I want boss and ignore defense on it. It was 30% boss 8% total damage and 53 mp recovery before I recubed.

Perhaps a paladin with its 94% mastery will reduce the standard deviation of crit and help to answer alot of problems.

Anyway, I did some thinking and wanted to do some stuffs with my excel but i didn't know how to play with binomial distribution function in it based on an earlier information.

959880 x x 1.38 / (1.3 x 1.08) = 943471 (compared to 942096)


Now supposed we have a step function with 25 575 steps.
The probability of falling into the last step is 1/25 575 = 0.00003910.
This means that the probability of your dmg showing the maxed dmg as in 943471 is 0.00003910.

Now 943471 - 942096 = 1375
25575 - 1375 = 24200
24200/25575 = 0.9463

Now P( of doing >= 942096 uncrit dmg ) = 1375/25 575 = 0.05376

Supposed in a zakum run where he gets to spam 6000 intrepid slash.

What is the probability of at least 1 intrepid slash hitting at least 942096 ?

probability of at least 1 intrepid slash hitting at least 942096 = 1 - (1 - 0.05376)^6000= 1- (0.94236)^6000
= 1 ( or close to 1)

What about the exact probability of getting that elusive 943471 in 6000 slashes?

= 1 - ( 1 - 0.00003910)^6000 = 0.21

Conclusion:

The observed data seems to fit the theoretical data with a marin error of 6.4%?

The problem is combo+crit interaction though, pretty much every other class just treats crit like a separate multiplier.

Everything seems to fit without critical. But yeah, like Stereo, this is only the problem with advanced combo and enrage, and perhaps even skills the boost certain skill damage by +%s, like passive boosts on shadowers, perhaps corsair's homing, etc. Remember how critical used to be (brandish % * advanced combo + critical %)? This is exactly what's happening again except I just can't seem to get a number close to the theoretical max at all. I just need like a critical potion of something, but a 20% crit weapon will be nice since I can sharp eyes myself, 10% crit ring, and passive 5%. Testing non crit isn't bad, since it's just uniform distribution, but with crit it's triangular and the chance of hitting high end just pummels.

This thread murdered my hero T_T

but with crit it's triangular

i want proof it isnt uniform. have a link or something?

Range and crit % are uniform, the product (and thus crit damage) is triangular.

looking for something a little more solid than a fact about how to make a triangle distribution. how do you know the game takes those two (randomed) numbers and multiplies them instead of multiplying minRange * critmin and maxRange * critmax and then selecting a single random number from the new range?

looking for something a little more solid than a fact about how to make a triangle distribution. how do you know the game takes those two (randomed) numbers and multiplies them instead of multiplying minRange * critmin and maxRange * critmax and then selecting a single random number from the new range?

Because with the selection of said number, by your statement, the game would be unable to determine the number's "color" (aka Crit = red, normal = yellow).

The game most likely follows a pattern such as (pseudocode, VB style)

Code:

Crit Chance = 0.05 (5%)

>RND(min,max) = Damage
>Crit = RND(0.00,1.00) (Crit yes/no)
>>if crit <= Crit Chance
>>>RND(critmin,critmax) = Crit Multiplier
>>>Display = Damage * Crit Multiplier
>>else
>Display = Damage
>>end if

-snip-

adding a graphic qualifier to crits is an issue outside of the probability distribution that crits follow. do you have any hard data saying it actually IS like your pseudocode, or just a guess? because i think a uniform distribution is more likely since about the only thing nexon gives us to help our average damage is mastery, and theyre stingy with even that. a triangle distribution has a much smaller standard deviation than a uniform distribution, so i doubt nexon would give us that benefit until i see something real saying otherwise.

Go take a bunch of critical hit damages and run it through Kolmogorov-Smirnoff test. Come on, is it ready that hard to guess/speculate that's it's multiplication of two random uniforms? If the outcome is just an modified uniform, then go ahead and test the new critical formula for me. I see a bunch of 560k+ non-crit intrepid on arms, how come I am not seeing a bunch of 700-720k crit then? Why must people want proofs when something is so obvious?

If you really want some hard data, go take down all the critical hits observed (don't use mine since chance attack isn't active 100% of the time). I can bet you if you group them in intervals, you'll see them resemble more of a triangular histogram than horizontal.

Everything is a guess, it's called reverse engineering, or what's it called, black-box testing?

I recommend testing the crit distribution using unbuffed normal attack on ordinary mobs (in a map containing just one type of mob, for preference). Then the question of Chance Attack, or how many Combo orbs were charged, doesn't come into play.

I also recommend it be done by someone with a packet sniffer, or "private server", that would make it possible to automatically and accurately keep track of all damage numbers (crit and not). Working with a video and manually copying the on-screen numbers is way too laborious and error-prone for my tastes. (Alas, I don't have either a packet sniffer or a pserver, so can't do it myself).

Go take a bunch of critical hit damages and run it through Kolmogorov-Smirnoff test. Come on, is it ready that hard to guess/speculate that's it's multiplication of two random uniforms? If the outcome is just an modified uniform, then go ahead and test the new critical formula for me. I see a bunch of 560k+ non-crit intrepid on arms, how come I am not seeing a bunch of 700-720k crit then? Why must people want proofs when something is so obvious?

unstable hero damage. i dont know, were these mechanics not in the nifty unpacked/debug files fiel got hold of?