yea probably "9" instead of "7" been quite some time.... also you don't seem to understand the point of my post :S
yea probably "9" instead of "7" been quite some time.... also you don't seem to understand the point of my post :S
Your implication seemed to be that:
Not being able to reproduce those missing '4's and '9's within a system would invalidate the system. I instead showed that this is no effect of pDif, but instead that Type Weaknesses were applied after final damage and then re-floored, not to mention I corrected your own anecdote pointing to this process.Originally Posted by Masamune
Even with you feeding me false data it was easy to solve. Not sure what other point you were trying to make besides "You're not clever enough to have figured out what I havent"...
No, he means that you can't make a model that generates gaps that don't show up in the data because parses of pDIF without piercing bonus are continuous. He compared it to the piercing bonus parses so you would know what he was talking about.
As general tips for producing clean data, his post was relevant.
Posting some semi-false anecdote about holes in his data as some grand mystery waiting to be solved seemed like an undermining of my method at first... So I solved it. Yes, I'm full of myself, but I wasted about 30 minutes plugging in slightly tweaked Piercing Bonuses in a double flooring system trying to find where both 4s and 7s disappeared instead of 4s and 9s before I finally went out and tested for myself.
Masa has pointed the piercing bonus thing in the past (literally years ago) and has said it is the only step that he is sure damage is floored before. You just didn't understand what he was saying.
I will toss on that I strongly suspect damage is floored before critical attack bonus is applied. It would be testable quickly with a Ranger WS and some ingenuity, but I am too lazy and sleepy.
I'll fire up that double flooring setup again and see if I can find a minimum crit bonus to give a missing integer...
Yeah. Even just 5% will produce missing digits. Not in easy-to-spot multiples of 5 like 0.25 does, but you'll skip 104 damage for example. 99 gives 103, 100 gives 105. Hell, any crit bonus at all will skip a number shortly after a multiple of 100. I can do this with DRK's CAB.
Got all the way to Wajoam Woodlands before I realized I had a ton of DRK/DNC data already XD
DRK keeps it's 8% CAB at 99 apparently. 237*1.08=255, 238*1.08=257. CAB is applied after final damage flooring too.Code:Melee Crits 234: 3 235: 1 236: 2 237: 5 238: 4 239: 6 240: 3 241: 8 243: 2 244: 9 245: 3 246: 29 + 247: 33 248: 22 ^ 249: 28 250: 32 251: 19 252: 25 253: 18 254: 8 255: 17 257: 16 258: 18
Thanks Byrth.
In tech terms, i meant this to prove/disprove a given formula with:
- DisplayedDMG = Floor(Floor(FInalDMGwithoutpiercingbonus)*PiercBon us)
or
- DisplayedDMG = Floor(FInalDMGwithoutpiercingbonus*PiercBonus)
1st case would be kinda hard to produce holes from a continuous formula producing only integers, 2nd case on other hand... Remember Raellia our convo @ otherwiki about number of flooring(s) SE might be using before displaying ? (i think there is only one)
That means can pinpoint even more closely the value for yur parameters RAE & LIA using this.... or change whole formula :S
Rolled up. Lots of new joyous goodies in the OP. I've got the 1.0 floor and ceiling in the same function and properly spaced too.
And now I'm gonna ragequit because of crits, particularly minimum crits but their overall slope below 1.5 cRatio too. I need to hand-pick some data to work with, and need some sub-1.0 cRatio data to work with too. I just can't get their range narrow enough without going back and redefining RAE and LIA for them (which is very, very likely.)
Raellia, there is something i don't understand:
I tried to see what kind of distribution your formula generates with simplifying it :
DMG = Floor[ BaseDMG * [ ((11/10)*cRatio-0.65 + rand1*0.85) * (1+rand2*5%) ] ]
i run this through 2k sample size and BaseDMG=59 and visualized generated distribution at different cRatios:
i don't see any spike at any cRatio o.O (and this is fundamental as those frequency spikes directly impacts averages)
Did i make a mistake somewhere in the formula ?
EDIT: for your crits generation problem, just change your primary random with :
Randx -> IF(RandZ < CritRate%; 1.2; 0) + Randx
See this is the vagueness I was talking about. I don't know about any frequency spike you're referring to except the spike at 1.0 because all I have is your summarized data. At what cRatio?
If you're looking for the 1.0 spike you have to use the 0.75-1.0 rule in the OP. This generates the 1.0 spike from 0.5 to 1.5 cRatio and requires a conditional. If you're talking about those little frequency spikes in crits (see my graph of the soboro crit data) I've got an idea for how Secondary is distributed that would solve that.
That simplified form is definitely elegant...
wow if yu don't get what i meant with frequency spikeS....
Look more closely at frequency distributions i made on last sheet on my googlesite and scroll down fast from topest graph to lowest and viceversa so you see how spikes behaves and when THEY appears, because there is not 1 but 4...
That's better, but I still can't see your graphs because not all of us can be assed to pirate use Excel and Google Docs doesn't render them.
And last time I tried to find a standalone version of Excel to view your spreadsheet it wasn't a new enough version.
Are they independent of 1-hander vs 2-hander? Are they in the same places irrelevant of ratio or do they scale with it?
As I said in the OP, distribution wasn't part of the plan at first, I just left it 'clean and even' while determining RAE and LIA.
ok my bad then indeed i didnot thought about readibiltiy compatibilty...
Ill show them in yur own distrib graphs after my VWs...
And yes I know Averages are off now. They were pretty gewd with static LIA, but now they lowball consistently. Moten pointed out that Primary was behaving in parses 'just as much up as down' or equally above 1.0 as below despite being imbalanced in spread, which would make precisely the small change needed to be hitting parsed averages.
I can get your data out of it just fine, I'll just have to generate my own histograms (and with something better than that website...) but after I work on getting crits to fit I'm gonna take the rest of the day off from this. A solid week of it has implored to me just how too much free time I have XD
Here are the spikes.
As for how they can screw averages, let's take for example cRatio=1.4 distribution should look like this*.
From this you can easily guess average dmg won't be in middle of the normal dmg values, but skewed toward the spike on the left.
Opposite skew happens when max pdif is capped.
EDIT: * were simulated with BaseDMG=50 and critrate=30%
There is also another signature about the behavior of those spikes:
If we consider for example the cap @ pDIF=3, its spike is 1st hardly noticeable at cRatio~1.65, then it gets more and more tougher in frequency as cRatio increases.
The skew of above vs. below I can agree with. It's what Moten pointed out.
The spike on the far left is apparently the 1.0 spike, because it should be appearing at 1.4 cRatio. It even has the 5% width so I can only guess you have 50 base damage because you didn't provide.
Hurpdurp, the upper range is crits. I was thinking the gap between 89 and 96 damage was something else.
If you're talking about the overall jaggedness, I don't see a damned thing wrong with it. Your distribution is going to be noisy in any sane parse but I have some ideas on how Secondary works that may produce predictable patterns of short spikes like that.
it's just a basic simulation, in my mind, the correct modelisation order is :
1. find a formula replicating ALL signatures
and only after this done :
2. Fine tune parameters to match parsed distribs
Yes it's the spike @ pdif=1, and the hole is simply what separates normal dmg from crits...
Masa, I updated the first post all the way through yesterday. not just added the graph at the end. I think you need to go read it again and find the part about the conditional on pDifs below 1.0 and 0.75. It gives the 1.0 spike perfectly even as pDif continues below it. Graph.exe just can't handle rands inside conditionals properly to show it, so can only show the spike developing above and below the 'crossover' threshold, but you can see it in my graph. You can see the 3.0 crit spike developing from 1.6363~ upward (yes, I know the exact number where this begins in my model).
1-hander data doesn't even reach the range you pointed out in the normal hits, but I can see some indication of an effect around 2.5 pDif in my own data (note this is at 2-hander's absolute maximum).
You're being hyper-paranoid about things I already show in my model.
Here I've changed Primary for crits from:
((11/10)*x-0.65+rand*0.85)
to:
((11/10)*x+0.45+rand*0.75)
This is effectively a 1.0 bonus inside Primary with some shortening of Primary's range that looks like a 1.0 bonus at common cRatios
Looks pretty damn good.
Is not a matter of being paranoid or anything similar XD, it's just im trying to reproduce yur results with what i have (Excel) and compare with parses, nothing else