A DPS Simulator 2 - SimFF

[pandabearcat]

Is this for blm only?

[Dervy]

Kenji, why don't you treat everything as a multiplier? Through my Testing I did in 2.55, DET works as a modifier and the same applies for WD. Makes it easier to type and more likely from a Developers stand point to use a formula like that.

[Kenji1134]

Kenji, why don't you treat everything as a multiplier? Through my Testing I did in 2.55, DET works as a modifier and the same applies for WD. Makes it easier to type and more likely from a Developers stand point to use a formula like that.

3 reasons.

1. It is a complete pain in the ass to weed out the individual coefficients of a series of terms multiplied together. So laziness I guess. =)

2. We have seen that in general stat weights do not seem to be static. People have done stat weight calculations with different levels of gear using multiplicative and 1st order formulas and gotten different results.
If the equation is in fact WD*STR*DET*A*B*C - Offset, which I could reasonably be from a programming standpoint, then we can say that it is in fact WD*STR*DET*Const. So if we then apply a % chance to 1 attribute, we expect the potency to change by the same %. We also expect that if the above is true, for any numerical change in a single attribute, we would see the same slope (change in pot / change in stat) regardless of whether you start naked and put on 1 piece of 190, or full 190 and take off the same piece of gear. However this is not the case. (To my knowledge)

3. Your formula in the form of (roughly) (WD/25+1)*(STR/9)*(DET/7300+1)-1, due to the +1's and -1 when expanded into a polynomial form becomes:

WD*STR*DET/1642500 + STR*DET/65700 + WD*STR/225 + STR/9 - 1

This is the same as performing the multivariable linear regression that I am using with the terms WD*STR*DET, STR*DET, WD*STR, STR, and an offset... Which I can try, using that set of terms to obtain the equation.

I have to run, but I'll put up a spreadsheet with data to demonstrate some of my points.
The "ideal" way would be to somehow have a large set of tests where we change only 1 attribute at a time, but we currently do not have a large selection of gear with which we could change secondary stats without affecting the main stat, same goes for WD and the main stat.

@pandabearcat: This formula is derived from BLM spells with the traits, damage mods, and potency amounts taken out, so it "should", I emphasize "should" work as a general formula for potency value derived from stats for any class's spells or weaponskills.
However you do have to add in the class specific modifiers, like Main and Mend, Heavy Thrust, and the potency on the tooltip of the attack/spell itself.
This is not accurate for autoattacks.

[SunnyHirose]

The "ideal" way would be to somehow have a large set of tests where we change only 1 attribute at a time, but we currently do not have a large selection of gear with which we could change secondary stats without affecting the main stat, same goes for WD and the main stat.

There is a rich selection of i70-i130 gear!

Before HW, I isolated DET at a few levels of AP and at 58 WD. I found it to be pretty dang linear. Many others have said that AP and WD are the same way in isolation but I never got around to trying it out.

[Kenji1134]

There is a rich selection of i70-i130 gear!

Before HW, I isolated DET at a few levels of AP and at 58 WD. I found it to be pretty dang linear. Many others have said that AP and WD are the same way in isolation but I never got around to trying it out.

By all means if you can run some setups to isolate stat variations then please feel free to do so.

The key things is gathering the data in a controlled manner. So ideally you'd want to use a strong ability to reduce rounding error, and always make sure that the additional effects acting on that ability are the same throughout, even if that means not using any at all.
As far as sample size, I have found that roughly 50-60 non-crit samples under the same conditions is enough to reduce the (standard deviation)/(max damage) to under 3%, which I would say is good enough. About 100 samples with larger damage numbers drives it to just about 2.5%, which I think is the minimum, since there is a natural +-5% swing on damaging abilities which (I assume) follows a Gaussian distribution.
Then the average of that data needs to be divided by the tooltip potency, as well as all passive buffs/traits, again like Heavy Thrust, Deliverance, Maim and Mend, Hot Shot, etc. This allows us to come down to a unified metric, the amount that your stats combined into a single value influence your baseline "potency".

[SunnyHirose]

The key things is gathering the data in a controlled manner. So ideally you'd want to use a strong ability to reduce rounding error, and always make sure that the additional effects acting on that ability are the same throughout, even if that means not using any at all.

I did one better: I tested at multiple potencies (sometimes with buffs) to calibrate around the rounding a little, and the tests were fairly long. A pitifully low number of data points, but they are HQ datapoints so neener neener. Sadly, they changed a few things around for HW and I ran out of retainer space, and won't have the time to do that kind of testing again in the foreseeable future, and excuse excuse excuse so here I am throwing up my hands and working on leveling things all over again instead.

since there is a natural +-5% swing on damaging abilities which (I assume) follows a Gaussian distribution.

It's uniform.

[Kenji1134]

Dervy, what's your Skype name, if you have a Skype account.
I'd like to talk to you in more detail about how to best model the potency equation.
IE. Congrats you are the new Puro!

[Dervy]

I'm on Holiday sadly and only have my mobile with me Haha, it'll be like that for three weeks sadly. Actually... on Monday I should be alright as I'll actually have WiFi. "Itsdervy" is my Skype I believe.

I uploaded the solver I used to get my formula anyway. It's nearly a carbon-copy of EMXs old sheet, but changed it up from my liking. I might be able to get a link for it...

[pandabearcat]

It doesn't matter to me really what the accuracy of the formula is (5% or whatever is fine) as long as it works for each class.

Its better to be able to roll them out then tweak later.

Plus, it will be super satisfying to post "dps rankings" and have everyone go crazy, esp if they aren't supposedly accurate yet.

[pandabearcat]

Hallo, just an update on the documentation. I ended up going through and cleaning all of the engine code up. I am -hoping- to get some documentation up by tomorrow, but that depends on my work schedule.

Now there is still going to be a revamp later on needed for Auras and Casting (mostly for BLM, to allow aura stacks and also casting resource cost modifiers). I am also planning to add 2 events, CastEvent (on a spell cast - can be instant or a casted spell) and ResourceEvent (whenever TP or MP changes) for ease of UI monitoring.

Next week there will also be a -huge- revamp to skills, moving from hardcoding to another script system that should allow for easier changes to data values (tp costs, mp costs, etc) without having to rebuild the entire program. However, until someone finds a way to pull spell info (like spell logic) from the client itself (similar to how simcraft does it) I will not make the effort to make skill logic also be scripted.