A DPS Simulator 2 - SimFF

Quote:

---

Originally Posted by Viridiana

The level 1 critters outside Gridania still calculate the same as the dummies outside Idyllshire, though, so I kinda doubt that.

---

Shrug. Fewer variables the better. Either way yall have all this data but I think are going the wrong way about getting info from it. Work smarter not harder!

Quote:

---

Originally Posted by pandabearcat

Assuming we just want to do a simple static test, all we need are...

1. Any random combination of gear that gives a set of stats.
2. Do one known potency attack (lets do w/o damage modifiers for now).
3. Do that enough times to get an average output - alternatively, feed all into the network if we want randomness to be a part of it.
4. Repeat 1-3 with other potencies and random sets of gear.
5. As an addendum, for now separate crits and noncrits and focus on getting a good formula for noncrits.

Ideally all this data is collected on a level 50 training dummy with no other modifiers, because I am almost certain enemy level/defense modifies damage.

The neural network can be exceedingly simple - input layer with X inputs (corresponding to all stats + potency + level if desired), one hidden layer with some number, lets say X*2 nodes, and one output node being the experimental observed potency.

Really though not sure why this wasn't done years ago =P Much easier to do than trying to find a regression line in a spreadsheet - and guaranteed to be accurate.

---

I linked an excel sheet on my google drive a few days ago in this thread, which uses the exact data collection methodology you described here.
So there's your data right there... Well, the War Flash data technically has Defiance and Maim in it, but they are factored out for the potency calc.

The specific attack used for the BLM data is mostly F1 under AF3, so 180*1.8 potency.
For the WAR damage tests it was an equal distribution of Heavy Swing (150) - Skull Sunder (200) - BB (280) potency.

So now all we need is said neural network. Happen to have one laying around? =)

PS. As for randomness, wasnt it confirmed (and likely still true) in 2.X that damaging abilities have a +-5% swing to them, and healing abilities have a +-3% swing?
To my knowledge the only things that are static are Flash and dots, though dots can crit. Flash cant crit.

Quote:

---

Originally Posted by Kenji1134

PS. As for randomness, wasnt it confirmed (and likely still true) in 2.X that damaging abilities have a +-5% swing to them, and healing abilities have a +-3% swing?

---

Yeh that still is the case. a WHM I'm talking to who's doing data collection for me atm confirmed it once more.

I don't have time to tailor one...but they are so generic you can sit down with a book for 30 minutes and write one in the language of your choice.

Might (highly likely) have generic packages online as well.

If you like web coding (euuugh) you can prob just set up a gitpage with simple upload and dynamically update your model.

Again the only hard part is collecting the data and feeding it in.

The actual network is only a few (maybe a hundred) lines of code.

EDIT: good news! I've implemented macros. Just not...the macro grammar yet. Oops. Oh well, I'm feeling tired, time for a nap.

Anything I can do on my SMN to help you guys? I can't program anything - thinking of in game dummy testing in whatever situations where I give you the data. I got ACT, and plenty of time.

For SMN the biggest problem is understanding how pets work.

I need to know how much damage they do relative to the SMN's stats, their attack delay, etc.

Heh - this is gonna be interesting. In 2.x they were a mirror of you, so Garuda's 100 potency Windblade hit 25% harder than your 80 potency Ruin. But they got the 3.X SCH pet nerf treatment, so now Ruin hits harder than Windblade. Do I just got ham on a dummy and give you the data along with my stats, or is there a specific test or stats I need?

I am unsure, I don't really play around with formulas because for the sim I can change them at any time.

Likely what we want is enough samples to get a good true average for one set of stats, then change some gear for another set of stats, and repeat.

We can then use the points to figure out how pet scales.

What would be better is if we had a good understanding of baseline stat scaling, so maybe pet scaling is just some scalar to that and we can slap it on and call it a day.

of specific importance to me is how pets scale with skill/spellspeed. That determines how many attacks they get and that is more important for the sim, to correctly simulate how they respond to commands and animation locks.

Quote:

---

Originally Posted by Viridiana

Wait, it does? I must be blind. @_@

---

I don't understand the format myself, assuming it's even visible. But the patch notes say so.

Since when did Ravahn add that feature, wut.

Quote:

---

Originally Posted by SunnyHirose

I don't understand the format myself, assuming it's even visible. But the patch notes say so.

---

Huh. So they do. I'll have to ask Ravahn how to get that info.

Quote:

---

Originally Posted by Dervy

Since when did Ravahn add that feature, wut.

---

I'll send you the link on reddit, since I'm not sure how the mods would like me linking to an ACT site as opposed to just talking about it. :p

Quote from Ravahn

Quote:

---

It's in the logs, but not on the UI. It is only reported for the active player, not everyone else.
I'm away from my PC so I can't say the message type #, but it should be one of the first 10 or 20 lines in the log after you start ACT. Just record your in-game str or int, and search for a line that includes it with a bunch of other numbers. Then change some gear, reopen the log file and you should see it update.

---

EDIT: Husten/Viridiana, what would be the best way for me to normalise AA_DMG in a solver which also includes AA_DELAY into the base damage calculation (not DPS) do you guys think?

Here's a gift for you guys, before I go on vacation. It's like Christmas in July! Only its August, and it's something I promised 2 weeks ago. Lol

v0.4.5 - Download
New additions to functionality
* sim.jitter added to script. Jitter is a uniformly distributed random variation on sim duration. Use for a more accurate contribution of SKS/SPS, and sustained dps output
* macro functionality added to script. Please see below for details
NOTE: Actor construction has changed. PLEASE see the sample rotations to see how your script needs to change to accomodate.

Two main updates, sim.jitter and macros have been added to the script.

Jitter:

Code:

---

# jitter is a uniformly distributed random +- variation to duration
# this is useful in obtaining a more accurate value of SKS/SPS
# as well as randomizing duration for more accurate sustained dps picture
# you can set this to very high (less than duration however...) for a very
# wide distribution of sustained dps. The higher you set jitter, the more
# trials you should run to obtain an accurate dps parse
sim.jitter=10

---

Macro usage:

Code:

---

[p1.PRECAST]
# this is a test of the macro functionality. it is not meant to be an optimal dps rotation
# this initial precast line will instantiate a macro
(){p1.macro.state.0}

[p1.GCD]
# this is a test macro rotation
# it is designed to showcase the use of macros in a static rotation, as opposed to
# a more priority focused rotation. You may find this useful for classes with zero
# procs, such as melee classes or SMN. As you can see, for DRG simulation,
# we still have the prioritized wheeling thrust and fang and claw
# note the end of the longer rotations, as they toggle between two states
# note that you don't need to name your macro 'state'. You can name it any alpha-only word
(p1.aura.enhanced_wheeling_thrust){wheeling_thrust}
(p1.aura.sharper_fang_and_claw){fang_and_claw}
(p1.macro.state == 0){heavy_thrust, impulse_drive, disembowel, chaos_thrust, p1.macro.state.inc}
(p1.macro.state == 1){phlebotomize, true_thrust, vorpal_thrust, full_thrust, p1.macro.state.0}

---

Macros should be useful in helping users describe "rotations" rather than "priorities". I don't like them cause its a cop out, but they fit FF14's extremely rigid rotation system (so dumb...) so there you have it. Please ask if you have any questions, but otherwise, please look at the samples to see how to use these functions.

Just to give an update on the UI. I finally had some more free time to work on it the past couple days and I'll try and put the finishing touches before I go on vacation for 2 weeks. I'll have my laptop with me, so worst case i'll just finish the last pieces on it.

It will support the base functionality to be expected initially and then I intend on adding new features in the future.

Guys halp.

I understand it's a linear regression, but when doing STR vs DMG @ 0 WD, 224 DET, my regression is completely different at level 50 to 60. I understand there's a chance that level 60 most probably did increase impact of Strength on Damage... But...

Level 50: (Collected in patch 2.5)
0.0856x + 4.8968

Level 60:
0.1153x - 1.5586

Wat? Why did level 50 +4.8, but level 60 -1.5? And isn't it odd Viridiana that -1.5 was also what my Damage Model solved for to give more accuracy? My 2.5 model also solved for +4.5, lol.

They've either changed the Damage Formula completely in 3.0, or there's some great unknown with Strength. I'm still gathering more data, so I'll let you guys know soon.

Also, just for added insult to injury

Level 50: 0WD, 277 STR, 224 DET
Min: 28
Max: 30
Average: 28.5

Level 60: 0WD, 277 STR, 224 DET
Min: 29
Max: 32
Average: 29.5

I understand that this isn't practical as DET scales differently now, but even with DET being butchered by a 1/5th nearly, I'm still out hitting with base STR.

Quote:

---

Originally Posted by Dervy

I understand it's a linear regression, but when doing STR vs DMG @ 0 WD, 224 DET, my regression is completely different at level 50 to 60.

---

On a different tangent, are we certain it's 0 WD for broken weapons? The reason I don't have any broken damage on the sheets I uploaded is because there's no display for it.

I wonder if devs read this thread and chuckle to themselves when we get closer and closer :) - Keep up the good work guys.

It's 0, or could be 1.

http://puu.sh/jti0V/a0bf0e75cb.png

Then Broken Weapon Min/Max Values were

MAX: 52
MIN: 47
AVG: 49.5

I guess it kinda fits 1 a bit better.

I'm doing more testing anyway, so I can confirm it again with a much larger spread of Weapons, going up to 72 WD.

I am on vacation starting today.

What would be very helpful is if...

1. People went and tested the sim damage output numbers, and if they actually correspond to same numbers ingame.
2. People went and put in whatever hardcoded rotation, and tested it to see if it is what you would expect in game. The priority rotation right now is something like 1013 dps, which is pretty decent w/o food buff, I think it was simmed with ilvl 179 gear and hive weapon.
3. People test if they can actually make a hardcoded rotation. It should not be hard with the new macro system.

See you guys in 1 week.

Dervy I have a new parse with different gear if you need it

https://youtu.be/xpqQABzzOuw

Here's a quick one off parse I did.
http://i.imgur.com/arNvXnP.png
And my modified version of the base script. (modified rotation and my current stats)
http://pastebin.com/QWv05mM6

One thing I notice directly comparing the numbers - what did you set leg sweep cd to? there's a significant difference between the parse and sim (for reference leg sweep is 20 second cd).

Idk. I was probably not paying attention while writing the numbers down. All the more reason to dump all that data into a spell database file so people can change it when I make dumb mistakes.

I don't have the source in front of me so I cannot check.

Numbers seem off. Prob have to revisit formulas and/or check to make sure how stacking buffs works.

Relative damage contributions seem actually fairly okay, so rotation and/or general potency check is about right.

I found the neural network toolkit in Matlab. Fed it my data, trained it, got very low errors... so now how do I get a meaningful equation of out it?
It seems like all the neural network does is train itself in order to process similar data. Is there any way to get the kind of equation we're looking for out of that?

http://www.mathworks.com/matlabcentr...nction-for-gen

I didn't read this cause if I was doing it I would just feed it into the neural network every time =D but this is what you want probably.

But that kind of defeats the purpose doesn't it?
We can already make models which match the inputs to under 0.5%, but they are only good for the input range, and maybe another 3-5% outside of the input data.
Granted since it is not overly difficult to feed new data into the model, we could continue to update the model to obtain stat weights.

But the real equation would be so nice to have. =)

Not sure what the question is.

If you mean "can we somehow guarantee we will have the right formula, the one the devs use" the answer is no.

If you mean "can we get a formula that will fit the data perfectly" the answer is yes.

In reality, there is no difference between the two as long as we can test for the accuracy of inter/extrapolation.

Its sort of how you can get the taylor expansion of any function, there is a polynomial approximation with some error.

You want less error, you put in more terms (in our case, perhaps more hidden layers, and more data to train).

Thought this might be a good place to ask, but I'm curious what the relation is for DET.

Currently testing on my MCN it takes 74 Det to increase my damage by 1%. Will this always be a linear increase? i.e. 74 = 1% (despite DEX/WD Changes?)

I ask because I was looking at the relationship for all stats and both SPD and Crit have exponential gains, where as DET just starts out better.

Crit is just plain ridiculous, and I estimate by a 25% increase over the base, the effectiveness of each point will double.

SPD seems to be better than both Crit/DET out the gate, until you factor in not having an increase to AA damage, and the exponential gains are much lower than Crit.
-----------------------------------------------------------------------

My curiosity about DET is due to the way Bard now works. Since in it's DPS stance it essentially does not have AA, and has a natively high Crit chance through Straight Shot.

So if it scales well off both the exponential gaining stats then I could see it being quite the power house if you are able to stack them both high enough.
(Crit seems to plateau with DET @ about a 5% increase at which point it just increasingly gets better)

Another point I was considering is the straight shot buff, which is natively 10% X Base Multiplier of 1.45 = 4.5% damage increase.

But....say @ a 25% increase in Crit (+~1056 Crit which doesn't seem that far off) you would have 10% X 1.7% = 7% damage increase.

----------------------------

SPD example +25 = 0.01 2.49s VS 2.5s = 0.4% increase
SPD example +25 again = 2.48s VS 2.49 = 0.4016% increase

So on a class without AA, point for point SPD should start out higher than DET and further increase with each increase in reduction to the GCD.

Yeah. 73 DET = 1%. Or, 1%/73 = 0.00013, which is my DET coefficient. It's the same as me doing 1/7290, which is what I actually use for my Damage formula. So you're dead on track. It's the reason why @Arisue_Neetsha uses my 0.000137 in his new updated JP formula. He ended up with the exact same conclusion as you have with his Bard.


I've got bard weights done already. Well, the only problem is sorting out potential GCD clipping with Bloodletter once we get really, really high GCD... So I cannot accurately calculate Skillspeed, but it's higher than a Monk and Dragoon.

Off the top of my head, they were:

WD: 11
DET: 0.14
CRT: 0.223
SS: 0.135

Anyway, I've sorted out my AA formula for Dragoon now, which does take into consideration Auto-Attack Delay! And surprisingly, it's more accurate than my WD Damage model for my Dragoon!

=(((WD/3*AA_DELAY)*0.029+1)*(STR*0.1484)*(DET*0.0001472+1)*BUFFS)-1.87

Or, if I convert it to whole numbers,
=(((WD/3*AA_DELAY)/34+1)*(STR/6.8)*(DET/6795+1)*BUFFS)-1

Quote:

---

Originally Posted by Dervy

Yeah. 73 DET = 1%. Or, 1%/73 = 0.00013, which is my DET coefficient. It's the same as me doing 1/7290, which is what I actually use for my Damage formula. So you're dead on track. It's the reason why @Arisue_Neetsha uses my 0.000137 in his new updated JP formula. He ended up with the exact same conclusion as you have with his Bard.

---

Wow, thanks for your quick reply. I didn't even think to look at your blog for that sort of information and could not find the Bard Guide page anywhere. I should just set your link as a Favorite lol.

I starting looking into it because I was trying to rationalize SE's reason for not having the MCN Turret scale off SPD (+Damage or otherwise) and through that I came to the conclusion that through the psudo AA damage that the Turret does (MCN has AA outside of Gauss Barrel, but I think they wanted the Turret to be considered the AA damage for MCN while in Gauss i.e. Turret is ~20% MCN damage similar to other classes with a baseline AA % of damage)

All that just to push MCN in a different ideal gear stat priority than BRD, as that 20% difference off the top for SPD puts it so far below DET that it will never catch up.

Don't think that stuff is ony blog anyway. Well, it is. But not explained. I'll be in putting a lot of stuff there this week. Bard, Monk, White Mage... Summoner maybe, and how AAs affect weights. The bard weights are with WM on btw, because who cares about it being any other way (lol!)


And yeah. It'll be weighted lower for a MCH due to your Turrets, but.. it won't be much. I mean a Bard without WM has an SS weighting of 0.103. If anything, SS will be around 0.09 for an MCH as you only have one DoT.

How does DET/WD/DEX scale with Turrets?

Quote:

---

Originally Posted by Dervy

And yeah. It'll be weighted lower for a MCH due to your Turrets, but.. it won't be much. I mean a Bard without WM has an SS weighting of 0.103. If anything, SS will be around 0.09 for an MCH as you only have one DoT.

How does DET/WD/DEX scale with Turrets?

---

You're right that the SS weighting for MCN wont be as bad as people think, although I feel it is quite complex due to the variables associated with the Turrets.

Single Target (Rook) it is ~20% of your damage, but toss in a fight like A2S where you have on average 5 mobs for the Bishop to AOE (300 Potency VS 80 Single Target) and it can quickly jump to 30-35% of your damage further reducing the weight of SS.

That's why I can appreciate what they did with BRD as SS is pretty consistent, as like you said why "weight" it outside of WM, or as people complain plays like BLM.

-------------------------------------------------------------------------

I don't have any concrete data on Turret Damage other than it seems to scale at a slightly higher rate (only tested against AA damage) on DEX + DET VS WD compared to the Player

+ 73 DET

270.5 VS 273.5 -> ME + 3
267.5 VS 271.5 -> Rook + 4

+33 DEX

270.5 VS 283.5 ME + 13
267.5 VS 282.5 ROOK + 15

Weapon -- Hive VS Antique (+56 DET - 9 DET - 2 WD)

270 -> 280 ME + 10
260 -> 267 Rook + 7

Hopefully you can understand what I wrote. It's the average compared against the original value.

--------------------------------------------------------------------------

Something else that has been bugging me:

When looking at the weight of Crit - is it different than the other stats on classes with "Increased Action Damage" due to the crit multiplier?

Nah. Multipliers like that don't affect any weights Weights. Same goes for things like Blood for Blood and Raging Strikes.... Well, might be different for jobs without AAs.

EDIT:

Well, just did some napkin math cause I can't sleep.

With your Turret, 271.5/267.5 = *1.014953271 boost. We can round this up to 1.015, or, a 1.5% boost in damage for 73 DET.

1.5%/73 = 0.000205479

1%/0.000205479 = 48.889

1/4889 = 0.000205479

So, going back to your base damage with your Turret:

267.5*(73/4489+1) = 271.4575342

So, we can assume that for Turrets at least, Determination scales 7290/4889 = 1.49 - 1.5 times better than it does for WD. Which is hilarious, as that means DET for Turrets scale better than it does for AA's.

Of course, I need A LOT more Data Samples to test this, but we can assume that it'll be around the 0.00017 -> 0.00021 range for DET scaling with Turrets.

Quote:

---

Originally Posted by Judge_Xero

SPD example +25 = 0.01 2.49s VS 2.5s = 0.4% increase
SPD example +25 again = 2.48s VS 2.49 = 0.4016% increase

So on a class without AA, point for point SPD should start out higher than DET and further increase with each increase in reduction to the GCD.

---

You don't compare 2.48 vs 2.49. You compare 2.48 vs. 2.5 and get exactly 0.8%.

Quote:

---

Originally Posted by Hustensaft

You don't compare 2.48 vs 2.49. You compare 2.48 vs. 2.5 and get exactly 0.8%.

---

But the relative dps increase increases.

End case, the difference between 0.2 second gcd and 0.1 second gcd takes the same SKS, but doubles dps, 0.1 to 0.0 second gcd is infinite times more dps.

Using induction backwards SKS increases in value the more you get of it, contrary to "haste" in other games which typically decreases in value because they are inverse, something like (Base GCD)/(1 + haste%).

Our "haste" is very backwards and at high levels ridiculous.

EDIT: also I suppose I should get back to coding hm...maybe.

Anything I can do to help test out machinist stuff or get numbers? Really like where this is going and would love to help get more solid numbers for our stat weights and whatnot.

People could always use more testing to determine the effect of stats, so just hit stuff with different gear. Possibly need testing on the effect of stats on the turret as well.

I sort of took a break from coding because I have new hours at work, 9-6:30 now whee. I've also been playing Sky Forge (which is super amazing, btw) so my free time is limited.

Anyhoo I should have a preview of the spell database tomorrow and hopefully people can look over and correct any obvious mistakes. Ideally the spell grammar will be done by this friday and then we will be DONE with the core engine and I can just start blowing through class additions (which are fairly easy).

After that point I anticipate I will take a break, then document all the code and put it on github. At that point I think I will probably be done with it barring any glaring bugs, and will try to find some peeps to maintain the code.

How do i get this exe to run its sending me to cmd and idk what to input? Aslo I know with Simcraft they would have 1-2 ppl helping with operations per class. It would be much easier if we were in some sort of chat/IRC.

This will be useful in determining class optimization.

http://www.fflogs.com/rankings/7

That Faust ranking tho.

Quote:

---

Originally Posted by Junified

How do i get this exe to run its sending me to cmd and idk what to input? Aslo I know with Simcraft they would have 1-2 ppl helping with operations per class. It would be much easier if we were in some sort of chat/IRC.

---

You need to type in the path to the sim file you want to run.

iirc the default packaged with the sim is Samples/drg.txt

You should take a look at the second post to see what kind of options there are.

Currently it doesn't support any command line operations (like simcraft, to on the fly modify settings) because I can't figure out the Irony interpreter lol. Oh well. Make the changes in the sim file instead. If you have any specific questions about syntax or why a sim file is not parsing let me know.

I know it seems like I died or something but truth be told I'm just having extremely long hours at work right now, 10 hours a day + commute, and when I get home I have time to go to the gym and play like 1 hour of Sky Forge than I go to sleep.