A DPS Simulator 2 - SimFF

[Dervy]

Pretty sure the scaling has completely changed Kenji. None of my old 2.0 models work with level 60.

Also my new model is much more streamlined where there's only 3 values you multiply/divide

[Kenji1134]

Pretty sure the scaling has completely changed Kenji. None of my old 2.0 models work with level 60.

Also my new model is much more streamlined where there's only 3 values you multiply/divide

The reason the model is so "big" was that it provided the least total and average error for the 80-odd point dataset. (Int, Det, WD, average B3 damage over 200 casts)
I did try further reducing the model, but it increased the total error too much... I think the "original" model was... Lets see, A = Int, B = (Det-202), C = WD...

A^2+B^2+C^2+AB+AC+BC+A+B+C+Offset = 10 terms... I did once try (for the hell of it) raising it to a cubic level equation, but after the first modeling iteration the coefficients for all the cubic terms were soooooooooo tiny that I just dropped them outright and reduced the equation order.

But yeah, odds are the scaling is different now, at least in some way.
Im about halfway through 59 I think, need to more gear to test with and I can start cranking away at it next week.

[Dervy]

Nice, because Black Mages need another "lead theorycrafter" anyway seeing as Puro has disappeared.

Do watch out though that 100 Potency isn't exactly 100 potency. I personally use Potency*0.01005, then -1 at the end of the entire calculation ( after buffs and all that) and it seems to give me less errors.

Hopefully you can confirm or deny my theory with BLMs juicy 4000 hits haha!

[Kenji1134]

Puro was asked to Monk by his static some time ago, and decided to stay monk.
He tried to hand his spreadsheet over to me back in like Feb, but I was already neck deep in masters thesis work. There's a few good blms who have kept the ship afloat, contributing insight, rotations, tips and tricks, and the like.
Personally I main a Warrior in raids (tanking since 1997), been doing that since T6, and playing BLM in T8 and T9. But as a mathematically inclined Male... I like big numbers and EXPLOSIONS!!! so I enjoy BLM and the engineering side of me just wants to make it as efficient as possible... So here we are.

Still, its nice not having anyone call me their "math butt-buddy" anymore. =P

[HeavensSword]

Nope, the code is currently made of duct tape and children's nightmares. You raise a really good point though, I think I will spend the next few days writing up documentation so people can write their own UIs.

Would be much appreciated.

I'm setting up a project now and including your libraries, will just wait patiently for how to use your API ( apply duct tape and invoke nightmares ).

PS: you look like deathwing

Hah, that's what I thought of when I first put that head piece on. Not wearing it anymore, but the forum avatar is slow like that.

[HeavensSword]

Just want to bump this thread.

[Waliel]

Hopefully you can confirm or deny my theory with BLMs juicy 4000 hits haha!

I think I average around 3700 with 100 WD, 907 INT and 443 DET :<

[pandabearcat]

Man commenting sucks. I can't remember the design choices behind half this code. Its sort of in limbo right now, a lot of it is pretty inefficient...some of it I need to overhaul before other classes (namely BLM, with its varying mana costs) can be implemented, and I need to sit down and figure out how the script and the engine is going to be able to handle multiple actors and targets.

The engine itself can easily support multiple actors but trying to figure out how that logic works in the script is going to be hard. I'm trying to go through my dps priorities and my motivations for AoEing and multidotting, etc, and figure out how to write those rotations.

[Kenji1134]

Ugh... collecting data is the shitty part.
7:21 parse to get a total of ~70 F4's. Then toss out crits, 64 left.
That brings the standard deviation divided by the max value down to about 2.5%, which should be sufficient given that the game naturally puts a 5% swing on damaging abilities.
Then about 7 test sets per weapon, starting from fully geared, taking off 1-2 pieces of armor/acc's to create controlled changes in Int and Det until naked... and repeat for different weapons. Gonna go with 97 WD, 83 WD, and 38 WD for the time being.
So 3 weapons * 7 tests * 7 minutes of spamming per test = ...147 minutes = ~2.5 hours... well shit...

Granted spamming B3 like I did the first time would be a lot easier and faster... But B3 is 240 pot, while F4 is 529 pot, a difference of 2.2x in terms of reducing rounding error, which will be important as my stats go down.
So, if anyone wants to help, then I need the following data:
Int, Weapon Damage, Det, and the average damage of F4 over about 60 non-crit casts. That tends to be enough data points to be accurate.

PS. The base value for Determination at 60 is 218.

PPS. Okay, so I finished 1 full set for the 97 WD weapon... This takes way too damn long... Decided to try using F1 instead of F4. 50 data points in 4 minutes instead of ~7, much easier rotation. 61.2% of the damage so there is a slight loss in accuracy at the lowest stat levels, but when naked the difference in final calculated potency between using F1 and F4 was 0.55%. I think that is within an acceptable error range given the amount of time saved.

So if anyone's got say full 180 w/ Hive weapon and doesnt mind collecting some data, the average of ~50-60 noncrit F1's, along with your Int, WD, and Det would be great.

[Kenji1134]

Alright, that was... a fair bit of time and spamming... anyway, this is likely not "terribly" accurate since it is based on only 18 data points, and I tried various stat combinations to make it somewhat logical, with zero intercept and all positive coefficients.
I would like data from someone with "top end" gear, since the model is only as good as the highest valued data points available.
So here is just something to consider. This model has an average error of 0.382% across the 18 data points, with a max of 0.98% error.

Raw Unmodified Potency = WD*Int/21748.8 + WD*(Det-218)/313453 + Int/908.779 + (Det-218)/3237.46

Not "THE" most accurate model setup, and again it could use more data points to increase the accuracy and effective range... but its worth a try.