81 replies to this topic

[Loroese]

BDF2000, on 16 November 2012 - 12:53 AM, said:

If you have time could you plug in another Ghom MP5 run for me?


Mp5 Ghom
144443.3 tooltip dps
5443 dps from follower
17APOC
3.14APS
56% CC
Spec:
Shards, Snap, Chain, Venom, Wicked

46 second kill


Namely I want to see if the ticks of venom weapon had the ability to contribute to CM procs and thus inflating my dps that way.

Also where do you find how much HP these bosses have?

I'll run that in a bit. I'm running some simulations to compare to some new results I obtained with my new razer naga mouse.

As for boss hp, I think there's an option to turn on the actual values, so I can see precisely how many hp they have quite easily. I can also see it in your vid.

Edited by Loroese, 16 November 2012 - 01:00 AM.

[BDF2000]

hmm, I feel dumb now, but ty for pointing that out

Forgot to mention that I was running both Evocation and Cold blooded.

[Loroese]

Effective multiplier is 3.52 so expected dps is 509k. Actual DPS is 559k. That's about a 9.7% higher dps than simulated.

Is your dps listed including the MW buff? Really the best way to compare would be to also record a kill without the buff.

For comparison the expected kill time is 50.5s.



Side note: I still can't seem to figure out why my shocking aspect dps is too low in the sim. I'm starting to wonder if maybe shocking aspect does have a proc coefficient. I reran my tests before using a razer naga that hit 123 at the same time every 0.05s while holding down my mouse button for WW to get what should be consistant results.

Edited by Loroese, 16 November 2012 - 01:37 AM.

[BDF2000]

Mp5 Ghom
130841.79 tooltip dps
5443 dps from follower
17APOC
3.14APS
56% CC
Spec:
Shards, Snap, Chain, Venom(turned off), Wicked
Passives:
Evo, CM, Cold blooded

49 sec kill

Edited by BDF2000, 16 November 2012 - 02:14 AM.

[Loroese]

Update for those still following my progress or just interested in the model: WW+Shocking Aspect is not as straightforward as I had previously modeled. I made a post on the official forums about my shocking aspect model in the hopes that someone knew more than I did or could see an error in my methodology but it turns out it's just really strange. My thread is here: http://us.battle.net...opic/7172727350

I discovered the problem by using WW with storm armor and shocking aspect, with no other spells. I recorded a data set consisting of approximate effective dps with shocking aspect, then I removed the shocking aspect rune and noted the effective dps. By taking the difference between the two I can approximate the dps gained from shocking aspect, and with the dps info it's pretty easy to estimate how many SA procs I expect from the fight. Then I used my simulator to tell me how many WW tics I should see for the fight which is easy to use to calculate the expected SA procs. The results are that I was seeing 2.5 times the number of SA procs as expected.

This afternoon I did some more data collection to determine that coefficient as a function of attack speed and came to the conclusion that the SA procs from WW do not completely match the CM procs. I ended up with an equation that I'm putting into my code that should give a more accurate estimation of dps with SA. I'll do some simulations tonight or tomorrow and share the results here.

My data:

Coef	3.693	3.294	3.280	2.873	2.770	2.480	2.167
APS	1.512	1.932	2.01	2.28	2.390	2.690	2.850
Tics per WW	19	24	26	28	30	33	36

My equation from linnearizing the above data:

WWsacoef = - 0.8737 * AS + 4.8035

Coef is the multiplier representing the excess SA procs, so if my attack speed is 1.512 and I expect to see 100 procs over a given fight based on the LoH tic rate, I would actually see 369.3, on average.

Notice how the Coef actually increases as attack speed decreases.


As far as the rest of my model goes, I'm hoping to finish up a few last edits/additions over this weekend. I want to add options for a couple FN runes and update the meteor and MB sections.

[maka]

Great work, and weird results indeed.
I don't use Shocking Aspect, but this is really erratic behaviour.

[Loroese]

I updated the second post to include these results, but since most of you are probably going to look here first, I figured I'd copy and paste them here also. I'll be updating the original post sometime in the next couple days when I finish up my editing of the m file.

Here are some simulation results combined with in game data to show how well my simulator is working. For each case I used the skills indicated in the tables. For the case of pinpoint barrier, I added 5 CC to the base stats. I used 114 max AP for Storm Armor and 94 without. I did a total of 3 case studies with varying overall attack speed. I used the same gear and just swapped between a 1.2, 1.5, and 1.77 APS weapon with 52% IAS on gear.

Base Stats:
APoC=19;
CC=0.505;
MaxAP=114; (94 with prismatic or pinpoint barrier)
APregen=10;
DSprism=0;
DSds=0;
Evocation=1;
CB=1;
Conduit=0;
SimDuration=VARIED;
Mobs=1;
AS=VARIED;
MobAttackRate=1;
DSabsorbs=1;
MobAttackDelay=MobAttackRate;
Simulations=300;
WWuptime=1;

ReactionTime=0.01;
Case 1: AS = 2.6904

Storm Armor	Shocking Aspect	Pinpoint	Shards	DPS Mult	Rel DPS	Simuated dps	Rel Simulated DPS
				3.01	1	3.09	1.00
		x	x	3.53	1.17	3.57	1.16
			x	3.42	1.14	3.43	1.11
		x		3.04	1.01	3.19	1.03
x			x	3.82	1.27	3.84	1.24
x				3.47	1.15	3.49	1.13
x	x			4.60	1.53	4.69	1.52
x	x		x	4.85	1.61	5.02	1.62

Case 2: AS = 2.28

Storm Armor	Shocking Aspect	Pinpoint	Shards	DPS Mult	Rel DPS	Simuated dps	Rel Simulated DPS
				2.95	1	2.83	1.00
		x	x	3.48	1.18	3.42	1.21
			x	3.12	1.06	3.14	1.11
		x		3.04	1.03	3.1	1.1
x			x	3.58	1.21	3.59	1.27
x				3.20	1.08	3.29	1.16
x	x			4.20	1.42	4.43	1.57
x	x		x	4.56	1.54	4.74	1.67

Case 3: AS = 1.824

Storm Armor	Shocking Aspect	Pinpoint	Shards	DPS Mult	Rel DPS	Simuated dps	Rel Simulated DPS
				2.69	1	2.61	1.00
		x	x	3.25	1.21	3.10	1.19
			x	2.90	1.08	2.88	1.10
		x		2.74	1.02	2.81	1.08
x			x	3.72	1.38	3.43	1.31
x				3.05	1.13	3.17	1.21
x	x			4.02	1.49	4.13	1.58
x	x		x	4.17	1.55	4.41	1.69

The dps columns are as follows:
DPS Mult = the number you multiply your char sheet dps by to get your effective dps
Rel DPS = relative fractional dps gain from the skill sets, compared to the all defensive build
Simulated dps = DPS multiplier obtained from the simulator. This should compare directly to the DPS Mult value
Rel Simulated DPS = relative fractional dps gain from the skill sets estimated by the simulator


As you can see the simulator has good agreement with the in game testing. It does appear to overestimate the shocking aspect contribution, which can be fixed by tweaking the empirical formula used. The Storm Armor results seem to not agree as well for Case 2 and 3 as Case 1, but I think that was mostly just RNG since the Storm Armor strikes can be pretty sporadic.

For reference, the empirical formula used is
WWsacoef = - 0.8737 * AS + 4.8035
where WWsacoef is the adjusted number of SA procs from WW per CM proc. In other words, if a tic from WW would proc CM, it also procs SA (WWascoef) times. This is only used for WW. CR, DS, and any other skills are all treated the same as CM at a 1:1 ratio.

[Loroese]

As promised, I have updated the mfile so that the meteor and MB simulators should be working effectively. I'm calling the current version 1.00 the release version. I also updated the original post to reflect this and I will post updates here as I make them, as well as keep an update 'log' in the first post and in the mfile with a version header at the top of the file for easy reference.

Also, if anyone is fluent in programming languages, you are welcome to use my code to covert the simulator into something more easily used. I haven't used C++ in years so I don't feel like trying to convert it into another file type, and I use MATLAB a lot so it works for me. I'm also going to post this on the official forums so if anything interesting pops up there I'll be sure to keep this thread updated.

[Bagstone]

I have no idea about Matlab, but if there was some documentation in the code (or at least how this works - most of the code looked like 20 year old Pascal code to me) I could think of re-writing this. I'd probably create a website rather than a C++ application though - if possible. Don't really know what the requirements are, but I can't imagine that the calculation is soooo extremely complex. Could do C++ or an Android app in Java or whatever, but that would require more time than a quick'n'dirty PHP thing.

This is really looking great and I want to use this, so I also have a personal interest here; don't know when I would have time to do that, though. Let me know if you can provide me with a very brief overview on how to read Matlab code, and I might look into it if I find the time and no one else is interested.

[Loroese]

@ Bagstone: The the language is mostly C++ as far as I can tell. There's just a bunch of IF statements for events centered in a WHILE loop that increases time every iteration until the sim duration is complete. I'd be more than happy to help you more if needed.


I had a crazy idea to finally make a few plots of how results scale with your base stats. What I ended up doing was starting with my base stats (49% crit, 20 APoC, 2.5308 APS) and changed one variable by a value Delta. I then repeated it for each variable. Delta = 2% CC, 1 APoC, or 0.1 APS so 7 delta means 14% CC, 7 APoC, or 0.7 APS more than the stats listed above. I then plotted the Effective DPS multiplier, FN uptime, and LoH return coef.

My build was FN - Deep Freeze, Diamond Shards, Shocking aspect, Cold blooded, but no evocation or slow time.

You can clearly see the WW breakpoints in the LoH gain per second and the FN uptime. The Effective DPS Multiplier also shows the WW breakpoints in the form of decreasing dps after the breakpoint, then a jump increase at the next breakpoint, etc. This is because increasing your attack speed while between breakpoints will increase your char sheet dps but will not lead to an increase in build dps, therefore the char sheet dps increase is misleading between breakpoints. I would guess that your effective dps would remain the same.

It's also interesting that effective dps actually seems to decrease as you raise Attack Speed past the 2.5 breakpoint, which is unexpected. As you increase Crit, it continues to increase, but APoC only has a very weak impact after the 2.5 APS breakpoint, likely because at that point you gain AP about as fast as you can spend it.

[BDF2000]

Does this simulator support simulations for Magic Weapon: Electrify on an arbitrary amount of targets(like 2 ubers)?  Just saw a post about how it does aoe dmg like crazy over on the official forums and it would probably be cool to have it be another subject for characterization of how it works.

Thread: http://us.battle.net...opic/7199030105

Edited by BDF2000, 20 November 2012 - 03:58 PM.

[Loroese]

I saw the thread, but so far it does not include electrify. I can make a rough estimate of dps gain though:

If you have 30 tics per WW per target, that's 3.75 electrify procs per WW per target, or an extra 37.5% weapon damage per target. WW does 252%, so that's about 14% damage gain. CR will  only proc 1/3 casts so it's mostly neglegible.

If your WW does 35% of your dps, electrify will add about 0.14*0.35*100 = 5% extra dps against uber bosses. So it seems against ubers it's not much different than the extra 5% from Force Weapon. It's probably more useful for higher level MP key farming, or trying to get a high XP high MP run.

I think that means the damage is +10% with 3 targets and +15% for 4+. I haven't tried it myself yet though, and from the post it seems like people think it does more than that. I probably missed something when considering 3+ targets but from a per target standpoint it makes sense to me that it would just be a linnear growth to the base WW damage. Granted, each WW hits multiple targets, but that doesn't change how it behaves against each target.

[apo86]

As far as I understood:
If we take your example on 4 targets, we have 30*4=120 ticks per WW.
That means 15 electrify procs per WW all of which will hit 3 targets.
That's 450% damage compared to 252%*4 = 1008% damage for the WW hitting 4 targets.
And suddenly it's a 44% increase :-)

Now if you factor in that most of the people discussing this stuff are above the 2.73 aps breakpoint, it's 144 ticks, 18 procs, 540% damage, 53% increase.

And I guess it also still gives the base 10% damage increase.

[Loroese]

@apo86, you didn't factor in the % of your damage that is done by WW. If WW only makes up 1/3rd of your overall dps, then you only see a ~15% dps increase, which is consistant with my math.

I posted some maths in the post on the official forums, and my conclusion is you would see about 20-25% dps increase from the combination of MW+Electrify, including the bonus from MW. This is taking into account that WW only makes up about 18% of your total dps against 4 targets (rough estimate though since I haven't messed with multiple mobs in my simulator much) compared to the 30-35% single target. Adding in the 10% from MW translates into about 19% DPS gain from the combination of MW+Electrify. This still neglects CR procs, which likely have a decent impact on the results against multiple targets since you get more casts off. Against 4 mobs CR makes up almost 50% of the simulated dps.

[BDF2000]

Why are you presenting the dmg done by electrify as a scalar of WW damage?  Why not just say it does X% of your tooltip dps extra at each breakpoint?

[apo86]

You are right. I skipped the last step and missed that your example was already correctly calculated.
~20-25% overal damage increase still doesn't sound so bad. Well, on the other hand, Stretch Time doesn't depend on how many targets are available... Wizards need more skill slots

[Loroese]

BDF2000, on 20 November 2012 - 10:42 PM, said:

Why are you presenting the dmg done by electrify as a scalar of WW damage?  Why not just say it does X% of your tooltip dps extra at each breakpoint?

It's just the first idea that popped into my head to estimate its dps. It's hard to put an exact % on it because different stats might change the WW contribution, so I was mostly generalizing based on reasonable stats.

[maka]

Loroese, thanks for the update. Is Slow Time included in this version?

EDIT: checked it out. It is included

Great work!

Edited by maka, 21 November 2012 - 05:47 AM.

[SeravokX]

Looking at your plots, the simulated loss of DPS as APS increases is troubling to me.

The only thing I can think of that would be causing that is your order of priorities of what spell to cast next for your simulation, or your simulated reaction time, or both. Could it be possible that as your APS increases that you're resetting DS and FN very, very fast, and so your simulation is casting them so often you're actually casting WW and CR less often?

On another note, wow you've changed a lot since you first posted this! The last I remember, this post was about comparing WW to Mistral Breeze and Meteor runes. I'm glad I took the time to re-read this. Very nice work.

Edited by SeravokX, 11 January 2013 - 09:33 AM.

[Loroese]

SeravokX, on 11 January 2013 - 09:32 AM, said:

Looking at your plots, the simulated loss of DPS as APS increases is troubling to me.

The only thing I can think of that would be causing that is your order of priorities of what spell to cast next for your simulation, or your simulated reaction time, or both. Could it be possible that as your APS increases that you're resetting DS and FN very, very fast, and so your simulation is casting them so often you're actually casting WW and CR less often?

On another note, wow you've changed a lot since you first posted this! The last I remember, this post was about comparing WW to Mistral Breeze and Meteor runes. I'm glad I took the time to re-read this. Very nice work.

Yeah, the initial release was mostly comparing different AP spenders. It didn't take long to figure out WW beats out the others very quickly as you raise your attack speed.

Regarding the DPS vs APS, there's 2 issues going on. Just to be clear, DS doesn't use an attack animation, only WW and FN do, so while higher APS means you can refresh your DS and CR faster, they do not take up attack animations so you can cast both of them at the same time as a WW for example.

The first issue is my SA model is incomplete. SA has a very strange interaction with WW and I don't know the exact mechanics of it. I initially thought it would just be a basic relationship like the LoH return. Instead, it seems to be procing from 2-3 times more often than it should, and the actual increase in observed SA procs seems to be effected by APS. My guess is I'm underestimating the impact of SA at higher APS, but I don't really have the time to get more data to try and improve my SA model.

The second issue is latency, which I call ReactionTime. That definitely impacts effective dps and just changing it for cases with the same stats besides ReactionTime lead to different dps multipliers. I think for the plots I made I used 100 ms because that's approximately what my latency is here in EST time zone.

The general thoughts behind the results is that at 2.5 I'm AP saturated so my WW dps won't increase as I increase APS. Because of the latency, the reduced CD of DS and CR from the extra WW tics has a negligible impact, so the DPS becomes mostly saturated. Add in the SA model and it results in the same or lower estimated dps at higher APS. In practice though, most people are seeing higher effective dps multipliers at the higher breakpoints, which again points to the SA model needing an update.

Edited by Loroese, 11 January 2013 - 09:58 AM.