Simulation accuracy: AMR matches real combat logs


#1

Hey everyone, I am always comparing our simulations to combat logs. The idea is we want to make sure they reflect actual fights, because when they do, you can get high-quality, actionable advice from the results.

I figured it might be of interest to other people, so I shared examples for each DPS spec, so you can see for yourself how well the sims match logs.

Here’s the blog post - and feel free to reply to this thread with any questions.


#2

That’s got to make you guys feel pretty good.
Thanks for all the effort you put in, for so little money too!

You might want to read #2 Enhancement shaman again, the given numbers don’t look close at all. :wink:
I suspect you pressed 8 & 0 in the wrong order.


#3

Good catch, I updated it :slight_smile: Yeah, it was pretty cool to see log, after log, how closely they matched. The one that threw me was the Fury Warrior, and then we realized they are jumping out early and the script didn’t have that. So, that’s how the “Fury Special” script came about, ha!


#4

Jumping out late. :wink:
I think you might need more sleep!
It’s been a bit hectic for you all the last week or so.


#5

I am not seeing that AMR is matching my simulations. Its actually pretty far off. From two points of view:

  1. Simulations against Krosus using your tool parse out at about 660K dps in the AMR recommended set (best in bags feature) and 654K in the set that I feel is the best - about a 7 ilvl difference AMR = 902, my personal best estimated set = 895. Now when I fight krosus on heroic difficulty (using my 895 set) - a fight lasting 3 mins and 30 seconds - I am doing 815K dps. See link here: https://www.warcraftlogs.com/reports/jy7Zd6hY8PGVBHTv#fight=24 (my toon is named Femfatale)
    So that is approx. 25% inaccuracy.
  2. When running my own simulations against the Iron Forge Raid Target Dummy I am parsing at about 710K dps over a 3 minute and 30 second span unbuffed in my ilvl 895 set. However, in the AMR recommended set I am parsing at about 650K dps. So the AMR recommended set is just under 10% less effective.
    Now this is an average of about 20 tests in each set all at 3 mins and 30 seconds.

So long story short - I feel the AMR Sim tool is not working well at all for Havoc Demon Hunters (im in the standard Felblade/Nemesis/CB build btw) And the Best in bags feature does not seem to be properly weighting secondary stats. I believe that it is weighing Agility over Crit and Mastery.

** one thing of note - This is math was done prior to my aquiring the 10% chance to extra hit on chaos strike


#6

I would like to make a request as a component of my post above - it would be great if you could post how you are weighting each stat. Both Primary and Secondary - in addition to this I would like to further request Tertiary stats be weighted. This would be extremely valuable for tanks I believe…
for example - I have an ilvl 865 tank set for my Vengeance Spec. I also have a ilvl 897 tank set for my Vengeance Spec. The ilvl 897 set has 2 legendaries and 4 pieces of 890 and 900 of heroic nighthold gear. Some would argue that for me to even consider the ilvl 865 set with no legendaries and no set bonuses over the 897 set would go a long way to proving that I am a terrible player… however the 865 set is comprised of 14 of 14 pieces having Leech on them and a large amount of Versatility. Versatility in this set is approx 6500 and leech is also 6500. This give me nearly 30% constant leech (augmented by the +healing that I receive from Versatility). This is massive for a Vengeance DH. I would be better able to properly evaluate the two sets if I had a weight stat for Primary, Seconday and Tertiary stats.


#7

As a side note to the request above. I realize the math example below(keep in mind these are made up numbers, but hopefully provide a helpful analogous)

Lets say this were true:
1 point of crit = 10 more dps
So if I went from 10000 crit (and that yielded 700K dps) to 10100 crit (with no other stats changing), I would expect my dps to go from 700K to 701K dps.
HOWEVER (and here is my point with this post) in reality - going from 10000 to 10100 is not the same as going from 5000 to 5100. That additional 100 points at 5000 crit rating is cannibalizing less than it would at 10100.
But I dont know what that math REALLY is, it would be nice to know.


#8

First, it sounds like you aren’t comparing apples to apples. To start, you want to compare exactly what you had in that log. Click the ‘import from log’ option on the simulator and link to the damage view for your character.

Then you need to pick the following options:

  1. The DW “No Miss” script that I talked about in that blog post. That’s because DW doesn’t miss on Krosus (an assumed in-game bug).

  2. We have the new feast going in soon, but until then, select ‘no food’ in the options. Then on the left, under your gear, override your agility to add 500. So it would be 17915.

  3. Select potion of the old war, check the augment rune box, and set the fight length to 219.

When I do those things to match your logs I get 742k DPS - about 90k higher than the ones you got.

Now the trick is to comb through and see what you did differently in the logs. Did you get lucky crit streaks? Lucky trinket procs and uptimes? Find those differences, eliminate them, and then we can see if anything doesn’t match :slight_smile:


#9

I looked at your log you posted compared to the simulation result. Here is the simulation I ran:
http://www.askmrrobot.com/wow/simulator/report/bdaa23eb1ba4483a8a9735349f149c4e

The first thing to consider for Havoc is that you will get highly variable results compared to other specs. Some reasons for this are:
1.) Large damage range on almost all your damage because everything is based off weapon damage.
2.) Getting more or less crits for CS/A will create a cascading effect because of the fury gain.
3.) Demon Blade proc chance is not an rppm, so can vary wildly from fight to fight.

Looking at your log, you were able to do 3 more Chaos Strikes than average for your fight length (this can easily account for 10, 15k DPS or more difference). Also, the damage of your Annihilations was on the high end, considering you got less crits than average for those. Your Jacin’s Ruse uptime was much higher than average, which also accounts for the higher damage/hit on your chaos damage.

Your DPS on the fight was 816k. The Simulation was 750k You fall well within 2 standard deviations of the simulated DPS. Considering you had a top, top end parse - this is really not outside of an expected result. I would be worried if a parse in the 99th percentile was even with the average simulated result.


#10

Do you want to re-post the other questions in their own threads? Then I will delete this post to keep this thread clear for combat-log-comparison questions only :slight_smile:


#11

well its your site, if you want it placed elsewhere - go for it. However, this location seems correct to me. I am still in disagreement with your logic. I would like to try again this week on heroic Krosus and see if a similar result presents itself. I bet that I am once again north of 20% variance from your simulation (as I have been nearly every time I have done the fight as close to the sim as realistically possible). was there a “miss” component to Kromog do you know?


#12

also - zoopercat - you said that you got 90K higher than my original post. is that using the set that I have now? with the artifact talents that I have. my test against the target dummy show that Chaotic Onslaught has yielded me an average of 70K additional dps. Now I have only had time to do about 10 tests and the range has been 50 to 90K. However, I would expect it to be higher right now by about the number that you listed in your response with just Chaotic Onslaught alone.


#13

In your 10 tests, what were you actual DPS values for each trial? With the actual values, we can calculate the standard deviation to verify whether we can draw a statistically meaningful conclusion. (There are well-known statistical formulas for it.)

Note that it’s a little tricky still… because there are more variables in-game than in a simulator, e.g. maybe your fight length timing was not perfect (one trial was a couple seconds longer than another), or you made a mistake in one trial and missed an ability. We can probably just gloss over that and say that any such variations are “random” (even though they might not actually be random, e.g. you might tend to make the same small mistake at the same time, or you might tend to collect data for a few seconds too long, but never too short.). But I think we could still get a good enough idea if we have all of your data.


#14

Can you tell me what you disagree with? Both Swol and I used the exact setup from your log (exact gear, talents, artifact path, etc). We do that so we can compare apples to apples (and you can do that with the ‘import from log’ feature). Then as swol pointed out, you got a lot of lucky procs, bringing your DPS up well above average.

To take a step back: simulation run thousands of iterations, so it gets some iterations with a ton of luck, some with bad luck, and some in the middle. It averages those out. So when you get lucky streaks, you SHOULD beat a simulation.

For Chaotic Onslaught, you can sim with it, and then without it, and you’ll see the DPS difference is large, as you suspected. So that is working right (if that was one of your questions).

And as yellow said, send over data and we will take a look.


#15

So, I looked at this a bit more, since I’ve been doing a lot of work on Havoc lately as far as verifying the simulator’s accuracy, and I’m 99% sure that we have all Havoc abilities simulating correctly.

The problem was your 850 trinket didn’t import right when I ran the first simulations, so the new one is this:
http://www.askmrrobot.com/wow/simulator/report/9c5e98e203a14cc581ce9abba12fd10d

800k, with a 45k std. dev. So, everything looks good. With those lucky procs and the extra crits on Chaos Strike over normal, it makes sense you’d be a couple % higher than the simulation.


#16

ok Swol thank you. That last Sim you posted is quite helpful. I will save it as I go through this week. thank you guys.


#17

Zoopercat - I dont know how to “select the DW no miss script” I dont see it as an option - and when I click “view/edit” I dont see how to properly replace whats there.

also… what does DW stand for?


#18

It stands for Dual Wield. The “no miss” script isn’t in the public scripts. From that result I linked, you can click on the script at the top and then it’ll take you to the script. Use the “copy” button in the top/middle of the screen to copy it to your list of scripts. Then once you reload the simulate page you will have access to it.

Our assumption is that this is actually a bug in-game, so we didn’t make the default Krosus script ignore the dual wield miss penalty. At this point though… seems like blizzard has no plans to fix it so maybe we should just make it part of the default script.


#19

was there a miss or no-miss functionality to Kromog or for Kologarn? might be good to know - if people did not miss on those either then it would allude to blizzard’s strategy for the programming behind these sorts of fights.


#20

you said that I could click on “script” at the top of the sim that you linked… I see no script button. Sorry, not trying to be difficult… I see two buttons - Add to Favorites and Browse