[7.2] Legion Shadow Priest Guide and Discussion

[Sesethi]

Dunno if it was mentioned, but any way we can download the trinket sheet as an excel file?

[davesignal]

That's a bit alarming. I mean since it's a universal 10% buff it's not going to effect stat weights or anything like that thankfully. I'll definitely get that fixed.

Not really a universal 10% buff, though, is it. Essentially this is a non-issue outside of Voidform, because Shadowform is going to be up anyway-- the 10% buff there is 'baseline'.

But Voidform? This would mean Voidform is simming 10% stronger than it should be. The sim is essentially weighted more towards Voidform than it should be, so the stat weights would likely reflect that. I mean, it won't be a huge issue-- haste is certainly going to still be the best stat, crit following, etc, etc-- but it won't be as accurate as it should be.

[Djriff]

Dunno if it was mentioned, but any way we can download the trinket sheet as an excel file?

You can copy and paste it.

Not really a universal 10% buff, though, is it. Essentially this is a non-issue outside of Voidform, because Shadowform is going to be up anyway-- the 10% buff there is 'baseline'.

But Voidform? This would mean Voidform is simming 10% stronger than it should be. The sim is essentially weighted more towards Voidform than it should be, so the stat weights would likely reflect that. I mean, it won't be a huge issue-- haste is certainly going to still be the best stat, crit following, etc, etc-- but it won't be as accurate as it should be.

Right, but it's not going to be a huge impact. Hopefully is fixed tonight.

[Jullyx]

Why is there so little fuzz about the insanity issue? Does anyone know the cause yet? It's building up to become truly infuriating at this point. I'm doingn so many dungeons, so I have to deal with it quite a lot. I always know when it's gonna happen and I know I don't actually lose any insanity, but it's still a pretty big dps loss.

[Dawnrage]

Why is there so little fuzz about the insanity issue? Does anyone know the cause yet? It's building up to become truly infuriating at this point. I'm doingn so many dungeons, so I have to deal with it quite a lot. I always know when it's gonna happen and I know I don't actually lose any insanity, but it's still a pretty big dps loss.

It's happened twice to me so far since the patch, so I don't know the exact cause. If you know when it's going to happen, can you tell what's causing it then?

[Jullyx]

It's happened twice to me so far since the patch, so I don't know the exact cause. If you know when it's going to happen, can you tell what's causing it then?

I would estimate it happens 5-10 times per dungeon, but it's not that noticeable when you don't get high stacks, as you only really drop the last 10 or so. I can say that it close to always happens when I move and swap targets a lot. It's very apparent that the insanity drains way slower than expected for a couple seconds before it all just disappears. I don't know if it appears that drain has reset but actually hasn't, or bugs in another way, but it needs to be fixed as soon as possible.

[spiritsurge]

Why is there so little fuzz about the insanity issue? Does anyone know the cause yet? It's building up to become truly infuriating at this point. I'm doingn so many dungeons, so I have to deal with it quite a lot. I always know when it's gonna happen and I know I don't actually lose any insanity, but it's still a pretty big dps loss.

What exactly is the insanity issue? I have nothing better to do at work then sit here on these forums and I haven't read anything about it. Is it the voidform randomly ending issue that i've seen a little bit of?

[Zoopercat]

Wanted to jump in really quick, some of you know me by now. I am from Ask Mr. Robot and keep an eye on forums in case people find bugs or have feedback.

I wanted to address DJRiff's comment quickly, regarding stat weights:

AMR's simulator essentially does the same thing, it's just not as robust.

Our stat weights are actually MORE robust than SimC. In fact, if you want to use SimC for stat weights, you'll want to analyze the reforge plots, not using the stat scaling function.

1. AMR uses 1000s of data points for each stat to find the best set of weights. SimC uses only 2 data points, 1 above your current gear and 1 below your current gear. That's why they can be so volatile (with so few data points).

2. AMR lets you customize how far out you want to plan for your gear. Stat weights evaluate how good a piece of gear is (as we all know). You can look at 'right now,' which gives you the best stat weights for the next couple of items. This is similar to the range of gear SimC would be good for. But on AMR you can also plan out further, like find the best gear for the next 20 iLevels, for example. It's all up to you.

3. You can now get stat weights for actual bosses on AMR. We published our Ursoc script recently, and later this week or next week we'll have a mythic+ script out. And we'll keep adding more. This is quite handy so you can get stat weights for real encounters, and it is super useful when evaluating logs. I've been comparing ursoc logs to our ursoc sims and it's pretty awesome to see how they line up and use that for a way to improve.

4. Our gearing strategy sims (stat weights) also simulate every item with a proc. Because as we all know, trinkets don't rank well with stat weights. So we sim them all instead and bundle that with the stat weight report. Then when you go to rank gear, all items with procs are ranked directly from simulations

If there's a feature missing in AMR, let me know. This is just the beginning, and we're adding more stuff all the time Don't be shy!! Hit me up.

[davesignal]

My main problem is that I can't get the thing to work. It always disconnects during long runs, or in the middle of several consecutive small runs, and can't resume after reconnecting which necessitates starting over. I'd love to actually see what that output looks like, but..

[N1gh7h4wk]

1. AMR uses 1000s of data points for each stat to find the best set of weights. SimC uses only 2 data points, 1 above your current gear and 1 below your current gear. That's why they can be so volatile (with so few data points).

So not sure about AMR simulation, but does this mean you run like 1k sims for every point of that 1k scale delta? or does that mean you run about sims for (nearly) every point of the statcurve and average over it?

[Kilee25]

Why do people still use SimC? Serious question. You are at the mercy of making sure the people who maintain it do a good job on it, and no offense, but even during the Twintop era it wasn't even accurate most of the time.

I started using SimC heavily at the end of Wrath. At that time, I could download the source code to the game and modify the items in it (ie trinket lists and set bonuses).

At some point between then and now, the software got a lot more complicated. The last time I looked at it, the resource files didn't have the same behaviors as they used to, so I couldn't figure out where to modify trinkets and check on their values.

The program in general has become a lot more complicated...

......

Our rotations, talents, etc.. are much more complicated than they used to be. A typical APL for an entire expansion was only 20-30 lines. These days APLs for even one set of talents and gear are like 150+ lines of code.

SimC is no longer a program that can be used by normal people. I would argue even that it's become so complex and unreliable that even veteran players and programmers have a difficult time getting anything meaningful out of it.

......

I hate how everyone dismisses what Ask Mr. Robot is doing so casually. Even if they weren't doing good work, having ANY sort of second opinion to SimCraft is the best thing that's happened to the game in years. Having ANY sort of alternative for players, since SimC is basically completely unsuable to average people... well you'd think more people would be excited about it. But everytimg I go on Discord or H2P, all I see is people hating on AMR and telling the community how crap it is. There's some kind of internal feud that's been happening towards AMR for years, and I never understood it.

.......

SimC is only as good as the APL it's using and the people who are programming it. It's been a very mixed bag over the years, and I never, ever, ever trust what someone posts unless I can test it myself and prove beyond some reasonable doubt that it's behaving the way that I would in game. Basically I have to do the work myself, or I don't trust what people post. TT and the rest of the work being done on H2P - I would basically just simply view it as another opinion to look at while I do my on work. I would compare and contrast my findings with theirs, and try to come up with unbiased opinions based on that. Having another point of comparison with AMR's tools is a freaking god-send to a person such as myself who takes an explorative scientific approach to it.

....

If I were programming SimC for S2M, this is what I would do. Instead of having the Sim end when the boss died, I would have it end when my character dies. I would limit how short fights could be with StM, then have the sime pop it at 35% health or so (basically whenever you get execute and all your resource builders going. Then I would just have the sim "play it out" until it died. Then I would treat whatever fight length came out of that as the final fight length from which to draw stats from. The stats would have their own incentive to trend towards longer S2Ms because most of your damage would come from longer VFs anyway.

From there it would basically just be spending an ungodly amount of time on the APL, trying to make it adapt to the various rotations that would change at different haste points. At a certain point, I am sure that the APL would dictate stat weights more than anything else. I'd probably go back to the work that I did for HFC, having the APL change into different "modes" based on various buffs and modes, etc...

But I'm not saying this would be better or worse... it's just what I would start experimenting with first.

[Yellowfive]

So not sure about AMR simulation, but does this mean you run like 1k sims for every point of that 1k scale delta? or does that mean you run about sims for (nearly) every point of the statcurve and average over it?

We generate around 2000 data points by default, with 1000 iterations per data point, or about 2 million "iterations" in simc speak. The number of iterations is irrelevant though -- all you care about is the margin of error and confidence interval. We use a slightly higher margin of error (0.25%) but generate a lot more data points. This generally gets much better results than using a lower margin of error (say 0.1%) with far fewer data points, as simc does.

You can more or less replicate the technique that simc uses by customizing our gearing strategy simulation to use a very narrow ilvl range. It will do a similar thing, but do it better. It's essentially a superset of what you can do in simc.
@davesignal It's pretty rare to get connection issues like that... must be something unstable between you and our servers. Our next update will introduce some extra error/retry mechanisms, maybe it will help. We do have hundreds of thousands of users doing simulations though... so it is definitely not a systemic problem with the simulator, it is something specific to you. Do you have any kind of unusual internet setup?

[davesignal]

We generate around 2000 data points by default, with 1000 iterations per data point, or about 2 million "iterations" in simc speak. The number of iterations is irrelevant though -- all you care about is the margin of error and confidence interval. We use a slightly higher margin of error (0.25%) but generate a lot more data points. This generally gets much better results than using a lower margin of error (say 0.1%) with far fewer data points, as simc does.

You can more or less replicate the technique that simc uses by customizing our gearing strategy simulation to use a very narrow ilvl range. It will do a similar thing, but do it better. It's essentially a superset of what you can do in simc.
@davesignal It's pretty rare to get connection issues like that... must be something unstable between you and our servers. Our next update will introduce some extra error/retry mechanisms, maybe it will help. We do have hundreds of thousands of users doing simulations though... so it is definitely not a systemic problem with the simulator, it is something specific to you. Do you have any kind of unusual internet setup?

Nope. Cable modem plugged into router, tower plugged directly into router. If I was on the wifi I could understand it, but I've got the only machine in the house hard wired to the router.

@Kilee25: Without going too deep into it, and not to discount the work they do for the community, I've felt for a long time now that the main aim of most theorycrafting is self-perpetuation. The main thrust of it has long since moved away from providing reliable, resilient guidance on how to get the most realistic good result and on to how to best produce the biggest sims, regardless of how fragile the assumptions and conditions necessary to produce them are.

The latter has its place, mythic raiding is definitely still a thing and there will always be a call for information on how to redline your DPS, but I feel like somewhere people lost the plot on providing reliable information on how to produce a robust set of assumptions and instruction that provides the greatest number of people with good results in the largest number of situations.

It's all too focused on the very highest end of raiding, in a time where the very highest end of raiding seems to be at its most anemic (compared to, say, several expansions ago).

[N1gh7h4wk]

We generate around 2000 data points by default, with 1000 iterations per data point, or about 2 million "iterations" in simc speak. The number of iterations is irrelevant though -- all you care about is the margin of error and confidence interval. We use a slightly higher margin of error (0.25%) but generate a lot more data points. This generally gets much better results than using a lower margin of error (say 0.1%) with far fewer data points, as simc does.

So it does a n-polynomial fit with n around 2k? Interesting. Or does it still do a linear average on data with higher error?
Also how do you handle so much computing time, for simc statweights we use around 27 different fight setups, running all those for 2 million iterations could take like 30- 40 days on a usual i7 core.

[dmfg]

Well davesignal, it's not helped by the current state of shadow, where
1. It's almost impossible to do meaningful DPS testing on dummies, and even in argent tournament you have 10 min CD
2. The difference in DPS from very small changes in VF duration can be huge
3. Performance is highly dependent on AS RNG
4. Performance is highly dependent on latency, to the point where it changes the rotation
5. The interaction between haste and crit is complex and Sims that analyse each independently don't give the full picture

And that's not even taking into account individual player skill or appetite for risk (how much chance of early STM death am I willing to accept in exchange for a better average?).

In short, it's never been more difficult to interpret Sims and how to use that data in practice. It seems to be more feelycrafty than ever before. That's not the fault of the sim coders or the TC, just the nature of the spec.

My interpretation of the TC is "if you can't reach VTor3, get more total (haste + crit), preferring haste but maintaining 16+% crit". But I bet everyone you all will give a slightly different interpretation. 16% crit might feel too low for some people, others might not care about crit. Some will value int even at the cost of haste. Etc etc. I don't think there's good quality blanket advice that works for SP.

[Deleted]

Does anyone else have strong differences between his own sim and the ones from wowprogress ? Mine are almost 100k lower for the same fight length. Are there any settings that should be used on simcraft to avoid the sim dying too early ?

[KablenTom]

Hi guys. Is there a good legendary item ranking? I got my 2nd legendary on Friday and my 3rd last night (sorry to those still without one!).

I've got the belt, wrists and feet now. Belt is a no brainer, but which of the other two to use? I'm conflicted since the feet are better stat-wise since they have haste, but the wrists seem like a better effect for DPS. Wondering how much their unique effects really do. I can see the feet being really useful in movement heavy fights but not used them in a raid yet so haven't tried them out.

Cheers

[N1gh7h4wk]

Hi guys. Is there a good legendary item ranking? I got my 2nd legendary on Friday and my 3rd last night (sorry to those still without one!).

I've got the belt, wrists and feet now. Belt is a no brainer, but which of the other two to use? I'm conflicted since the feet are better stat-wise since they have haste, but the wrists seem like a better effect for DPS. Wondering how much their unique effects really do. I can see the feet being really useful in movement heavy fights but not used them in a raid yet so haven't tried them out.

Cheers

The Boots barely procc, while the Bracers give you avg. ~1% steady damage increase.

- - - Updated - - -

Does anyone else have strong differences between his own sim and the ones from wowprogress ? Mine are almost 100k lower for the same fight length. Are there any settings that should be used on simcraft to avoid the sim dying too early ?

Just ignore Sims from wowprogress.

- - - Updated - - -

I started using SimC heavily at the end of Wrath. At that time, I could download the source code to the game and modify the items in it (ie trinket lists and set bonuses).

At some point between then and now, the software got a lot more complicated. The last time I looked at it, the resource files didn't have the same behaviors as they used to, so I couldn't figure out where to modify trinkets and check on their values.

The program in general has become a lot more complicated...

......

Our rotations, talents, etc.. are much more complicated than they used to be. A typical APL for an entire expansion was only 20-30 lines. These days APLs for even one set of talents and gear are like 150+ lines of code.

SimC is no longer a program that can be used by normal people. I would argue even that it's become so complex and unreliable that even veteran players and programmers have a difficult time getting anything meaningful out of it.

......

I hate how everyone dismisses what Ask Mr. Robot is doing so casually. Even if they weren't doing good work, having ANY sort of second opinion to SimCraft is the best thing that's happened to the game in years. Having ANY sort of alternative for players, since SimC is basically completely unsuable to average people... well you'd think more people would be excited about it. But everytimg I go on Discord or H2P, all I see is people hating on AMR and telling the community how crap it is. There's some kind of internal feud that's been happening towards AMR for years, and I never understood it.

.......

SimC is only as good as the APL it's using and the people who are programming it. It's been a very mixed bag over the years, and I never, ever, ever trust what someone posts unless I can test it myself and prove beyond some reasonable doubt that it's behaving the way that I would in game. Basically I have to do the work myself, or I don't trust what people post. TT and the rest of the work being done on H2P - I would basically just simply view it as another opinion to look at while I do my on work. I would compare and contrast my findings with theirs, and try to come up with unbiased opinions based on that. Having another point of comparison with AMR's tools is a freaking god-send to a person such as myself who takes an explorative scientific approach to it.

....

If I were programming SimC for S2M, this is what I would do. Instead of having the Sim end when the boss died, I would have it end when my character dies. I would limit how short fights could be with StM, then have the sime pop it at 35% health or so (basically whenever you get execute and all your resource builders going. Then I would just have the sim "play it out" until it died. Then I would treat whatever fight length came out of that as the final fight length from which to draw stats from. The stats would have their own incentive to trend towards longer S2Ms because most of your damage would come from longer VFs anyway.

From there it would basically just be spending an ungodly amount of time on the APL, trying to make it adapt to the various rotations that would change at different haste points. At a certain point, I am sure that the APL would dictate stat weights more than anything else. I'd probably go back to the work that I did for HFC, having the APL change into different "modes" based on various buffs and modes, etc...

But I'm not saying this would be better or worse... it's just what I would start experimenting with first.

Usually we try to keep Simc apl simple, but for something that complex as the s2m gameplay and decision making has become, you can't just write it in a few lines. A pretty neat and simple APL is for example the Elemental Shaman one.
I think the Problem about looking at it casually is missing important points. Like: no fight is patchwerk. except for patchwerk etc.
Simulation craft is a program made for TC, not for a casual approach. That's why we put so much effort on bringing decent average statweights, and having sources that you can look into and maybe even tweak your toon for a specific encounter type.
The oversimplification of AMR is what is a thorn in our side. It's actually just not that simple.

[Yellowfive]

The oversimplification of AMR is what is a thorn in our side. It's actually just not that simple.

What oversimplification are you referring to? A huge thorn in our side is people throwing out comments like that with nothing to back it up. If you want to talk specifics, I can show you how the AMR simulator is actually a superset of what simc can do at this point, and is much more prone to correctly-authored rotations, and generating more realistic scheduling of events. I think it's great when the two simulators can verify each other, but in general I like ours better from a user and accuracy perspective.

And with regards to your earlier post:

So it does a n-polynomial fit with n around 2k? Interesting. Or does it still do a linear average on data with higher error?
Also how do you handle so much computing time, for simc statweights we use around 27 different fight setups, running all those for 2 million iterations could take like 30- 40 days on a usual i7 core.

I threw out that "2 million iterations" number because people seem to think it matters... they see "2000 setups" on our gearing strategy report and think it means we did only 2000 iterations, which is wrong, we did about 1000 iterations for each of 2000 data points. That varies wildly of course, we go by margin of error, not number of iterations.

About half of those points are for stats, half are for special items. We do a multiple linear regression fit on 1000 of the data points for the stat weights.

[Kilee25]

I was present in the early days when AMR first appeared, and I was on the back end of the priest community during that time. My perspective on what happened was... two things.

One, the initial representative for AMR came into the forums and was argumentative and came across as a bit of a parasite. They would ask for all our theorycrafting, in a TLDR format, then if we helped they would criticize it or just start trying to poke holes in it. At some point, zoopercat kind of took over and that first guy went away. She had a much better "bedside manner" than the first guy (damn I can't remember names). But that initial impression kind of stuck, and most of the people who were "in charge" of the priest community are still around and in the same positions. In some ways the "priest community" is like this old stubborn elephant that never forgets or forgives a slight no matter how large or small.

The second thing is is sort of an "uncomfortable truth". AMR is competition for the entire TC community, and a threat for class-based fan sites. Class sites in general have a tremendous stock built around their theorycrafting. Guides are generally a dime a dozen, and people in general can kind of figure out what to do by just playing around with it and talking to their peers. Theorycrafting, and specifically stat weights, are what sets a site apart from the rest. If you establish yourself as an authority, or heck THE authority, on stat weights... well you essentially control the spice at that point. Nobody is going to give that up or concede that AMR is a viable alternative. You also have the fact that TCers are something like mini-celebrities in the wow community. You only have to look at that fairwell thread to Twintop to get an idea of the fanatical level of stardom that position gets you. AMR is essentially taking a piece of that power out of the class community and relocating it to another place. And that other place is a site that also generates some revenue for those who work there. TC with SimC in the fan community by and large is free, and open source. So AMR basically carries with it this stigma of being not as morally altruistic.

Maybe I went too far in some places, but that's basically my view of why the hostility and lack of leniency towards AMR over the years. It has nothing to do with accuracy or method, but those things are attacked first because the truth is less noble and PC. I think also, there's a lot of people who don't know anything about it and are just parroting what the community leaders say. If somebody like Twintop says AMR sucks, the average fan site goer is just going to take that at face value and propagate it out to the rest of the community as loudly as they can. Even the community leaders are not immune to this hive mindset.