Custom Sims / Client DL

Ask Mr. Robot
21

Updating the numbers/mechanics for each spec takes at least half a day, and then making a good rotation that handles all these talents takes at least a day.

It would be an interesting thought experiment to consider whether the simulator could derive good rotations itself on the fly.

Each spec (including talents) has one or two resource levels, and perhaps a half-dozen procs that affect the choice from about a dozen abilities. The combinatorics don’t seem intractable.

Seems like a good candidate for a genetic programming approach. Once you had a pool of some moderately viable rotations, the optimization process might converge pretty quickly.

22

A few people have kicked around the idea of rotation optimization… it is certainly possible theoretically. I think it would be possible to arrive at a pretty good rotation automatically, but getting the “best” rotation would be extremely hard. There are a lot of variables that can be changed in a rotation so it is difficult to bound the problem to a tractable size.

That said, we have always been proponents of using relatively simple rotations that get you very close to the theoretical maximum output. The theorycraft community at large, on the other hand, is obsessed with absurdly complex rotations that squeeze every tiny bit out of the simulator, even when such changes are arguably gaming the simulator more than modeling the game. (And even when such tweaks are only relevant in a very narrow set of cases requiring specific builds, gear, etc.)

The point being that it would be difficult to get the community to accept a really good auto-generated rotation if some person is able to find even one tweak that is a little better some of the time.

It would be a fun project though!

As Swol has mentioned, once the initial push is done here, we’ll revisit some simulation features down the road. We are definitely sad about letting the simulator fall to the wayside for now – we put a LOT of effort into it and it was probably the most fun project we’ve done with WoW. We are just really time-limited right now and need to focus on other features first, like talent optimization. The simulator is just not the right tool for that.

2 Likes
23

Can we hope that if the simulator is not revived, the source code will be published on github?
I think there will be a lot of people who want to continue developing it. At least, I am interested in it.

24

I am a bit disappointed as I think simulation is generally superior to modeling but we’ll have to see. If you look at the number of possible combinations of talents it does get exponentially more complex but I think in most cases the comparison would be between a low number of builds as opposed to trying to sim every discrete component across all others. I can’t think of a spec that has more than 4ish builds in the talent tree and maybe 3 in the class tree. That doesn’t seem too bad considering you can force them to use their own hardware in the simulator client.

It sounds like you have picked a direction and it’s too late to change course now but if the opportunity arises later maybe instead of the talent sim setup you had for pre-DF talents where every selection doubled simulation time.

I also don’t understand why formulation is significantly less resource intensive from a development standpoint. In order to spit out good numbers you already have to work spell coefficients, cooldowns, etc. into a working rotation, right? Or are you just crunching based solely on observed performance? If the latter, surely sample sizes for underused talents or setups have tiny sample sizes.

It’s too bad as I often used the simulator to discover (and perform well with) off-meta builds or even specs.

3 Likes
25

Simulation is certainly superior for exploring the nuances of rotations and figuring out how to play the game better/differently. We are mainly concerned with picking gear, so that has to be our first priority. Simulation turned out to be rather limiting for that goal.

We have been looking at the talent trees to see how many different combinations of talents need to actually be calculated to generate some useful data for helping people pick talents. Even if you choose a starting point with a few nodes more or less locked in for a “build” - the amount of ways you can fill in all the other choices is very large. On top of that, to give all the setups a fair shake, you have to test them with at least a few combinations of stats, and then you have to try at least a few builds from the class tree as well.

We have never simulated every possible combination of talent/gear - we always pruned it down to something relevant or representational when making the gearing strategies.

The up-front work is comparable, but it is much faster to maintain and adjust. If people tell us that they don’t think the optimizer is picking the right gear, we can adjust the math and post an update and people will have new rankings right away. With the simulator, we first have to get the simulator to actually do something that will produce the result people want, then run all the simulations, then we can update the site with the new rankings.

Often, the rankings people want to see do not actually follow what the simulator finds. Things become popular that do not fall into what a “patchwerk” simulation will spit out - using only simulation to rank the gear ends up being very rigid. “Formulation” gives us the flexibility to change the model however we want very quickly to match up with what people want to see.

I’m still thinking about where the simulator falls in the future of the site. It’s a really cool tool that we maintained for the last 6 years… but it saw very limited use. I had always hoped it would bring more people into theorycraft because of the rotation editor and find widespread adoption in the TC community because more people could effectively be involved in rotation development. That never happened. We also have never been able to effectively explain to people that it is open source. The “class modules” were all available as the wiki. We weren’t writing code to make the simulator work, we were using the wiki, the same way anyone else could. The only part not open source is the engine, which is equivalent to the part of simc that only like 2 people know how to tinker with, and that no one ever looks at. We created a way for people to work on the simulator without being programmers… and no one was interested!

I do feel bad that the few people who were using it now don’t have this cool tool to use. Like I said, maybe we could put it out there with the UI updated to hook into the new talent trees, and then let people update the classes if they want to. We always envisioned it as a community-driven tool. Maybe it could become that now that we don’t have to update it ourselves to keep the gear optimizer working.

1 Like
26

Well it would need to be well separated because people like me are going use this other tool to compare to the gearing suggested by AMR and challenge AMR’s results :stuck_out_tongue:

27

Yeah you’re right, forget it, no simulator :wink:

1 Like
28

The first mythic week started and I realized how much I miss the simulator. I literally can’t get the damage/survival balance right now. All things have to be tested during runs, which is simply not reproducible.
Also now there is no reference point to the “maximum” hps and I do not understand where I am relative to it.
And of course the lack of ability to compare different builds is an irreplaceable loss to all who used the simulator.

This frustrates me terribly.

p.s.
I want to test how I can increase my generation of Holy power and how it affects my survivability. But I can’t, because I don’t even have gears with haste. I used to be able to do it in the simulator in a couple of minutes…

1 Like
29

Can we at least see, what rotations are you using for each spec, was a big fun of yours simple arcane rotation, and used to learn rotations in amr simulator

2 Likes
30

I am also not happy with the Simulation section gone. I used the function pretty regularly and where quite happy with the accuracy of the results. I dont mind using the local client to run the sim on my own computer but this is just a bummer.

1 Like
31

Thanks for letting us know - I want to reiterate that we didn’t remove the simulator because we thought it wasn’t a great tool. We wouldn’t have done all the work to create and maintain it for all those years if we didn’t think it was high quality.

Right now it is a question of time/prioritizing. It’s not off the table for some simulation to come back to the site, but we have a very limited amount of development time - we’re already working long hours to get the optimizer and new talent optimizer working and polished up ASAP.

After we clear that hurdle, I can look at bringing back some simulation (especially healing) on a bit of a less intense schedule. I do feel bad for the people who were using it (like myself, ha!) - I miss it too.

1 Like
32

I would even pay more if that’s what it took to bring it back. For what this tool does the value is pretty high.

I realize that doesn’t mean anything in isolation because you would need a lot of people to commit but it may provide a data point for you.

1 Like
33

i also used the AMR Sim to discover off meta speccs and play them well. Your Sim was soooo much better and far more easy to use than Simcraft. I dont mind have my pc run a 3 or 4 hour sim so he can give me decent off meta build even when i am afraid that there are currently only builds with Netherportal/gul dans viable.

2 Likes
34

I really want this back as well even if it was for a premium cost. I want to sim my healer so bad that I’m looking inside simcs class modules trying to make sense of anything wishing I could add the spell data and crap myself but it’s hopeless.

I just don’t understand how these math models that don’t use sims work, especially to somehow be better in choosing your gear? Like the order you use your spells doesn’t affect stats?

I agree it was annoying that TCs and blind sheep community constantly shunned AMR though I read many times that they used AMR for the easy to access spell data wiki stuff. But they also say there is 0 value is simming healers which just isn’t true. I imagine that the people who make QE Live would also have to see the value in simming healers, it’d be awesome to see y’all collab if the healer sim stuff was brought back.

1 Like
35

I don’t think the math model will be better than the simulations were. I think, for the purpose of picking gear, it will be comparable. The benefit I’ll have over simulation is that I can more easily adjust the results to fit community trends. An example of this is the slider I just added to adjust the value of haste independently of the other stats, since haste is so popular for multiple specs that don’t actually scale well with haste.

The healer simulations were the slowest of our simulations, so creating new data took a loooooong time. And that was without the massive number of stat/talent combos now relevant in dragonflight.

There is a good chance I will bring back at least some of the healing simulation for fun because I enjoy messing around with it. I think the healing simulation worked really well for coming up with play styles that actually work, and were also off-meta.

This is where we started back in the day, and came to the same conclusion. Simc does not have the pieces necessary for a healing simulation. The most important part is the ally targeting logic that we then exposed via the rotation editor. And then we also added AI healers that were programmable via the boss script, as well as the incoming damage pattern. Really, all the pieces were there to set up any scenario one wished. I’m still disappointed that the TC community refused to engage with the tool. I built it as a free tool for anyone to use and offered multiple times to set up rotations and scripts per their specifications as examples to get them started.

36

I’ve not played WoW for a long time now so only just discovered the simulator has been depreciated, at least for the time being.

I’m still in a few WoW Discord servers, one of them is for an addon I used to help with the in game display of de/buff and CD. As there’s a new class this expansion it doesn’t have anything and there were lots of questions about how to make one. It’s an old addon with no in game configuration so it’s a steep learning curve for new users!

So I came to look at the Guide section to help me try and visualise how it would play, I didn’t find the Wowhead guide very useful. Perhaps it would be better if I could log into the game.
When I saw the guide wasn’t there I thought I’d just do a simple sim with logging turned on and look at the output, I often found that was the best way of working out how something was supposed to play. :slight_smile:
With that gone too I’ll leave it to them to climb that learning curve!

We know that @Sienss and I did a lot of poking at the Wiki, I like to think I helped but answering my questions might have slowed you down more than helped you @Swol!
I can think of half a dozen of us who, I know, would run a lot of sims to get a better understanding of what was happening.
How many more people were simulating?
As a percentage of registered users and total with unregistered users?

Looking through this thread there are about a dozen people who are expressing how much they’re missing it. Some of you even acknowledge how baffling the anti-AMR crowd was, I had a more than a few in game, guild forum/Discord and even official Blizzard forum “discussions” about it and mostly gave up as it never changed any minds, it wasn’t worth my time. :frowning:
I’d just state I used it and if questions were asked I’d justify why, occasionally people would look into it more but if they get questioned by a raid leader/member about why Z instead of X/Y they’d not have the knowledge to explain why and would just swap so they didn’t have to put up with it. :frowning:

I understand why Team AMR aren’t spending time on it now and I’m sure they’re feeling the loss more than the rest of us, they spent a lot of time and effort making it and used it themselves.
The rest of us needed to be louder in our support over the last six years, it’s easy to see that now but I’m sure the rest of you felt like I did, it’s too hard to educate sheep!

Stepping back a bit it isn’t just a small community of gamers who have to deal with this.
Since the rise of the Social Media “Influencer” experts have been left scratching their heads about why people listen to them vs listening to the actual experts!
From dieticians, personal trainers, midwives, doctors and scientists.
Over the last three years it got worse, it became even more apparent that people are more willing to listen to a pretty face, with no expertise in anything relevant, than they are to medical professionals and scientists who’ve dedicated decades of their lives to the subject at hand.

So it isn’t surprising that they’ll listen to popular streamers/raiders/loudest voices over the people who built a simulator for a video game. Some of those loud voices worked with SimC but often only tweaking rotations to game the simulator more than actually implementing changes in the class. I can sympathise with them about why they thought they should be listened to, they were putting more effort in than the people just asking the questions, but at some point - pretty early on - they’d have realised that explaining the intricacies was much harder and time consuming than stating “this is best, everything else sucks,” and leaving it at that. It didn’t help the question askers understand why, but it took up less time

You’d think that but I remember them popping into Discord and saying something like, “you’re doing it wrong,” but when questioned they didn’t elaborate and said something about maybe coming back later to discuss, but I never saw them return. I was looking forward to that discussion.
At least I think it was Discord, it might have been a forum post in which case you’d be able to find it.

Well this got longer than planned, I should stop typing now!

One last thing though.
I think the move from forums to Discord made any kind of useful theorycraft discussion almost impossible.
Also I suspect the percentage of people doing simulations here would struggle to be 1%, the percentage of us who tested out modified APL would be two orders of magnitude smaller.
Most people just want the easy answer and do what it says, trusting that the people telling them/making the site know more than they do.

2 Likes