Suggestion: Optimize for two specs via their optimal stat green zones

Serious discussion
, ,
1

Hi. I noticed that some classes in some patches are quite lucky from the perspective of easily getting gear that is in AMR’s “green zone” in the optimal stat graph for more than one spec. Naturally, other classes aren’t so lucky. And after looking at the optimal stat graph for a while, I just got an idea that could be extremely helpful for users like myself. Can AMR devs add the feature of trying to find the optimal stat combinations for two specs of a character simultaneously?

This may be possible to do manually but it would take a significantly long time for most users, and it would defeat the purpose of subscribing to an excellent and quick service like AMR. The idea here is that AMR’s algorithm would start searching for the stat combinations from two specs where each secondary stat, like crit, has the lowest difference if that stat in one combination for the first spec is subtracted from the same stat of the combination for the other spec. So, crit here minus crit there is the lowest, haste here minus haste there is the lowest, and so on. And the algorithm goes through those stat combinations from the top (greenest) down (the red zone). Then, AMR shows the best stat combination to the user that wants to gear up for two specs simultaneously. And what would make this feature amazing is if AMR can also show how much DPS the user is sacrificing from each spec by doing this, which would hopefully sometimes be a low or negligible percentage for most users. And if one of the specs is tanking or healing, the DPS can be replaced with a TUFness score and HPS, for example.

I hope that I explained what I mean clearly and that the AMR’s decision makers can consider adding this feature, since it can be extremely useful for users who want to save a lot of time while gearing up a couple of specs that they want to use almost equally.

Thank you.

2

I think I understand what you are saying. That would be really hard to do in the optimizer. Blending together the data for two different gearing strategies and doing two BiB optimizations at the same time… Not sure that is something that is going to happen, heh. We can always keep it in mind, but I think that it’s one of those features that would be way more difficult to pull off than it seems at first glance. There is also an inherent level of ambiguity to the results that would make it difficult to say for sure if it’s working “right” or not.

3

I’m not that knowledgeable in data analytics but it did feel doable to me, mainly because of what AMR establishes in that there isn’t really a single optimal combination of stats for most or all specs, or in other words, a spec doesn’t have to maximize two stats and ignore the other two, for example, as most players do manually, so there’s always a chance that a combination that works for one spec is pretty close to a combination that works for another spec. As I mentioned in my OP, I’ve seen this working already without it being a feature for the Demon Hunter class and also for the Warlock to some extent, for example, mainly because the DM and Demonology Warlock tend to accommodate a semi-balanced weight of stats, so it is possible to make two stats moderately higher than the other two, for example, instead of have them all perfectly balanced, and see this as an optimal combination for another spec while still being optimal for the active spec. This is all happening by Blizzard’s class design and scaling now, but it strongly feels doable by AMR as an objective, at least in many cases. And as I said, it’ll be a really useful tool for people who want to get the best return on their time investment from the vault, for example, if they really want to juggle a couple of specs for a class, like tanking and DPS spec or a healing and DPS spec.

4

Hey !
I would like to know why you would like to do so ?
I only reason i see is you want to focus on gear around two stats “usefull” for two spec at the same time.

What i do is having two page of AMR one in the first spec and the second in the second spec and look what loot they have in commun ? (in the upgrade finder MM+ for example)

Is it what you would like to do ?

5

AMR doesn’t really try to optimise specific stats in isolation. From what I understand, they do a large number of simulations testing the dps that different stat combinations will do, then use machine learning to use this data to figure out what arrangement of gear will do more dps than another.

Having the right objective function that you are trying to optimise for is important.

I think at the moment, they optimise for the highest dps/hps, etc in your highest priority spec, and then keeping what enchants/gems/conduits that are being used by that spec, optimise for the highest dps/hps for the next highest priority spec. In other words, they focus on optimising one spec at the expense of others. For most people, who mainly just play one spec, and occasionally play others, this works fine.

I think you are looking for some kind of multi-objective optimisation, since you play 2 or more specs about equally, and don’t want to completely sacrifice the performance of your second spec for the best performance of your first spec. For example, if you lose 50 dps in your main spec, but gain 100 dps in your second spec, you would be ok with that trade off.

The trick in doing this would be to come up with the correct method to combine these competing objectives together, and no doubt, it would be different for everyone. A naïve way would be to optimise the sum of dps over the different specs, but I don’t think this would have the intended effect, since 2000 spec 1 dps + 1000 spec 2 dps would be treated the same as 1500 dps for both specs. You could also look at include minimising the difference in dps between the two specs, but this also could have some unintended effects as well.

There is a lot more complexity involved in optimising for multiple objectives at the same time, not to mention the additional complexity in processing these calculations.

6

At various points in the past we have kicked around doing things like this… and there are other theorycrafters who are big on this idea – they generate stat priorities for e.g. 5 different scenarios, then do a weighted average depending on how important they think each scenario is.

We were never able to get behind that approach… it seems plausible until you start thinking about it too hard. You really end up with an “optimization” that isn’t very optimal for anything.

There are other approaches… something like agrocasey said. We could explore different methods for combining the scores of two different setups then optimize that. Certainly doable with enough code… but it would also be quite a bit more CPU intensive, not to mention the UI implications adding complexity to the site.

While these kinds of things aren’t totally off the table, I’d have to think about them a lot more. My intuition says that if your stats are optimal for one setup, they are at least “fine” for another setup. Major swings in the value of stats from one setup to the next within the same spec are rare. So the first thing I’d want to do is an analysis of how much benefit such an approach would theoretically give a player.

7

I’d like to do this because I experiment with several classes to find what’s most fun and at the same time competitive with other classes in terms of effectiveness as a tank and my DPS meter as a DPS (while I don’t forget my utility, interrupts, stuns, etc., of course). This experimentation of mine can be quite time-consuming and slow down gear progress on all classes, especially with Blizzard’s current RNG-based reward or gear-progression system. Accordingly, I would like to have one set of “acceptable but competitive” gear for my Frost & Blood DK, as one example. It’s not just about checking a Vault drop that is acceptable for both specs, as you suggested, and it is indeed something that I do. It’s also about prioritizing which M+ dungeons I run on each class, for example. I can’t do this for both specs while feeling mildly confident that I’m doing something acceptable without some sort of “math” behind my approach. It is why I always subscribe to AMR when I return to the game and pretty much never play capped characters without constantly optimizing them on AMR.

I don’t want to focus on just Frost on my DK, thus suffer from a lot of downtime waiting for tanks in my M+ runs and also gimp my ability to tank with my DK on higher keys as my gearing strategy is focused on DPSing as a Frost. And I don’t want to focus just on Blood either, thus give up my ability to produce any meaningful DPS in higher M+ keys and also suffer from burnout due to only tanking. And because I take hiatuses a lot, it’s hard to stay in a guild or keep in-game friends for long, so I cannot rely on guildies carrying or helping or tanking for me when I want to take a break from tanking. Finally, the third scenario, focusing on both Blood and Frost on my DK is not an option unfortunately due to the time limitations; as I indicated, I’m also experimenting with a Demon Hunter, a Hunter, a Warlock, a Mage, and I may even miss my Druid.

Moreover, I rely mainly on M+ for my gearing, but my ultimate goal from this gearing up is to be raid-ready. So, at least in a few of those classes, I will need to invest even more time specializing in single-target legendaries and maybe even a few pieces of gear later on.

Thus, I’d really like to be able to balance two specs in my gearing strategy and dungeon prioritization, simply because there’s no time to divide my attention upon several specs individually while my attention is already divided upon several classes.

And I’m fully aware that I won’t be topping DPS meters at the beginning while doing this, and I’m fine with that. I just don’t want to look utterly pathetic and useless as a Frost DK in higher M+ keys when I feel like taking a break from tanking, to refer back to my earlier specific example.

And as I said above, some classes seem already really helpful in fulfilling this need of mine without AMR catering for it specifically, like the DH, where I can find certain stat combinations that look green for both, Vengeance and Havoc. I hope that things like this stay the same always in the game. And what I’d really like to see from AMR is a sort of optimization feature that shows me green bars for two specs for most – if not all – classes, even if not bright green, without fooling me obviously but actually base this on mathematically sound algorithms.

1 Like
8

I mostly gear up in M+ at the beginning, before being raid-ready, so I almost never look at the projected single-target DPS in AMR’s guides when exporting a specific character. What I rely on and aim for is seeing green bars under “Your best gear has:” based on the understanding that every spec can be viable with a wide variety of stat combinations.

And this is why I thought that what I’m looking for may be done fairly straightforwardly with the method I described: programmatically go through the stat combinations in two specs from the best down, calculate the difference between the same stat numbers in both specs by subtracting the crit in the best Frost DK stat combination from the crit in the best Blood DK stat combination, for example, and so on, add up all those differences, then sort those differences from the lowest to the highest, and the lowest [may] just be the stat combination that balances the Frost and Blood specs for the DK best, for example.

9

Thank you for considering my suggestion and contemplating it at least, yellowfive.

When you have a moment, please read my other replies above, as they have a lot of details about what a user like me is going through and why he needs a feature like this. And I invite you to keep in mind that this approach is rather “casual,” almost exactly the opposite of “min-maxing,” so I would imagine that it would be appreciated by a wider range of users who play the game more casually than those who laser-focus on one class and/or spec to be the best at it in a raid. You know better how many “casual” players use your service with a subscription, but what I would imagine is that the service is used by both hardcore and casual players, with the casual players being the bigger percentage by far, especially because AMR is much easier, more user-friendly, and faster than Simulation Craft.

And forgive my ignorance if this sounds like an uneducated or counter-productive approach to achieving my goal mathematically and programmatically, but perhaps a rather primitive idea that I have in mind now, which I hope can be enhanced or improved a lot with your team’s knowledge, is one that depends on a fact that AMR emphasizes a lot: Contrary to popular belief, there isn’t one stat combination that is “best” for a spec. Rather, there are always at least a few – if not many – stat combination ranges that can be optimal for any given spec. Accordingly, the idea that I proposed above, and to use the Death Knight as an example again, is that AMR goes through the best stat combinations for Blood and Frost from the top downward, compares each stat in both specs, subtracting the crit in the top Blood stat combination from the crit in the top Frost stat combination, doing the same for each other stat of the four, then adds up the total of the absolute values of those differences, then does the same for the next best stat combinations for both specs, and so on, then after a few hundred (or thousand or million?) comparisons, sort the total differences from the smallest to the biggest, and treat the smallest difference to represent the most optimal stat combination that balances both specs for the DK, Blood and Frost.

Then, if you want to go a much appreciated step beyond that, AMR can then show the user some sort of a negative DPS percentage for that stat combination on each spec, Blood and Frost, to show the user how much DPS they’re losing by using that balanced stat combination instead of using a default one-spec-focused stat combination that AMR has always used.

10

So thinking out loud a bit: say I have two specs that I like playing in M+, Blood and Frost. As the website is currently conceived, I’ll generally favor one of those specs a bit more than the other, let’s say Blood. So I’ll optimize Blood with no compromises, then create a Frost set that is as good as I can get without messing with my Blood set.

What kind of compromises does that Frost set need to make? It can’t change gems and enchants. It can’t change conduits. And that’s it. So the next question: how much performance is my Frost set going to lose by using potentially imperfect gems/enchants, and maybe having to use a different soulbind? In 95% of cases, not enough to care.

I have generally played a tank for most of WoW, and even though I spend most of my time tanking, when I want to maintain a good set of gear for both specs, it is not uncommon for me to set my DPS spec to a higher priority and then build my tank set around it. This is because almost any stat is at least decent for toughness, so you lose very little by compromising the stats slightly to favor DPS. And a lot of people prefer a more DPS-oriented tank build anyway, and crit/haste (which are almost always preferred by DPS specs) are almost always the best for tank DPS as well.

Which comes to… maybe a fancy optimization that tries to blend two strategies into one set of gear ends up being not as good as simply using two sets of gear, prioritizing one of them, as the site currently works? It would be nice to try and quantify that, but all of my experience over the last decade has come to that conclusion, especially for casual play.

I certainly play like this – I choose gear based almost entirely on my tank spec, but still end up with a decent DPS spec. I might make a few exceptions… like if I know a boss drops a really good DPS trinket I’ll switch to that loot spec, or as I said above, optimize my DPS spec first to favor the soulbind tree for that spec, since I don’t lose much by using the 2nd best soulbind for tanking. (I find conduits and soulbinds to be very underwhelming for toughness in general.)

I think the only place that this “hybrid” strategy/build makes sense to me would be the upgrade finder. And the real question it’s trying to answer is: “Which of these activities that I can do has the highest chance of rewarding me with a good upgrade for either of my setups?” It would be cool to have a feature that answers that question directly, but I don’t think a hybrid optimization is the best way to do that. It would be better to think a bit more about that feature specifically and come up with an approach that uses the current multiple-setup optimization to get an answer.