Mr. Robot and Legendaries

[Yvaelle]

Part of the AMR mini-war is also an echo effect from a time long ago when AMR was actually full-trash for Shadow. It got a lot better a couple expansions ago, but the stigma remains.

There are of course still very valid concerns about - as example - Kilee's statement that he can't 'trust' AMR, even on little things - but as an attempt to make a far more accessible simulation system for the average raider, for most specs of most classes - it does very well IMO. For Shadow, it's really not as bad as it is oft-maligned - but for the serious theorycrafter - it's not enough.

It's all about picking the right tool for the job though - a screwdriver (AMR) is more than enough to hang a painting in drywall, handing a powerdrill (high competency with SimC) to the uninitiated is likely to result in either buzzed screwheads, or a hole in the wall: in some scenarios the screwdriver is actually the better option.

A great example is how most(?) other specs tend to compare themselves based on Patchwerk sims - which are easy enough - but Shadow typically compares itself on a broad spectrum of talent choices, encounter types, gear levels, stat distributions, etc: and then we usually report on the general trends for most - and the specific for the detail-oriented.

There isn't a replacement for the kind of care or support a community provides - AMR is no threat to communities - on the contrary - by making data-driven theorycraft accessible to the masses, it encourages the community to engage in theorycrafting and developing and critiquing theorycraft: it empowers us all by making us accessible. AMR therefore empowers communities, even if this is sometimes annoyingly achieved by convincing people of incorrect information, and then relying on theorycrafters to explain the errors or nuances that were either misinterpreted (by the initiate spriest) or miscalculated (by AMR).

[Yellowfive]

There are of course still very valid concerns about - as example - Kilee's statement that he can't 'trust' AMR, even on little things

He was actually talking about simc in his post, not AMR. But I would expect he has a similar perspective about any simulation, including ours.

It's all about picking the right tool for the job though - a screwdriver (AMR) is more than enough to hang a painting in drywall, handing a powerdrill (high competency with SimC) to the uninitiated is likely to result in either buzzed screwheads, or a hole in the wall: in some scenarios the screwdriver is actually the better option.

From my perspective, this analogy doesn't quite fit, or it would at least be better if you flipped it: AMR is the powerdrill and simc is the screwdriver. The reason for this is simple: I created the AMR simulator to do more than simc, not less. It's not just a pretty UI slapped onto a less-capable simulation engine. Most people just haven't explored all of the features, and think it's just the one big green button that says 'Simulate'.

Talking only about the core simulators first: AMR and simc are fundamentally similar. They take similar kinds of inputs (tell it what you want the player to do, tell it what kind of fight you want, tell it how the spells work, press go) and produce similar kinds of outputs (what spells were used, in what order, buff uptimes, DPS, etc.). And thus they produce very similar results when configured similarly.

simc stops there -- it is a pretty "raw" program. It has no real UI, it has virtually no other features built on top of the core simulation engine. Everything else is left up to the users to essentially program for themselves... either through modifying C++ code, writing python scripts, or some combination thereof. This is one of the fundamental things we wanted to work on with our simulator -- requiring programming skill significantly limits who is able to do interesting things with it. And even for the people who are able to do this... they are left with the problem of efficiently communicating the results to users.

When I started... I considered just modifying simc and adding stuff onto it. But a few things prompted me to write a new one: One, unless I sat down and got someone to go through all of simc with me, it was actually easier to just write a new simulator from scratch. Two, everyone benefits from having two independent models of the game. Three, I already had a working simulator from way back in WotLK. I have also written two other simulators in the past, so it was much more efficient to just revive that work.

Our core engine, while similar, actually has a few improvements. Our fundamental model of how the player chooses actions is more realistic. simc does things that are literally impossible to execute in-game, and while the impact might be small overall... there's just no reason to continue doing it. We have done extensive comparison of our simulation results to logs, and we think it is closer to what a real world-class player is physically able to do within the confines of the game, and should thus give more actionable results.

So now we get to what AMR can do that simc can't... it's really tough to list it out without it just sounding like a PR laundry list... the differences are so vast at this point. The AMR simulator can:
- easily load and modify any character with a good UI
- let you modify any rotation, boss script, or spell directly on the website, and run it immediately without messing around with code
- keeps an infinite history of all simulations you have ever done, and lets you easily share any report with anyone on the internet by simply copying the URL
- lets you share your rotations or boss scripts so that other users can subscribe to them and get your changes in real time
- import characters directly from combat logs
- do advanced custom simulation batches and let you visualize or download the results
- distribute large simulation batches across multiple computers or server farms even, without having to do any significant setup
- create gearing strategies that plug into the AMR optimizer with the press of a single button
- has full support for tanks and healers (still working on adding some more stuff for healers in particular though)

As a long-time simulation user and theorycrafter, these are the features that I have wanted for years. And I happen to have the time and appropriate skills to do them, so... here they are, enjoy!

All of the arguments and differences between the simulators... it's all the configurable pieces. A lot of people latch onto the rotations... "simc has a better rotation so it is a better simulator." The rotations are designed to be changed in a simulator, and it takes five minutes to do it. Our simulator focuses on different boss scripts. If you want to make the two simulators match exactly, well... you need to use the same boss script! These are fundamentals that simulation experts get, but that the more "casual" participants ignore or overlook. Kind of blows my mind... people have been trained to think that the entire simulator is a set-in-stone black box, rather than something easily modified. And to a large extent... simc is a black box from the rotation on down. They can get pretty gnarly. Improving on that with a better UI isn't just "fluff", it is fundamental to doing better theorycraft and having a larger TC community.

The AMR simulator does a lot more stuff than simc from my perspective. (And it is also free.) It's open and transparent. It's almost absurdly powerful for figuring out how to play a video game. And I'm always looking for feedback to add more stuff to it -- just use it and hit me up. Is there something about simc that makes it better or easier to do what you want to do that I've missed? Just let me know -- that is why I poke my head out every so often. This is one of the most fun programming projects I've done in a long time, it is not hard to convince me to add something

[MendUS]

He was actually talking about simc in his post, not AMR. But I would expect he has a similar perspective about any simulation, including ours.




From my perspective, this analogy doesn't quite fit, or it would at least be better if you flipped it: AMR is the powerdrill and simc is the screwdriver. The reason for this is simple: I created the AMR simulator to do more than simc, not less. It's not just a pretty UI slapped onto a less-capable simulation engine. Most people just haven't explored all of the features, and think it's just the one big green button that says 'Simulate'.

Talking only about the core simulators first: AMR and simc are fundamentally similar. They take similar kinds of inputs (tell it what you want the player to do, tell it what kind of fight you want, tell it how the spells work, press go) and produce similar kinds of outputs (what spells were used, in what order, buff uptimes, DPS, etc.). And thus they produce very similar results when configured similarly.

simc stops there -- it is a pretty "raw" program. It has no real UI, it has virtually no other features built on top of the core simulation engine. Everything else is left up to the users to essentially program for themselves... either through modifying C++ code, writing python scripts, or some combination thereof. This is one of the fundamental things we wanted to work on with our simulator -- requiring programming skill significantly limits who is able to do interesting things with it. And even for the people who are able to do this... they are left with the problem of efficiently communicating the results to users.

When I started... I considered just modifying simc and adding stuff onto it. But a few things prompted me to write a new one: One, unless I sat down and got someone to go through all of simc with me, it was actually easier to just write a new simulator from scratch. Two, everyone benefits from having two independent models of the game. Three, I already had a working simulator from way back in WotLK. I have also written two other simulators in the past, so it was much more efficient to just revive that work.

Our core engine, while similar, actually has a few improvements. Our fundamental model of how the player chooses actions is more realistic. simc does things that are literally impossible to execute in-game, and while the impact might be small overall... there's just no reason to continue doing it. We have done extensive comparison of our simulation results to logs, and we think it is closer to what a real world-class player is physically able to do within the confines of the game, and should thus give more actionable results.

So now we get to what AMR can do that simc can't... it's really tough to list it out without it just sounding like a PR laundry list... the differences are so vast at this point. The AMR simulator can:
- easily load and modify any character with a good UI
- let you modify any rotation, boss script, or spell directly on the website, and run it immediately without messing around with code
- keeps an infinite history of all simulations you have ever done, and lets you easily share any report with anyone on the internet by simply copying the URL
- lets you share your rotations or boss scripts so that other users can subscribe to them and get your changes in real time
- import characters directly from combat logs
- do advanced custom simulation batches and let you visualize or download the results
- distribute large simulation batches across multiple computers or server farms even, without having to do any significant setup
- create gearing strategies that plug into the AMR optimizer with the press of a single button
- has full support for tanks and healers (still working on adding some more stuff for healers in particular though)

As a long-time simulation user and theorycrafter, these are the features that I have wanted for years. And I happen to have the time and appropriate skills to do them, so... here they are, enjoy!

All of the arguments and differences between the simulators... it's all the configurable pieces. A lot of people latch onto the rotations... "simc has a better rotation so it is a better simulator." The rotations are designed to be changed in a simulator, and it takes five minutes to do it. Our simulator focuses on different boss scripts. If you want to make the two simulators match exactly, well... you need to use the same boss script! These are fundamentals that simulation experts get, but that the more "casual" participants ignore or overlook. Kind of blows my mind... people have been trained to think that the entire simulator is a set-in-stone black box, rather than something easily modified. And to a large extent... simc is a black box from the rotation on down. They can get pretty gnarly. Improving on that with a better UI isn't just "fluff", it is fundamental to doing better theorycraft and having a larger TC community.

The AMR simulator does a lot more stuff than simc from my perspective. (And it is also free.) It's open and transparent. It's almost absurdly powerful for figuring out how to play a video game. And I'm always looking for feedback to add more stuff to it -- just use it and hit me up. Is there something about simc that makes it better or easier to do what you want to do that I've missed? Just let me know -- that is why I poke my head out every so often. This is one of the most fun programming projects I've done in a long time, it is not hard to convince me to add something

Considering the "sim" rotation (healers can't be simmed ) for Discipline has Penance BEFORE Schism, still, after months of it being brought up, is rather comical.

http://www.askmrrobot.com/wow/theory...e&version=live

Bad inputs = bad outputs. This is part of the reason people can't trust AMR / have a bad perception of AMR. The only way I could see someone using this service is if they are brand new and doing easy content where min-maxing doesn't matter. If someone is going to go through any more effort of improving their character, AMR isn't really an ideal choice.

[Supliftz]

i remember when AMR suggested disc use PW:S relics :thinking:

[Swol]

Considering the "sim" rotation (healers can't be simmed ) for Discipline has Penance BEFORE Schism, still, after months of it being brought up, is rather comical.

http://www.askmrrobot.com/wow/theory...e&version=live

Bad inputs = bad outputs. This is part of the reason people can't trust AMR / have a bad perception of AMR. The only way I could see someone using this service is if they are brand new and doing easy content where min-maxing doesn't matter. If someone is going to go through any more effort of improving their character, AMR isn't really an ideal choice.

If there was an obvious bug in a SimC APL, you would just fix it and tell people to use the updated APL. You wouldn't say that the entire SimC project is bad and not worth using.

Just... change the Discipline rotation to do what you want. There is a "share" link at the top. Post your updated rotation on your blog/discord/whatever and tell people to use your rotation when they create gearing strategies. The entire AMR simulator project is set up to make it easy for theorycrafters to test their ideas and then share those ideas with their fans/readers.

I haven't put much work into tweaking the default healer rotations lately because I have a "version 2" of healing simulation in the works based on all of the feedback we have collected since legion launch. Once I finish that I will update all the rotations accordingly. It will be available well before Tomb opens up.

[MendUS]

If there was an obvious bug in a SimC APL, you would just fix it and tell people to use the updated APL. You wouldn't say that the entire SimC project is bad and not worth using.

Just... change the Discipline rotation to do what you want. There is a "share" link at the top. Post your updated rotation on your blog/discord/whatever and tell people to use your rotation when they create gearing strategies. The entire AMR simulator project is set up to make it easy for theorycrafters to test their ideas and then share those ideas with their fans/readers.

I haven't put much work into tweaking the default healer rotations lately because I have a "version 2" of healing simulation in the works based on all of the feedback we have collected since legion launch. Once I finish that I will update all the rotations accordingly. It will be available well before Tomb opens up.

I think my point is mostly that simming healing is a fools errand. Healers don't have "rotations" and healing sims aren't really possible in a way that would be beneficial. (Note, I am looking at this from entirely a healing perspective)

I'm all ears to something new; I'd be interested to see what you've been working on once it's released, but I'm still skeptical.

[Kilee25]

Re: simming healers.

Back in Vanilla, I wrote very barebones healing simulator on my computer for my own use. Bare in mind that the game was much simpler back then. The goal of the simulator was to try to find a set of gear that could cast the longest amount of time while keeping up renew and chaining greater heals. It would then spit out a set of stat weights providing for intellect, spirit, sp, and mp5. It would factor in how long it could cast without going oom, etc...

It was a very, very crude simulator, but the stat weighs I got out of it were much better than "nothing".

My point is that having "something" to go by is better than nothing. Even if it's flawed, as a theorycrafter you can use it as an additional data point in order to develop your ideas further. Having additional tools, to me, is invaluable.

I also agree completely about being skeptical and needing to go into it with an air of doubt, but I tend to think that's a good policy with everything...

[Yellowfive]

Yeah healers are a bit tricky -- we're going to try a different approach and see how it works out, feedback always welcome.

MendUS -- I understand that up until this point most people have thought healers "can't be simmed." But as a programmer and tool builder, I think that's based on a preconceived notion of what it means to "sim" a healer. Many people have made spreadsheets for healers. These spreadsheets do things like calculate how much heals will heal for given your stats, and use formulas to try and factor in buffs and procs and things like that. Some of them will take a healing breakdown from a log and try to weight things based on that, etc.

A simulator could be written to perform essentially the same tasks, but I believe it could do it more accurately and with less errors, and in a more verifiable manner. A simulator is essentially a "brute force" approach to coming up with a formula for how much a spell heals, on average, as a function of all your stats and stuff. Rather than try to use fancy formulas to determine how crit interacts with mastery interacts with buff X and buff Z... we just... do it, and average the results. Less assumptions are made, and thus there is less chance for things to go awry. That is one of the major motivations for simulation in general.

Spreadsheets are also notoriously hard to verify... looking through someone else's complicated spreadsheet formulas is, for all practical purposes, an impenetrable black box. A simulator is made up of several very simple pieces: spell X works like this, spell Y works like that, this much rating increases your crit by Z%, when put together, here's the output. It is much easier to find bugs this way, and verify that the results you are seeing can be believed.

So at this point, it's a matter of adding features to get the healing simulation to model what we want. We have been modeling how DPS players think for years now... I don't think that it is impossible to model how a healer thinks. Most healers I would argue, follow a relatively simple set of rules when they are playing. We just need to beef up the simulation to give that healer rotation logic the input it needs to execute those rules.

I think that a lot of people get hung up on the idea of trying to model every little piece of damage that every player in a raid would take, and model how every player is behaving and dodging (or not dodging) damage, etc. I don't think that level of detail is necessary to get useful results. We can remove some of that "noise" and treat it as a statistical phenomenon. All WoW simulation removes at least some variables to keep the simulation tractable. We don't model every possible eventuality for DPS specs like... someone on your team dying unexpectedly, the tank moving a boss out of your range when it is technically not necessary, random small player mistakes, etc. The idea is that we can get good information without worrying about every detail like that. (We DO think that it is important to have more realistic boss patterns for DPS specs though, which is why we have reintroduced that with our boss scripting of actual bosses. But the idea there is to get the "major" pattern of the fight, not every possible detail.)

Anyway... "show don't tell" right? We'll make an announcement when the new stuff is ready, and you can check it out. The main thing I ask is that you approach it as a tool, or an additional piece of input for doing healing theorycraft. Really try it, don't just do a 5-second simulation with the defaults and say "lol not what I think I already know." Use it, configure it, look at the results, and give it the benefit of the doubt: if it is different than what you already assume, why is it different? Is this actually some new and useful information, or is it just an artifact of how the simulation was written that doesn't reflect real game situations well enough? Then give me feedback either way, and we'll continue to refine it.