Thread: Simcraft guide

Page 1 of 7
1
2
3
... LastLast
  1. #1
    The Lightbringer Darkfriend's Avatar
    10+ Year Old Account
    Join Date
    Mar 2011
    Location
    Wisconsin
    Posts
    3,285

    Simcraft guide

    Since now is one of the best times in an expac to learn how to improve ones knowledge of a class, in preparation of a new tier, I wrote up a simcraft guide for one of my guilds. It is here. Please provide any feedback, if anything is unclear, and most importantly if anything is incorrect.

    Hope this helps.

    Simcraft is an amazing tool in the arsenal of the serious raider. The ability to derive accurate stat weights for a variety of fight types and lengths, and then utilize those weights to perfect your character is of great benefit. However, a caveat. As with any tool, the usefulness is limited by user knowledge and problems which exist inside the tool itself. This guide is an effort to aid in the learning of how to get stat weights and use a tool such as AskMrRobot to min/max. As it is a first draft, errors in the guide may exist, and I would appreciate any being pointed out so I can correct them.

    As well, note that errors in Sim-c individual class/spec modules *may* exist. This will depend on how well your class can be modeled, and how good those who edit the modules are at what they do. For example, the fury module is amazing. The DW one is incredibly solid, but Unholy, due to the nature of the spec isn't quite as accurate for true DPS results, but is solid on stat weights. Likewise there is currently a "feature" on AskMrRobot where Amp trinket weights are being increased even on custom weights, while Sim-C has already done so, effectively double dipping. Be aware if anything incredibly odd seems to be going on. It may be a bug.

    On to the fun stuff! First step is to make sure you have the latest version of Simcraft. Download it here. http://simulationcraft.org/ After doing that, unzip it and open the application named SimulationCraft.

    WELCOME!



    Head over to the Import tab, and find your character.





    Make sure everything is right for the fight type you want to simulate. Check glyphs/talents/gear/trinkets, etc. Now click "Import!"



    This page is the meat and potatoes of Simcraft. It's the operations list for actions. Although these can be edited to test different things out, Simcraft can be buggy if you don't phrase things correctly. Basically, don't change anything here unless you know what you're doing. Look but don't touch!



    Next step is to setup the parameters for your specific sim. To do that click on the tab "Options". Those options with arrows are the most important to adjust. Generally 10k iterations is enough, fight length should be adjusted to mirror length of typical fights, as should be fight type. Personally I use 450 seconds near the start of a tier, which is 7.5 minutes, and 300 seconds near the end, which is 6 minutes. Fight types I generally juggle between patchwerk and light movement for the classes I play. If you hold your curser over the fight style drop down menu, it will describe what each type of fight changes. You can also add threads if you have a PC which can handle it to increase speed.





    These options will give you average DPS over 10k iterations for that fight length and style. There is one extra step for stat weights. (I'm assuming all buffs/debuffs will be present in 25m. Those can be changed in buffs/debuffs tab)

    For weights head over to "Scaling". Shown is an example of what I sim for a duel wielding strength class. You can run all, but it takes much longer and most of those numbers aren't relevant.



    And that is it! Hit "Simulate" in the bottom right, and wait! You'll get a page such as this once finished.





    Normalized numbers are simple normalized numbers where your primary stat is always 1. Those are useful for determining upgrades in your head through napkin math. (In the case of higher item level but worse itemized loot. Is it an upgrade? You can use those numbers to determine the "value" of each piece to decide)

    Once you get those numbers take a look at the "Scale factor" numbers. Those are what we want. Head over to http://www.askmrrobot.com/wow/gear. Import your character. Hit "Edit Weights". Type in the results from your Scale Factor page. Hit "Save". Hit "Optimize." Fix anything that is off that *Does not look sketchy*. AMR is wonky at times.

    UNEQUIP THOK'S TAIL TIP BEFORE OPTIMIZING.

    AMR is increasing the value of mastery by the amplification amount if it sees you have on a Tail Tip, so if you have on a tail tip it's increasing the value by an extra ~7-9% ABOVE what it should be, which will skew results. Same for haste, although that really shouldn't impact much if anything.







    And there you go!
    Last edited by Darkfriend; 2014-02-15 at 03:26 AM.

  2. #2
    Herald of the Titans
    10+ Year Old Account
    Join Date
    Feb 2013
    Location
    Germany
    Posts
    2,503
    This actually deserves a sticky and should be in a forum for every class..
    Quote Originally Posted by Ryan Cailan Ebonheart View Post
    I've done nothing wrong. I'm not the one with the problem its everyone else that has a problem with me.
    Quote Originally Posted by MilesMcStyles View Post
    I don't care that other people don't play the content that I enjoy.

  3. #3
    The Lightbringer Darkfriend's Avatar
    10+ Year Old Account
    Join Date
    Mar 2011
    Location
    Wisconsin
    Posts
    3,285
    Well it's pretty rough ATM. There are things that need to be added for other classes (plot points, editing list, syntax guide.) It's more a basic "How to sim and use the results to optimize" guide currently. Any feedback let me know, and if it helps let me know!

  4. #4
    Stood in the Fire Mogai's Avatar
    10+ Year Old Account
    Join Date
    Feb 2012
    Location
    Sweden
    Posts
    383
    i'lltake this opportunity to thank you for this guide!. i believe you posted a shorter version in another thread a few months back. really hepled me get my stats correct!

  5. #5
    inb4 "simcraft isnt accurate". It's only as accurate as you define the latency, fight style, and action priority list. I religiously use simulationcraft, and I tweak the default action lists, or make them from scratch, to get the best possible numbers for myself.

    FWIW, I'm really anal about having accurate stat weights. Anything less than 1 million iterations is useless to me. A stat weight sim usually takes about 3-5 hours for me, depending on the complexity of the action list and the amount of stats I find weights for, though having a 6 core i7 @ 4.6GHz does help the computing speed some.

  6. #6
    Awesome guide Dark, well written and well worth a sticky!

    Quote Originally Posted by Ssateneth View Post
    inb4 "simcraft isnt accurate". It's only as accurate as you define the latency, fight style, and action priority list. I religiously use simulationcraft, and I tweak the default action lists, or make them from scratch, to get the best possible numbers for myself.

    FWIW, I'm really anal about having accurate stat weights. Anything less than 1 million iterations is useless to me. A stat weight sim usually takes about 3-5 hours for me, depending on the complexity of the action list and the amount of stats I find weights for, though having a 6 core i7 @ 4.6GHz does help the computing speed some.
    just...why?

  7. #7
    Quote Originally Posted by Ssateneth View Post
    inb4 "simcraft isnt accurate". It's only as accurate as you define the latency, fight style, and action priority list. I religiously use simulationcraft, and I tweak the default action lists, or make them from scratch, to get the best possible numbers for myself.

    FWIW, I'm really anal about having accurate stat weights. Anything less than 1 million iterations is useless to me. A stat weight sim usually takes about 3-5 hours for me, depending on the complexity of the action list and the amount of stats I find weights for, though having a 6 core i7 @ 4.6GHz does help the computing speed some.
    Yeah that is over kill to the extreme. You are not doing yourself any favors but it is of course your time. You will see differences between the two but they won't be nearly large enough to make a difference. I have heard people recommending to go to 50,000 iterations, but again you shouldn't see any significant changes. You may see a few more uses of an ability, but you have to remember with 50,000 + iterations you are going to see variations that you wouldn't always see, simply because you are seeing so many more variations. This doesn't necessarily mean more accurate.

    I do think is something that is much easier to see with a caster class, as cast times can become variable based on things like Haste procs and spells can proc other spells which change your rotations up. Warriors, and most Melee classes by contrast are pretty static with that regard. Regardless of how many times you run the test you are only going to get so many possible GCD's, at most you will see a little bit higher rage generation which could lead to a few extra abilities used. Or you could see less rage generation and less abilities used.

    Keep in mind of course, it's all RNG. Regardless of how many times you run it, all you are doing is putting a law of averages behind the RNG, and the results should always be taken with this in mind. Feel free to experiment at your leisure but I believe you will be very hard pressed to find many people who will recommend going higher than 50,000 at the most.

    - - - Updated - - -

    While I am on this topic;

    To speed up your simulation, go ahead and uncheck Analyze Attack Power, Weapon DPS and Off-hand Weapon DPS/.

    As seen here:


    Reason:
    Currently we have no items that add Attack Power, so analyzing it doesn't do us any benefit, we have no choice between Str and AP gear.

    Likewise, its neat to see the impact Weapon/off-hand DPS has on our damage, but the gain from DPS is so high, that higher ilvl weapon will almost always be worth using.
    I say almost always because it is possible that in extremely unlikely circumstances (using a higher DPS Tank weapon that has +Armor or some other non-DPS stat on it?), that you could gain more out of the extra stat than you could out of the extra DPS. However, under current loot/stats, that isn't even possible, so its a non-concern until loot like that starts appearing.
    As is, until they start making +weapon damage gems/enchants again, there is no benefit to simming the weight, since it doesn't compete with other weights. Either we get more weapon damage from a higher ilvl weapon, or we do not. We never have a choice of "more weapon damage vs more Str/Haste/etc".

    The gain:
    Less stats to sim means quicker simulations.
    Last edited by Archimtiros; 2014-02-08 at 11:25 PM. Reason: Redundancy

  8. #8
    Dreadlord sjsctt's Avatar
    10+ Year Old Account
    Join Date
    Feb 2013
    Location
    Atlanta, GA
    Posts
    935
    Advanced section for seeing how to change gear in the CharDev incoming ?

  9. #9
    Quote Originally Posted by sjsctt View Post
    Advanced section for seeing how to change gear in the CharDev incoming ?
    It's a WIP.

    Depending on what changes in the game, Simcraft and AMR go through next expansion this will be rolled into my new Arms and Fury PVE guides.

  10. #10
    The Lightbringer Darkfriend's Avatar
    10+ Year Old Account
    Join Date
    Mar 2011
    Location
    Wisconsin
    Posts
    3,285
    Quote Originally Posted by archimtiros View Post
    it's a wip.

    Depending on what changes in the game, simcraft and amr go through next expansion this will be rolled into my new arms and fury pve guides.
    ok jalopy 2.0

  11. #11
    The Lightbringer Tzalix's Avatar
    10+ Year Old Account
    Join Date
    Mar 2011
    Location
    Sweden
    Posts
    3,118
    Quote Originally Posted by Archimtiros View Post
    Yeah that is over kill to the extreme. You are not doing yourself any favors but it is of course your time. You will see differences between the two but they won't be nearly large enough to make a difference. I have heard people recommending to go to 50,000 iterations, but again you shouldn't see any significant changes. You may see a few more uses of an ability, but you have to remember with 50,000 + iterations you are going to see variations that you wouldn't always see, simply because you are seeing so many more variations. This doesn't necessarily mean more accurate.
    I always run 50k iterations. I'm just a bit of a perfectionist.
    "In life, I was raised to hate the undead. Trained to destroy them. When I became Forsaken, I hated myself most of all. But now I see it is the Alliance that fosters this malice. The human kingdoms shun their former brothers and sisters because we remind them what's lurking beneath the facade of flesh. It's time to end their cycle of hatred. The Alliance deserves to fall." - Lilian Voss

  12. #12
    The Lightbringer Darkfriend's Avatar
    10+ Year Old Account
    Join Date
    Mar 2011
    Location
    Wisconsin
    Posts
    3,285
    Quote Originally Posted by sjsctt View Post
    Advanced section for seeing how to change gear in the CharDev incoming ?
    Probably. As well as tricks and stuff.

    Example, on this line


    The talents= line, 0122120 etc, you can change those to change talents without logging in and out. 0= first, 1 is second, and 2 is third in each line.

  13. #13
    Quote Originally Posted by Ssateneth View Post
    inb4 "simcraft isnt accurate". It's only as accurate as you define the latency, fight style, and action priority list. I religiously use simulationcraft, and I tweak the default action lists, or make them from scratch, to get the best possible numbers for myself.

    FWIW, I'm really anal about having accurate stat weights. Anything less than 1 million iterations is useless to me. A stat weight sim usually takes about 3-5 hours for me, depending on the complexity of the action list and the amount of stats I find weights for, though having a 6 core i7 @ 4.6GHz does help the computing speed some.
    simcraft isn't necessarily accurate, but it's the best we have.

    the options for fight styles are hardly all encompassing and you assume that the person modeling your specific class and spec has accounted for everything, which i find sort of doubtful.

    you spend 3-5 hours on a million iterations so you'll get perfect stat weights but no fight perfectly fits into the example fight styles or exact circumstances and (although i don't have any evidence to prove the models aren't accurate) you assume the models are perfect, and all of these factors will have an effect on stat weights at the end of the day.

    i remember at the end of cataclysm when someone realized heroic leap wasn't getting a benefit from single minded fury (which would have affected stat weights btw); how do you know there's absolutely zero chance of anything else like that right now?

    anyway, all i'm saying is that simcraft is a super useful tool to get stat weights but you can't pretend that it's perfect and there's no chance it's wrong.

  14. #14
    The more people use Sim-C (or any tool for that matter) the higher the chance is that such issues can be found and then taken care of. We have a lot of minor bugs or issues that we've needed to fix over the past couple of months, and it's constantly improving, due to the fact that more and more people start to use it and experiment with it.

    Regarding different encounter models, you can make your own models if you know how to do some coding. I know Collision made a model for Galakras and Garrosh a while back to be able to help us figure out which trinket combination would be optimal for those fights and it worked out well enough. We've sense been able to test the things out on live and see that the models where close enough to the truth to be able to give accurate information.

    The amount of iteration you need to do in order to get statistically worthy results are between 10,000 and 50,000 and therefor those are the "standard" in Sim-C, anything less and you'll get uncertain results, and any more than that is just overkill. Sure, you "may" get a 0.001 change in some of the stats values, but the likelihood that you'll get some different results are slim at best. If you feel like you need more than 50,000 iterations in order to feel certain feel free to do more, it's your own time and CPU power you're using.
    My Stream
    NollTvåTre Looking for Raiders

  15. #15
    Quote Originally Posted by Eranthe View Post
    simcraft isn't necessarily accurate, but it's the best we have.

    the options for fight styles are hardly all encompassing and you assume that the person modeling your specific class and spec has accounted for everything, which i find sort of doubtful.

    you spend 3-5 hours on a million iterations so you'll get perfect stat weights but no fight perfectly fits into the example fight styles or exact circumstances and (although i don't have any evidence to prove the models aren't accurate) you assume the models are perfect, and all of these factors will have an effect on stat weights at the end of the day.

    i remember at the end of cataclysm when someone realized heroic leap wasn't getting a benefit from single minded fury (which would have affected stat weights btw); how do you know there's absolutely zero chance of anything else like that right now?

    anyway, all i'm saying is that simcraft is a super useful tool to get stat weights but you can't pretend that it's perfect and there's no chance it's wrong.
    Very true you should take everything with a grain of salt, but going to back up what Warriorsarri (who has done a ton of work optimizing SimC for us btw) said above.

    Simcraft is obviously a simulation. It runs off of pre determined variables which makes it failable, but it does take a certain amount of error into consideration. For example it can be set to randomly miss ability usage to simulate player error. It obviously isn't 100%. As a player you may tend to screw up your WS/HS more often but get your CS phases near perfect because you know that is the most important part and what you focus the most on.
    Ultimately it doesn't matter though, it simulates a deviation from "perfect" which accomplishes the same goal. I wouldn't say Simcraft is perfect, but wouldn't call it wrong either. The goal is to make it "realistic". With enough tuning in the settings that is pretty doable, though even at default it is close enough that the minor deviation won't change too much.

    It is a tool like any other, and it is only as good as the work that is put into it. Warriors have one of, if not the best Simcraft profiles, and are a fairly static class to boot, which means ours will be pretty damn accurate vs many other classes that don't get much work put into them. This is mostly due to having a really good core of Warrior theorycrafters and testers.

    The more work the profile gets, and more people that use it the less bugs and mistakes we will find.

  16. #16
    To add to what Archimtiros said, if you have work put towards our Sim-C module, or any other work that you would like to help make our theorycrafting even better, don't hesitate to contact either me, Arch, Darkfiend or Collision.

    The more we are that's working on improving the quality of our Theorycrafting community, the stronger it will become, and by improving the quality of our tools and maths, the bigger the chances are that we'll be able to influence the decisions that Blizzard will make regarding the warrior class in the upcoming future in such a manor that would be beneficial for all of us.
    My Stream
    NollTvåTre Looking for Raiders

  17. #17
    Just for shits and whatever, here are the bugs that I'm aware of for Fury:

    Storm Bolt: FIXED
    In Game: Rolls once to determine if MH attack hits, and if that is successful it rolls again for the off-hand attack. Possible for MH to hit and OH to miss, but not for MH to miss and OH to hit.
    Simcraft: Rolls separately for MH/OH.

    How does that affect us? Only matters if you're not main hand capped on hit/expertise... which is rather unlikely at this point in the game. If you're under exp/hit cap, simcraft will very slightly undervalue additional hit/expertise. I'll fix this sometime soon.

    Raging Blow Meat Cleaver: FIXED
    In Game: If the main target RB fails to land, all will fail. Otherwise, everything will land.

    Example:
    3 targets, target 1 is selected for the actual raging blow attack, the game will run a hit test on all targets that will be getting hit by raging blow.
    Target 1: Hit
    Target 2/3: All parried

    Even though the game throws parrys at you in your combat log, the ability actually hits all 3 targets.


    However, if the hit test fails on the target you are actually using raging blow on, it will not meat cleave whatsoever.

    The purple "Badge" on the right shows that meat cleaver was active on myself when that event happened.

    Simcraft: Duplicates RB into 2-4 different raging blows, that can all separately fail/succeed.

    How does that affect us? Average-wise it doesn't, but it does mean that our aoe dps will be slightly more rng/variable than simcraft says. Probably won't fix until we know more about WoD.

    Whirlwind: FIXED
    In Game: MH/OH are not linked, determine hit/miss separately on each target.
    Simcraft: MH/OH are linked, performs a single check on both attacks, if successful hits all targets, if fails misses all targets.

    How does this affect us? Inflates value of expertise/hit pre-cap. I actually meant to fix this, just never got around to it... I'll probably do it this weekend.

    AFAIK, everything else is modeled about as well as it could be. Due to Blizzard removing hit/expertise as a stat in the game, they'll likely revamp linked attacks to be more consistent, and they will all be fixed in a future build.

    Arms doesn't care about OH/MH interactions so it doesn't have to worry about all of this, same for prot.

    The SMF action list probably needs some tweaking overall, as I haven't done much with it this tier.
    Last edited by CollisionTD; 2014-02-13 at 10:02 AM.

  18. #18
    Quote Originally Posted by Warriorsarri View Post
    To add to what Archimtiros said, if you have work put towards our Sim-C module, or any other work that you would like to help make our theorycrafting even better, don't hesitate to contact either me, Arch, Darkfiend or Collision.

    The more we are that's working on improving the quality of our Theorycrafting community, the stronger it will become, and by improving the quality of our tools and maths, the bigger the chances are that we'll be able to influence the decisions that Blizzard will make regarding the warrior class in the upcoming future in such a manor that would be beneficial for all of us.
    i don't think i'd be able to contribute, but is there a place where you all discuss theorycrafting in semi real time?

  19. #19
    We have a skype chat in which we are doing the main bulk of our work, but in order to not clutter it more then is needed we only have the people doing the work on there.
    My Stream
    NollTvåTre Looking for Raiders

  20. #20
    Quote Originally Posted by Warriorsarri View Post
    We have a skype chat in which we are doing the main bulk of our work, but in order to not clutter it more then is needed we only have the people doing the work on there.
    It's super exclusive, we have a secret handshake and everything.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •