Best in Bags Adaptive Strategy Optimizations (Havoc)

I’ve been making some manual tweaks to optimize my BiB results from the Adaptive Single Target strategy for Havoc lately. Some of these optimizations are based on what I believe is a limitation of the Adaptive strategies. Some appear to be some minor nuances of stat allocations. All BiB recommendations and tweaks in this post were done using the Adaptive Single Target strategy on 10/21/2017 around midnight CDT.

First things first, here is my AMR export string:

$54;US;Dragonblight;Hadronicus;Righteous;6;1;110;13:35,8:800,2:800,14:21;1;.s1;4;2222311;1000,1,1,1,1,1,1,1,1,2,1,1,1,1,1,1,314,32,131,1,1,1,134;4,4,4,4,4,4,4,4,4,1,1,1,1,1,1,1,1,1,4,1,1,16,1;137407b1582b3337b3418,142310b1567b3337b3418,137476b1582b3337b3536;1,1739,40,4,-290,-491,-2,1,1,0,0,0,0/2,739,39,-8,-769,2,4,1,1,0,0,0,0/3,732,39,6,-780,489,-487,1,1,0,0,0,0;.s2;5;3222133;1096,1,1,1,1,1,127,1,1,1,1,1,1,1,1,92,35,71,63,1,1,1,131;1,1,1,4,4,4,1,4,4,4,4,4,4,1,1,1,1,1,4,1,1,15,1;137326b1572b3336b3418,147076b1512b3337b3562,151008b3179b3336b3397;1,1739,40,-4,-543,265,-267,1,1,0,0,0,0/2,509,40,3,-553,268,-398,1,0,1,0,0,0/3,640,40,-9,-669,132,-133,1,0,0,0,0,0/151011b3189b3337b3396,639,41,-10,-540,-1,4,0,0,0,0,0,0/141255b1572b3528b3573,506,41,-10,-539,266,-265,0,0,0,0,0,0/140825b1507b3528b3573,507,32,13,-683,131,-1,0,0,0,0,0,0/140825b1507b3528b3573,508,41,-10,-540,2,-133,0,0,0,0,0,0;.q1;121299s6b1582b1755b197;6530s16b-2815x137407y4903z-4834;1s17;4622s12b1092b1759x14104e5427;1795s14b-2967b969b236b1720b45x0;212s9b-3533b1567b1730b81;32s2b-3378b1542b1755b199e10;51s11b-1969b241b1610b110x0e-10;2794s5b-1921b1730b81;1769s8b-1821b1740b81;5174s3b-1607b1759e456;2730s13b-2058b1825b225;119s15b-2060b1834b226e-448;1s10b-2065b2031b34;1s1b-2040b1814b227;1s7b-3520b1454b2031b34;.q2;121310s14b607b965b236b1720b45x151583;7401s13b-2972b78b10;120s17;1s16b32x-14257y9750z3932;3612s2b2738b111x572e5437;8s12b-1759b1759x0e-10;1958s15b-1988b1755b199e8;49s9b-3496b1567b1730b81;2877s5b-1811b1730b81;1769s8b-1821b1740b81;7915s11b-1896b1814b227e-7;109s10b-2066b2031b34;1s1b-2040b1814b227;1s7b-3520b1454b2031b34;1s3b-2065b2031b34e455;215s6b-2055b2021b34;.r;_;.inv;6948;14929;11593;9632;29886;14424;1700;4083;11133;11940;4892v110b1716b96;150b-1205b965b236b1720b45x151583;21;2538;49;1;186;2;6;0;0;2;1;1;1;1;2;0;3;0;313;0;0;0;0;1;0;0;0;1;0;201;2364;840;0;4;863b-2972b78b10;121b32x-14257y9750z3932;52b-120b68b20b1011b1708;3b-2808b69b20b989b1730;0b-2807b68b20b1005b1714;0b-2808b69b20b1001b1718;0b-2807b68b20b987b1732;962v0b-2716;46v0b0;2549b2767b111x572e5437;1126;7;0;0;16;23b-1993b1760b198;2b-1998b190b1801;19b-3488b1522b1775b172e-10;42b-1952b1780b81;297b41b111;181b-3528b1530b1956b45;36b-1981b1745b199;1b-1939b1740b81;9b-2814b958b1775b236;76b-1996b1759b237;7b-3531b1550b1745b199e456;87b-1949b1750b199;37b-1954b1755b199e-448;19b-1949b1750b81;13b-1881b190b1801;7b-1956b236b1528b82;41b-1846b1764b237;1852;191;121;384b-1762b1759x0e-8;23b-1759b1759;5b-111b70b41e8;24b-111b111;1b-111b70b41;341b-2008b246b1720b45;77b-1981b1745b81;3b-1881b190b1801;21b-1966b1774b199;474;368b-2048b2029b12e0;0b-2056b2042b14e0;1b-3488b1447b2029b12;0b-2041b2029b12;1b-2041b2029b12;1b-2021b1830b179;0b-2044b2042b14;1b-2056b2042b14;1407;0;15;1;7;3;0;134;232;14;395;0;209b-2021b2021b45;6b-3532b1476b291b1528b237;4b-2096b2051b45;2b-2081b2025b11;17b-2021b2021b45;0b-2066b2021b45;70b-3532b1446b2029b12;117;243b-1956b1956b45;96;131b-2101b1994b62;51b-1996b1804b237;50b-2071b2026;22;47;307;1;2;155;22b-3487b1511b1785b81;2b-1871b1790b81;26b-1921b1839b237;5b-2046b281b1610b110;33b-1991b1800b197;121;9;1;217;0;1108;188;443b-75b111;184b-1988b1755b199e0;2b-1979b251b1529b81x-21333e-8;2279;146;2;1;1;1;1;14;88b-1921b2031b34;1b-2065b2031b34;3b-2080b326b1720b33;5b-2039b1814b227e1;1b-2066b2031b34;3b-2065b2031b34;6b-2055b1830b224;4b-2054b301b1720b34x21336;5b-2065b2031b34;1b-2065b2031b34;69b-2065b2031b34;2b-2055b2021b34;16b-2065b2031b34;1b-2095b2061b36e7;2b-2067b2031b34;2b-2075b1849b225e448;0b-2064b2031b34e0;62b-2060b1834b226;0b-2065b311b1720b34x-21336e-453;63b-1820b1422b172b237;90b-2066b2021b34;8;96;60;358;1;197b-2020b1986b45;1597b-3531b1530b1956b45;2b-3532b1531b1956b45;858v0b16b1e-4;465b-2013b1759b237;17b-374b138b60;0b-223b223b131;2b-354b223b131;3b-359b167b59;0b-206b148b59;122;433;0;0;0;1;0;1;71b63b111x21333e1;14;708b-2098b2056b86;83;2;289;2;0;0;1;0;1;1;0;3;1;42;13;0;0;132;4;63;25;91;9;1;11;1;0;0;1;1;1;0;1;0;0;1;2;0;1;1;0;1;0;2;0;0;0;3;0;1;2;0;1;1;0;0;0;1;32;1;13$

Here are the simulation reports showing a net ~6.2% DPS increase from manually overriding some of the Adaptive BiB suggestions:

Original BiB recommendation - 1,378,097 DPS

Swap Soul of the Slayer for Anger of the Half-Giants - 1,396,154 DPS
Note: this is the setup that is currently being recommended for me by BiB today using export string above.

Edit: Sorry for removing clean formatting here, the forums would only let me include 2 links since I’m technically a “new user”. Maybe a mod can clean this up for me?

Swap Cinidaria, the Symbiote for Delusions of Grandeur
not_Not Available

  • 1,444,130 DPS
    Note: this swap actually changes 3 gear slots because switching from a legendary belt to shoulder would break the T20 4 piece set.

Swap Vial of Ceaseless Toxins for Crit Stat Stick
not_Not Available

  • 1,453,576 DPS

Swap 930 crit/mastery boots for 935 haste/crit boots
not_Not Available

  • 1,463,205 DPS

So, the first thing that stands out is that Delusions of Grandeur are responsible for about half of the overall increase in DPS between the lowest and the highest sim. This is the part that I would expect the BiB optimizer to have a difficult time with. I am not sure what fight duration was used to produce the adaptive strategies, but it is most likely a duration that does not favor this particular legendary.

In the sims that I ran, DoG was always sufficient to produce a 3rd Meta cooldown that I otherwise would not have been able to use for a fight duration of 320s +/- 16%. I used this duration based on my guild’s most recent Mythic Goroth kill. Prior to this week, I had not done a Mythic Goroth and I was running the same set of sims to improve my BiB results with a 170s +/- 16% sim to represent our most recent Heroic Goroth kill. In that situation, DoG provided 2 meta uses instead of 1, so I got similar results by locking in that particular item.

Aside from that, BiB seems to be slightly overvaluing the Vial of Ceaseless Toxins. I’m not really sure why that is, but I don’t really have a problem with it. I actually think this trinket is my best option on any fight that has adds that die quickly enough to proc the cooldown reset on the trinket. For pure single target though, the crit stat stick is outperforming it in the sim and likely would in practice as well because it doesn’t require any play input to optimize.

The adaptive strategy is also slightly favoring mastery over haste for me right now. Mastery is still a very low value stat for Havoc. I’m sure haste begins to lose effectiveness once we get close to being effectively Fury capped by GCDs. With over 50% crit and 19% haste, I may actually be fairly close to that point. However, running these sims is still showing haste slightly ahead of mastery.

One other thing worth noting: a few of these sims have no gem in the Anger of the Half-Giants ring. The export string I used during these sims was made immediately after I got that ring, before gemming and enchanting it. The export string above was made after gemming and enchanting the ring, but should be identical otherwise. The BiB tool was not always putting a gem in the ring after I updated my locked slots. I’m not sure why this was happening. The missing 200 stat gem may have more to do with some of the minor DPS changes between sims than the actual gear changes themselves.

The biggest takeaway that I get from all of this is that the BiB optimizer might benefit from an ability to customize the fight duration for which you are trying to optimize gear. For Havoc Demon Hunters, DoG is among the best legendaries in situations where it can get you an extra use of Metamorphosis and it is among the worst in situations where it cannot. I know other classes likely have similar situations occurring with the Convergence of Fates trinket. Allowing some customization on fight duration for BiB might help people build different gear sets for different fights to take advantage of these items when they are at their best and swap them out when they are at their worst.

You are right that the fight duration matters significantly with cooldown reduction effects. The strategies are run with default fight lengths set in the boss scripts. For Goroth that is 270 seconds (4.5 minutes). That is a typical “progression” fight length for a single target fight.

If I run the BiB suggested gear vs the gear you manually tweaked at the fight length used to calculate the adaptive strategy, I see these results:
BiB [1440k DPS] (Not Available) vs Manual Tweaks [1474k DPS] (Not Available)

The difference here is 2.36% DPS. Definitely on the highest side of what we are trying to achieve (2% is our goal), but, still a good overall suggestion. Just using the new “customize” feature and locking in your preferred legendary items, the DPS ends up being [1465k DPS] (Not Available), which is very close to the set you manually tweaked.

Havoc is particularly difficult to get the margin of error low on because so many of the legendary items are close in value, and the T20 set bonuses interact very strongly with particular talents, making it complicated on the back end.

The difference in DPS between the Vial and stat stick is very small. Ranking an item with a proc vs an item that just has stats requires us to combine our predictive model that we build for stats with our cached simulation results for the item with a proc. Getting that ranking “perfect” is just not going to happen. In this case we can see that if we simulate a set of gear with Vial vs the other trinket and no other changes we see 1465k DPS vs [1476k DPS] (Not Available), a difference of 0.68%. I’d consider that a very good relative ranking of those two items, considering they were ranked instantly without doing an actual simulation.

Behind the scenes, we have created a predictive model that will estimate the DPS of any given combination of items. That is how we examine every single combination of gear in your bags in a few seconds. That is also why there is necessarily some “fuzz” with the whole process. We trade a little margin of error for speed. It is our opinion that in-game, you will never be able to effectively notice a 2% difference in theoretical simulated DPS between two sets of gear. For those who don’t agree with that, we made it quick and easy to simulate specific sets of gear, as you have done. We feel that is the best compromise available.

Customizing the fight length on adaptive gearing strategies will be impossible, unfortunately. You would have to run a whole new adaptive gearing strategy for each fight length. For CDR items, I think it would be best to run some simulations with those particular items at fight lengths that you think are relevant to your raid team’s typical kill times to see if they would be better to equip or not. I know it’s not as nice a solution as pressing a button and getting an answer, but we’re a bit limited in this case by raw compute time/cost.

I appreciate the detailed write-up of your issue.