Theck's Simcraft Thread - WoD/6.x

Also, as a side note, if you guys want to optimize the prot gear profiles for me, be my guest. I find that sort of gearing exercise to be a bit of a slog, so I'm happy to let other people do that while I investigate other matters or code features.

I outsourced our gear sets to someone else, so again, I have no idea if they're properly optimized. In practice it doesn't make a huge difference in sims, but probably still worth doing for completeness.

theckhd wrote:The default gear profile I've been using in sims here was put together back before we were aware that haste was weak. I just never bothered to update it much, because it mostly affects stat weights, and I shifted my attention to L90 sims. Once I get back to working on L100/T17 sims, I'll probably update it to be a more mastery-heavy set.

That said, I have no idea why you'd want to stack crit - I don't think we've seen it outperform mastery.

Your right, I should switch to mastery. I'll do another comparison today and report back results.

theckhd wrote:That's not a problem with the importer, it's a problem with your talent input string. The battle.net talent format uses 0/1/2 to indicate the three talent columns, with (I think) a "." to indicate no selection. So e.g. ..2..1. would be just Sacred Shield and Light's Hammer chosen.

SimC's talent strings use 1/2/3 to indicate the columns, with a 0 meaning no talent selected. So "talents=0030020" would be the same build as above. SimC is smart enough to interpret the battle.net format as long as it's part of a URL, but it expects you to use the 0-3 format if you're using "talents=", which is what the automation tool is doing.

Ah...now that makes sense. Thanks.

theckhd wrote:How are you measuring effectiveness in this case? I'm a bit skeptical that 125 bonus armor is even a 1% difference, let alone 2% or 4%.

It was a 6% TMI loss, a 4% DPS loss, and a 3% HPS loss. Thats why I said 4%. This was before the bonus armor included in the set. Removing the mythic and replacing with LFR resulted in a 4% loss.
Once I added in the shirt bonus armor item, TMI became 4% loss, DPS 2% loss and HPS 1.5% loss.

theckhd wrote:

Thels wrote:

theckhd wrote:Footer:

Code: Select all

enemy_initial_health_percentage=20


(this just keeps us in execute range the entire time, so we get a larger variation between results and thus more accuracy)

Whenever I try that myself, I get the following error:

Simulation setup error: Unknown option 'enemy_initial_health_percentage' with value '20'
Simulation failed!

Well, it worked back when I performed those simulations. I'll have to check and see if someone has removed that option. What revision of the sim are you using?

simc-603-2-win64

I'm working my way through step by step optimizing Ret for current L90 content, running into hurdles all over the place, but trying to get all this under my belt so I can make a much smoother approach for when L100 hits.

Another problem I had run into was AMR not exporting SimC profiles correctly for the new SoO gear, but they fixed that now, and switched entirely from trying to provide SimC with the correct stats to just providing the item ids, bonus ids, gem ids and enchant ids.

I'll see what I can do about level 100 stuff later on, hopefully before WoD actually releases. :P

theckhd wrote:Also, as a side note, if you guys want to optimize the prot gear profiles for me, be my guest. I find that sort of gearing exercise to be a bit of a slog, so I'm happy to let other people do that while I investigate other matters or code features.

I outsourced our gear sets to someone else, so again, I have no idea if they're properly optimized. In practice it doesn't make a huge difference in sims, but probably still worth doing for completeness.

So I'm interested in doing this however I have a question....
When using the automation import, it seems to put every paladin against the same boss. Could we get an option to split the automation actors among X simulations and auto queue them? I'm expecting quite a lot of combinations if I test every item in every combination (Excluding gems and enchants).

Or better yet, I can split them into multiple files, is there a command line I can run to feed these files into the simulation and have it output the results.html? There is about 1.4 million combinations of gear for us pallys and that only uses 1 ring and 1 trinket.

---edit:
I'm dumb, I should have looked at the documentation first. I'll try the commandline out.

The whole point of the automation tool was to generate multi-actor sims so that you wouldn't have to make a bunch of different sims and run them all separately.

That said... the proper way to optimize gear is really not to run all 1.4 million combinations in the first place. Work smarter, not harder.

Indeed. There are 2 decent ways of going at it:

1) Try to build a middle-ground set, and then run for stat weights. Optimize the gear set to match these stat weights, and then run again. Stat weights may shift around after optimization, as you pass a saturation point. Then simply backtrack a bit, or add weights, until you have a set where the weights between the top runners are tied. Alternatively, there could be a best stat that remains best even if you stack it to the max, in which case things are easier, but that's not as likely.

2) Alternatively, you could build different gear sets that prioritize different stats, and then compare them to each other, see which stats are best to prioritize that way. You might still want to find some middle ground.

You could also use the reforge plots to find your middle ground between two top-running stats.

As for trinkets, you probably want to start out with more neutral trinkets as you test the stat weights, then when you're done with the stat weights, test each trinket individually. When you've found the two best trinkets, test the stat weights again, to see if they still work well with those two trinkets.

Thels: regarding enemy_initial_health_percentage, it was changed to an actor-specific option. So something like

Code: Select all

enemy=Fluffy_Pillow
enemy_initial_health_percentage=20


would work. But using the option all by itself generally won't (I say generally because there are a few cases where it will 'accidentally' work, like i you put it at the very front of the simulate tab).

I'm going to add something like "override.initial_health_percentage=20" so that we have a real global option for it.

Thels wrote:Indeed. There are 2 decent ways of going at it....

Its weird, I'd rather spend days writing code and figuring out an automated process of doing it than doing it by hand and figuring it out logically....

1.4 million just wasn't going to get done fast enough. I tried with 2 threads and that's 30 mins per group of 20...ugh. 12 threads gets it to 4 mins, which is better but not enough for the full group. I reduced the sample set to 7k by getting rid of the neck, fingers, and trinkets.

I'll then write more code to analyze all the results and spit out the top 25 DPS, 25 HPS, 25 TMI, and 25 DPS/TMI ratio (unless you guys think there is a better ranking to use). Then use those gear sets to add a neck, then repeat and add 2 rings, repeat and add 2 trinkets and that should be close to the global maxima without getting too stuck in an inefficient local maxima.

My next run I'm going to try 18 threads and hope my PC still lets me do other work.

Or you could...

1) Put together 1 "guess" gear set based on assuming e.g. mastery>haste>crit
2) Generate stat weights with that gear set
3) Use wowhead link on report to go to wowhead and rank gear in each slot using those stat weights
4) Make new gear set out of the top choice in each category
5) Repeat steps 2-4 a few times until you have reached a stationary set, or you see oscillation between two gear sets.

While I enjoy a genetic algorithm on a cold winter night, I felt I could whip something out faster with brute force. Although something interesting showed up in the results that I was not expecting. Balance of stats seemed to be useful than stacking....here is examples:

If you take off the back, neck, fingers, and trinkets then I was able to compare the weights of the stats from the default profile vs weights from alternate gear profiles topping DPS, TMI, and HPS.

To come up with these percents I added all the stat values together and then took the stat and divided by the total.

Default:
Mastery = 30%
MS = 21%
Crit = 18%
Vers = 8%
Haste = 20%

The brute force versions:

Top DPS:
Mastery 12.5%
MS 25%
Crit 28%
Vers 4%
Haste 30%

Top HPS
Mastery 17.5%
MS 40%
Crit 19%
vers 8.5%
Haste 14.5%

Top TMI (Lowest TMI value)
Mastery 31%
MS 16%
Crit 21.5%
vers 23%
Haste 8%

These are percentages relative to each other, not percentages on the character sheet.

If anyone is interested I can upload the simc files but without back, neck, fingers, and trinkets its more of a fun exercise than really useful at the moment.

Next 2 days, running sims of the top 25 DPS, top 25 HPS, top TMI (lowest) and top DPS/TMI (highest) combined with neck and fingers. Then I can do the next top 25 groups + back and trinkets.

Balance instead of stacking seems to be the key here unless you are going after a specific meter to top.

Can we realistically get 23% of our bonus stats as versatility? I thought it was pretty limited in this tier.

viperjason wrote:Balance of stats seemed to be useful than stacking....here is examples:

I'm not sure those examples support your hypothesis.

When simming stat weights, we generally get Mastery>Versatility>Crit>else for TMI, and Crit/Haste/MS excelling for DPS. Thus, one would expect those stats to dominate in their respective columns. However, if you're using real gear, you don't have unlimited ability to min/max.

For example, if mastery is your strongest stat, at most you might expect to get ~50% mastery (i.e. all gear is mastery/x). In practice, it's probably less than that due to itemization - maybe you have to choose between a mastery/haste piece and a versatility/crit piece, and haste's poor showing causes the vers/crit piece to outperform, dropping your mastery representation further.

To really show that balance is advantageous, you'd need to show that a gear set that prioritized mastery even more highly performs worse in TMI. Another way to do this is to use reforge plots to show that there's a point at which you'd want to reforge mastery into something else.

Unless you meant that balance is useful because different stats are better for DPS and TMI (HPS is basically meaningless), in which case I think the argument is flawed on a more fundamental level.

You are right that I completely forgot that 50% is the max since most pieces of gear have the total stat pool split.

So far those stat priorities you provided theck seem to be holding for TMI and DPS. May turn out all my code was for naught, but it was a good exercise.

marsbubble wrote:Can we realistically get 23% of our bonus stats as versatility? I thought it was pretty limited in this tier.

23% doesnt mean 23% versatility. It just means that if you divide versatility stat by all the secondary stats together on all of your pieces, 23% of the stat values is versatility.

Adding necks and rings and you start to get a better picture:

Top DPS has haste slightly above crit/MS
Top TMI focuses Mastery and in fact hates haste. 0 haste on the gear used so far with the lowest TMI numbers....I wonder if this will continue on the last simulation set.
Top HPS still has multistrike focused slightly above Haste
I was also doing DPS/TMI and the highest ratio has a pretty even haste/mastery build.

So far nothing unexpected.

Last simulation set should be done in a day if I'm lucky. I'll post the full gear set and then people can see what they think.

What is the TMI diffenence in a full dps optimisation vs a full TMI set up? Is there a massive difference in it?

Reason I'm asking is my BrM co tank is fully after BA/Mastery/Vers build (same as our higest TMI) however as standard his best TMI is lower then ours so in theory if I were to go dps build I'm thinking our TMI would be on par (if that makes any sense lol)

Sorry if this question is in the wrong thread!

Are you finding that even with the added uptime of ShoTR that haste provides it still is a loss in DTPS/TMI compared to Mastery/Crit? This is what my Sims are saying as well I just want to make sure I am not alone. Also I am seeing Haste > Crit in my sims for dps but Crit may be that stat we take in order to increase dps AND survival.

Sexybeast wrote:What is the TMI diffenence in a full dps optimisation vs a full TMI set up? Is there a massive difference in it?

Reason I'm asking is my BrM co tank is fully after BA/Mastery/Vers build (same as our higest TMI) however as standard his best TMI is lower then ours so in theory if I were to go dps build I'm thinking our TMI would be on par (if that makes any sense lol)

Sorry if this question is in the wrong thread!

I understand what you are saying, I am trying to sim that out atm with different stat weights. Most of the heroic dungeon gear comes with every variety so as crit is very good for both DPS and DTPS a crit/x where x is a mix of multistrike/haste/mastery would probably be a good medium ground. Everything of course will be dependant on how hard things hit but atm from what i have tested on beta DTPS is really important especially in challenge modes where I was getting wrecked.

Donatist wrote:

Sexybeast wrote:What is the TMI diffenence in a full dps optimisation vs a full TMI set up? Is there a massive difference in it?

Reason I'm asking is my BrM co tank is fully after BA/Mastery/Vers build (same as our higest TMI) however as standard his best TMI is lower then ours so in theory if I were to go dps build I'm thinking our TMI would be on par (if that makes any sense lol)

Sorry if this question is in the wrong thread!

I understand what you are saying, I am trying to sim that out atm with different stat weights. Most of the heroic dungeon gear comes with every variety so as crit is very good for both DPS and DTPS a crit/x where x is a mix of multistrike/haste/mastery would probably be a good medium ground. Everything of course will be dependant on how hard things hit but atm from what i have tested on beta DTPS is really important especially in challenge modes where I was getting wrecked.

Awesome, thank you buddy :)

Sexybeast wrote:What is the TMI diffenence in a full dps optimisation vs a full TMI set up? Is there a massive difference in it?

Reason I'm asking is my BrM co tank is fully after BA/Mastery/Vers build (same as our higest TMI) however as standard his best TMI is lower then ours so in theory if I were to go dps build I'm thinking our TMI would be on par (if that makes any sense lol)

Sorry if this question is in the wrong thread!

Again, I dont have a full picture here because I am missing trinkets and back but here is what I have:

Highest DPS is 19620 with TMI of 116951.91
Lowest TMI is 114999.88 with a DPS of 19131

Remember this is missing trinkets and back.

I will explain methodology once the sims are done...hopefully by tomorrow.

I did not include enchants since that would just explode the number of combinations.

Final results are in and do improve upon the default config. Not significantly but a small improvement in all categories.

Focus on HPS, TMI and you significantly improve upon the default config.

This is my suggestion to the current items used for the simcraft prot pally T17M profile:

Code: Select all

head=lavadipped_greathelm,id=113894,bonus_id=567
neck=gluttons_kerchief,id=113882,bonus_id=567
shoulders=pauldrons_of_guiding_light,id=115565,bonus_id=567
chest=battleplate_of_guiding_light,id=115566,bonus_id=567
waist=ironcrushers_collar,id=113950,bonus_id=567
legs=legplates_of_guiding_light,id=115569,bonus_id=567
feet=iron_bellow_sabatons,id=113961,bonus_id=567
wrists=bracers_of_visceral_force,id=119331,bonus_id=567
hands=gauntlets_of_guiding_light,id=115567,bonus_id=567
finger1=seal_of_unquenchable_flame,id=113922,bonus_id=567
finger2=spellbound_runic_band_of_elemental_invincibility,id=118308,bonus_id=567
trinket1=horn_of_screaming_spirits,id=119193,bonus_id=567
trinket2=tablet_of_turnbuckle_teamwork,id=113905,bonus_id=567
back=gronnstitched_greatcloak,id=113873,bonus_id=567
main_hand=kromogs_brutal_fist,id=113927,bonus_id=567
off_hand=kromogs_protecting_palm,id=113926,bonus_id=567


This provides impred DPS by 4%, improved HPS by 1.5% and improved TMI by 4%. I was even suprised that the 4 set was maintained in the DPS selection. Was not expecting that.

If you sacrifice a bit TMI, you can get significantly better HPS slightly better DPS by going the top HPS route.

Code: Select all

head=lavadipped_greathelm,id=113894,bonus_id=567
neck=gluttons_kerchief,id=113882,bonus_id=567
shoulders=overdriven_spaulders,id=113990,bonus_id=567
chest=battleplate_of_guiding_light,id=115566,bonus_id=567
waist=ironcrushers_collar,id=113950,bonus_id=567
legs=legplates_of_guiding_light,id=115569,bonus_id=567
feet=sabatons_of_fractal_earth,id=113936,bonus_id=567
wrists=bracers_of_visceral_force,id=119331,bonus_id=567
hands=gauntlets_of_guiding_light,id=115567,bonus_id=567
finger1=seal_of_unquenchable_flame,id=113922,bonus_id=567
finger2=spellbound_runic_band_of_elemental_invincibility,id=118308,bonus_id=567
trinket1=horn_of_screaming_spirits,id=119193,bonus_id=567
trinket2=forgemasters_insignia,id=113983,bonus_id=567
back=gronnstitched_greatcloak,id=113873,bonus_id=567
main_hand=taners_terrible_spine,id=113862,bonus_id=567
off_hand=kromogs_protecting_palm,id=113926,bonus_id=567


The set that has the best DPS/TMI percentage is listed below:

Code: Select all

head=hardened_greathelm,id=113910,bonus_id=567
neck=gluttons_kerchief,id=113882,bonus_id=567
shoulders=pauldrons_of_guiding_light,id=115565,bonus_id=567
chest=battleplate_of_guiding_light,id=115566,bonus_id=567
waist=uktars_belt_of_chiming_rings,id=113976,bonus_id=567
legs=legplates_of_guiding_light,id=115569,bonus_id=567
feet=sabatons_of_fractal_earth,id=113936,bonus_id=567
wrists=bracers_of_visceral_force,id=119331,bonus_id=567
hands=gauntlets_of_guiding_light,id=115567,bonus_id=567
finger1=seal_of_unquenchable_flame,id=113922,bonus_id=567
finger2=spellbound_runic_band_of_elemental_invincibility,id=118308,bonus_id=567
trinket1=blast_furnace_door,id=113893,bonus_id=567
trinket2=tablet_of_turnbuckle_teamwork,id=113905,bonus_id=567
back=gronnstitched_greatcloak,id=113873,bonus_id=567
main_hand=kromogs_brutal_fist,id=113927,bonus_id=567
off_hand=kromogs_protecting_palm,id=113926,bonus_id=567


You want the best TMI then that is below but you risk a decent amount of DPS:

Code: Select all

head=helmet_of_guiding_light,id=115568,bonus_id=567
neck=choker_of_bestial_force,id=113952,bonus_id=567
shoulders=overdriven_spaulders,id=113990,bonus_id=567
chest=battleplate_of_guiding_light,id=115566,bonus_id=567
waist=ironcrushers_collar,id=113950,bonus_id=567
legs=legplates_of_guiding_light,id=115569,bonus_id=567
feet=sabatons_of_fractal_earth,id=113936,bonus_id=567
wrists=bracers_of_martial_perfection,id=113871,bonus_id=567
hands=gauntlets_of_guiding_light,id=115567,bonus_id=567
finger1=razoredge_blade_ring,id=113908,bonus_id=567
finger2=spellbound_runic_band_of_elemental_invincibility,id=118308,bonus_id=567
trinket1=battering_talisman,id=113987,bonus_id=567
trinket2=blast_furnace_door,id=113893,bonus_id=567
back=gronnstitched_greatcloak,id=113873,bonus_id=567
main_hand=kromogs_brutal_fist,id=113927,bonus_id=567
off_hand=kromogs_protecting_palm,id=113926,bonus_id=567


HTML results are here:
http://codelandia.com/Defaultvs.html

I can provide all of the sims, but that is over 2gigs of data :)

The report should be sufficient. To guide any further work, you should pretty much ignore HPS entirely. The only two things we're likely to care about are TMI and DPS, and the trade-offs (e.g. how much of one we gain or lose to improve the other).

Looking at those results, I think "TopDPS2TMI" is the clear winner. Very minimal TMI loss for a significant DPS gain. The curious (though perhaps predictable) part is that this set stacks both mastery and haste:

Top TMI: 1177 Mastery, 817 Bonus Armor, 785 MS, 637 Vers, 510 crit, 432 haste

Code: Select all

          Raid-Buffed Unbuffed  Gear Amount
Strength       4770      4543   4543
Stamina        8567      7199   6007
Health       514020    431940      0
Spell Power    7602      6362      0
Crit          15.64%    10.64%   510
Haste          9.54%     4.32%   432
Multistrike   16.89%    11.89%   785
OVersatility   7.90%     4.90%   637
DVersatility   3.95%     2.45%   637
Attack Power   7602      6362      0
Mastery       23.70%    18.70%  1177
Armor          3858      3858   3041
Bonus Armor     817       817    817
Tank-Miss      3.00%     3.00%     0
Tank-Dodge     5.00%     5.00%     0
Tank-Parry    25.42%    24.23%   510
Tank-Block    30.71%    26.51%     0


TopDPS2TMI: 1116 Mastery, 963 haste, 834 Bonus Armor, 702 MS, 592 Vers, 484 crit

Code: Select all

          Raid-Buffed Unbuffed  Gear Amount
Strength       4770      4543   4543
Stamina        7986      6671   5570
Health       479160    400260      0
Crit          15.40%    10.40%   484
Haste         15.11%     9.63%   963
Multistrike   15.64%    10.64%   702
OVersatility   7.55%     4.55%   592
DVersatility   3.78%     2.28%   592
Attack Power   7591      6353      0
Mastery       23.15%    18.15%  1116
Armor          3875      3875   3041
Bonus Armor     834       834    834
Tank-Miss      3.00%     3.00%     0
Tank-Dodge     5.00%     5.00%     0
Tank-Parry    25.28%    24.08%   484
Tank-Block    30.26%    26.03%     0


The top DPS set obviously stacks haste and crit, as you'd imagine.

Excellent sleuthing work Viper. I need to run a few sims of my own, but I suspect I'll switch the default profiles over to this mastery>haste>else setup.

Perhaps, but I wonder how much of the difference arises from the fact that the TMI gearset uses a Stam trinket while the DPS2TMI one does not. The haste factor could be incidental.

Yep, one of the things I want to check is whether normalizing the trinkets changes anything. Also want to see whether talents affect this in a material way (he used Empowered Seals - initial sims suggest that the results aren't any different for Seraphim or HS).

Yeah, looks like it's all about the trinkets. If you replace the trinkets in each setup to match the ones used by TopTMI, you get this (-NT means "normalized trinkets"):