5.2 Protection Paladin stat weights

[Theck]

I played around with AMR this morning to determine what, if any, modifications need to be made to the stat weights for 5.2.  While the tweaked versions seem to work reasonably well, I think they could be made more robust.

The current (tweaked) stat weights were developed before the "Force Hit/Exp cap" option was available. As such, I had to exaggerate the valuation of hit/expertise to encourage the algorithm to cap or near-cap those stats.  This was further exacerbated by the relative rarity of rating on gear.  It also meant I had to crank up the value for stamina for gemming, e.g. to encourage the algorithm to use hybrid haste/stam gems instead of pure haste.  However, those issues are behind us now, so I think it makes sense to adjust the stat weights to a flatter scale.  This should help give good results with and without the "Force Hit/Exp cap" toggle checked, and eliminate some of the odd stam-stacking behavior that users have been complaining about.

In many cases it will be worth picking up non-stamina (hit, haste, even parry/dodge) socket bonuses.  The low valuation on avoidance coupled with the high stamina weight tended to ignore these, and often mastery/haste socket bonuses.  Again, this was primarily because of the exaggerated gap between the "top-tier" stats (stamina, hit, expertise) and "second-tier" stats (haste, mastery).  A flatter scheme should correct this, a well as make it easier to include dodge/parry socket bonuses, which would be nearly impossible in the old scheme. 

This morning, I worked up stat weights that approximate the different gearing strategies outlined in my recent blog post.  I've tested the Control strategies pretty extensively, less so for the avoidance strategies since those are pretty robust to begin with (no caps to worry about).  The goal was to come up with a set of weights that:
1) followed the appropriate gearing scheme
2) generally use hybrid gems to efficiently pick up useful socket bonuses (a boon to 10-man raiders, who have long lamented my stam-heavy bias)
3) tend to favor higher ilvl pieces a little more than before (armor is an oft-overlooked smoothing stat)

Here's what I came up with.

Code: [Select]

                     Gear Strategy
Stat Weight   C/Ha    C/Ma    C/Ba    C/Av     Av    M/Av
Stamina        1.5     1.5     1.5     1.5     1.5     1.5   
Hit            1.1     1.1     1.1     1.1     0.5*    0.5*
Expertise      1.09    1.09    1.09    1.09    1.09*   1.09*
Armor          1       1       1       1       1       1
Mastery        0.8     1       1       0.8     0.5*    1
Haste          1       0.8     1       0.5     0.5*    0.5*
Strength       0.6     0.6     0.6     1       1       1
Dodge          0.5     0.5     0.5     1       1       1
Parry          0.5     0.5     0.5     1       1       1

*fungible weight - chosen to be low enough to deprioritize, anything lower is likely fine as well.

Dropping the stamina weight to 1.3 will convert from a balanced, hybrid gem strategy to a very aggressive secondary-stat gem strategy.  Bumping the stam weight up to 1.8+ will shift back towards a very stam-heavy gem strategy, with the level of single-mindedness increasing as you go higher.  2+ will start using fewer and fewer hybrids, 3+ will almost exclusively use stamina gems unless you get a stamina socket bonus or need the hybrid to achieve hit/exp caps (if that option is enabled).

[yellowfive]

Thanks for the info -- we'll take a look and throw it into our next round of stat weight updates on the live site.

This whole week is going to be "stat weight week", so expect frequent updates until next tuesday or so when we hope to have everything pretty stable.

[Theck]

Yep, understood.  My suggestion would be to just stick with the Control/Mastery (default), Control/Haste, and Avoidance (M/Av on the table, really a Mastery-focused avoidance build) gear sets.  The rest aren't really interesting enough to warrant a preset, and people who want to fool around with it can always just plug in the custom weights based on the table.

[yellowfive]

Yeah that sounds good -- I was playing with the weights you posted a bit this afternoon, look pretty solid in my tests (I usually do a test at pre-raid (~470 ilvl), raid (~500) and max (~535) to see if anything goofy starts happening).  I hope to work them into tonight's update, along with a few tweaks for other classes.

There are some tweaks to trinkets and set bonuses coming soon as well -- these should impact the best-in-slot optimizer more than the mod optimizer.

[Theck]

For 5.3, I'm going to suggest a few adjustments.  First, here are slightly updated stat weights for the three gearing strategies:

Code: [Select]

                     Gear Strategy
Stat Weight   C/Ha    C/Ma    M/Av
Hit            1.05    1.05    0.5*
Expertise      1.02    1.02    0.5*
Armor          1       1       1
Mastery        0.7     1       1
Haste          1       0.8     0.5*
Strength       0.5     0.5     1
Dodge          0.35    0.35    1
Parry          0.35    0.35    1

Setting:  Max    High       Med     Low
Stam      1.5+   1.34-1.4   1.2     < 1

*fungible weight - chosen to be low enough to deprioritize, anything lower is likely fine as well.
The Control/Haste weights are the ones I'm using at the moment after playing around with them.  I've dropped mastery and dodge/parry/strength with respect to haste to reflect the stronger-than-expected effect of Seal of Insight.  I've also dropped hit and expertise a little to bring them closer to haste.  Partly, this is because many of us are actually having trouble staying under hit cap.  With the "force caps" option, hit and expertise don't need to have significantly higher stat values than haste anyway.

As far as the stamina settings, they're probably OK as is.  I think that 2+ is probably too high for the "Max" setting, as 1.5 or more suitably forces full stam gems.  1.34 is the minimum for the "High" setting, as that's the cutoff where it will generally stop using hybrid gems in yellow/red sockets.  Anything between 1.34 and 1.4 should work here, I tend towards around 1.35ish.  Medium and Low are probably fine where they are now (1.16 and 0.3ish if I recall correctly).

I would also recommend setting Control/Haste as the default.  I no longer believe there's any advantage to be had from the Control/Mastery build (more in this blog post).  It could probably be removed entirely, but since it's already there it won't hurt anything to leave it.  It may have some niche usefulness in heavy AoE settings (though the pure avoidance build is probably even stronger in those cases).

[yellowfive]

Thanks for the feedback -- we'll take a look and work those into one of our upcoming updates.

Yeah 2x your next-highest weight is probably a bit overkill for the "max" stamina setting (which in itself is probably overkill most of the time).  We're still working on a few ideas/tweaks to bring back some more flexibility around targeting specific stamina levels, but still wading through the expected barrage of minor 5.3 item data issues.

[Theck]

Just an update: after playing with it some more, I've had better luck using hit/exp values that are even closer to haste, like hit at 1.02 and exp at 1.01 (for haste @ 1.00).  It cleans up some of the gemming issues and tends to get the optimizer to pick better solutions (for example, doing this finds a solution that wastes less hit/exp over cap in my current gear).

I'd also suggest that the default enchant be changed.  Colossus hasn't scaled well (i.e. at all) with gear, and while I still think it's a valid choice, Dancing Steel and Windsong are competitive at this point.  In fact, I think Windsong is probably the strongest all-around choice for heroic modes just by virtue of pure DPS.  In normal modes, any of the three are legitimate choices, though Dancing Steel will probably lead in DPS. 

Windsong is probably the best default option for everyone now, simply because it serves multiple functions: high DPS compared to Colossus, similar survivability gain, much cheaper than Dancing Steel.

[Theck]

Uh.. well this is odd...


Custom weights are:
Stam: 1.35
Hit: 1.02
Exp: 1.01
Haste: 1.0
Armor: 0.99
Mastery: 0.7
STR: 0.5
dodge/parry: 0.35

I also have min ilvl set to 522, so I'm not sure why it's even considering Kil'rak, but it seems clear something is wrong here.  Also note that it's using the gem sockets of the T15 ret chest instead of the gem sockets for Chestplate of Violent Detonation, thus giving it an extra socket.

Then I hit "save as" and saved as "theck-bugged" again for your convenience, and this happened (same settings):



I think the huge increase here is because it's switching me to the tank meta again, but what struck me as odd is that the BiB score is exactly the same as the previous screenshot.  Given that the second one looks like the same bug we encountered before, I'm guessing this is what it was optimizing to (but not displaying) in screenshot #1?

Color me puzzled.  I have a strange knack for breaking AMR, I guess?

[yellowfive]

We can make the default enchant change -- we would basically have to hard-code it to ignore colossus (which is an estimate converted to armor, so the relatively high weight on armor tends to favor damage absorption effects).  Either that or apply a custom reduction to colossus (so anyone who has it equipped doesn't get an artificially low score), or reduce the weight on armor... not sure how that would impact gear suggestions though, would have to try it out.

There is a display bug in that first screenshot... it's showing all of the modifications from your "my gear" tab, but the items are from the "best in bag" tab.  Not sure how that happened... I tried to reproduce a similar display issue using your profile and a few others, but couldn't.  I would try clearing your browser cache first... might make the issue in the first screenshot go away.

For best-in-bag, we basically ignore all filters, and inspect all available items in your bag.  The idea is that it shouldn't pick bad items in the first place... not really sure why it would choose that weapon.  I downloaded your latest inventory data and was able to reproduce the problem, so I'll take a look and see what's up.

[yellowfive]

Actually... I think that I see the problem.  In your bag data, you have do'tharak locked with dancing steel.  Thus it thinks that your next-best weapon with colossus will score better.  If we adjust the score on colossus, that problem should go away.

We probably need to add an option for tanks to use the DPS meta gem... we could probably try to come up with a way to score it by putting a weight on attack power for tanks... but you would have to adjust AP/armor weights to seemingly arbitrary values to make it choose one meta or the other (even though we do try to calculate pretty good equivalences for the absorption effect and damage proc... so maybe it wouldn't be that arbitrary).  What would be kind of arbitrary is estimating the amount of AP a tank would have for the damage that the proc does.

[yellowfive]

I did find one other bug with the best-in-bag optimizer where it was holding onto a sub-optimal solution even though it scores worse than your equipped gear, that has been fixed.  Really obscure bug with a bad object reference, was hard to track down.


So in this update, I made some tweaks to a few of the estimates for enchants and meta gems for tanks.  Firstly, I reduced the value of colossus and the legendary tank meta gem proc quite a bit -- the idea is that these rather random damage reduction effects are, by nature, of less value than static armor because you can't rely on having them when you need them.  This reduction allows more "reasonable" weights to control which enchant/gems are chosen.

I did not change the default weights for protection paladins yet, but my proposal would be something like this:

stam: 1.35
hit: 1.05
exp: 1.02
haste: 1
mastery: 0.7
armor: 0.6
strength: 0.55
dodge/parry: 0.35


With these weights, the chosen weapon enchant would be dancing steel.  If you turn on "exclude expensive enchants", it would choose windsong.  The score for colossus is very close to windsong, so bumping up armor slightly would choose colossus for people who prefer that.  If someone prefers windsong in all cases, a slight reduction in strength would start choosing it.

As for the legendary meta, if you add some reasonable weight for attack power, it will choose the DPS meta instead of the tank meta.  e.g. setting it to strength weight / 1.05 (a very logical value), would pick the DPS meta.

How does that feel?  The adjustments are live so that you can try weights like these -- if no one sees any problems, I'll go ahead and update the defaults weights.

Now the last question... which legendary meta gem to suggest by default... they are both very good for what they do, but the damage on that legendary meta is very high... pretty tempting for any tank.

[Theck]

Dancing steel is probably the weakest enchant of the three for survivability, but it's a very efficient way to trade stamina/survivability (from Colossus) for DPS.  That's why people would take it. Similarly for Windsong - above 170k Vengeance or so, Windsong out-DPS-es Dancing Steel.

Colossus may still hold the survivability title, but I'm not sure.  It was ahead in T14, but not by a whole lot, and Windsong's proc rate was buffed since then.  Combine that with heavier-hitting bosses and it should probably fall behind Windsong, I think. I worked out its time-averaged stamina equivalence a while back and it wasn't very much (under 50, if I recall correctly). 

Those weights look reasonable.  I tend to drop hit and exp even further ( like hit=1.02, exp=1.01).  When you're not able to cap, they take higher priority than haste, but when you have an abundance of rating they really should only be valued as high as the next best stat you could reforge them into.  I've noticed some oddities with the slot-specific ordering of gear when I have hit/exp much higher.  In particular, the order of the ret T15 chest (522) and Chestplate of Violent Detonation (535) jump around a lot depending on the hit valuation, which suggests to me that even when "relative" ranking is chosen it's doing something funny and overvaluing the hit on the latter.

However, I just tested your weights out and they give me identical results to my hit-depressed version thanks to the lower armor weight, so they're probably fine.  The armor weight is a bit arbitrary anyway, and only serves to change how much you value raw ilvl upgrades. 

I'm mixed on the meta.  I think AMR should probably suggest the tank one by default, because in general survivability should be the first goal.  That said, it is currently a huge (~10%) DPS upgrade to take the DPS meta, which most tanks find very attractive.  My best advice is to give them both exactly the same value, but suggest the tank one by default if the DPS one isn't already equipped. 

[Theck]

By the way, it's doing that thing where it returns an optimization that has a lower score than the current setup again.  I'll leave myself logged out in the gear set that's causing it.  I'm using the stat weights from your post, and the only modification to the gear is to lock in the helm gems (meta).  It's trying to do full stam gemming despite the stam weight being 1.35, and returning a net loss of ~190 score.

[yellowfive]

I just loaded up your latest, and I locked only your current helmet.  The BiB tab gives a lower score -- the reason though is that it wants to put the tank meta gem into your other helmet.  If I lock both of your helmets (which gives BiB no way to use the tank meta instead of DPS meta), it says that your current setup is already optimal.  If I add a weight to Attack Power, I don't need to lock anything -- it says that your current setup is both optimal and BiB.

It seems that with your particular gear, there is a solution that uses mostly stamina gems and expertise reforges vs. one that uses mostly exp/stam gems and non-expertise reforges that scores similarly.  When a couple items swap around, it switches between the two sometimes.  I would have to run a test to see if the mostly-stamina gem solution is really better in the case where it wants to pick it, or if it's just really close.  Regardless, the new optimizer that will go into beta soon (now that 5.3 stuff is behind us) should make that a non-issue and always return the single best choice.

[Crozekiel]

Dancing steel is probably the weakest enchant of the three for survivability, but it's a very efficient way to trade stamina/survivability (from Colossus) for DPS.  That's why people would take it. Similarly for Windsong - above 170k Vengeance or so, Windsong out-DPS-es Dancing Steel.

Colossus may still hold the survivability title, but I'm not sure.  It was ahead in T14, but not by a whole lot, and Windsong's proc rate was buffed since then.  Combine that with heavier-hitting bosses and it should probably fall behind Windsong, I think. I worked out its time-averaged stamina equivalence a while back and it wasn't very much (under 50, if I recall correctly). 

I was actually just coming over to the forums to see if 5.3 had changed something about weapon enchants that I had missed.  I have noticed in the last few days that all the tank specs (except dks) are coming back with dancing steel as the best enchant on AMR (not just on my tank, but other tanks in my guild as well as the generic BIS characters).  I thought that seemed quite odd considering how low strength (and for that matter, parry) are on the default (control/haste) weights.  I have been using Colossus up until now, and would really rather not waste the sha crystals swapping to Dancing Steel to lose survivability.