Jump to content


Photo

Zerack's DK Gear Optimizer v2.6.9 (03-MAR-2010)


  • This topic is locked This topic is locked
213 replies to this topic

#21 Zerack

Zerack

    Piston Honda

  • Members
  • 219 posts

Posted 24 August 2009 - 11:55 PM

I've re-uploaded version 2.1.1 with a very small change fix (I didn't change the version number) that should fix the problem with trinkets. Basically, I had copy and pasted some blocks of code from my gear set evaluation w/ trinket procs into my final gemmed set evaluation and neglected to change a variable. Unfortunately that variable was still in scope in the gemming, so pulling from it didn't cause any errors.

You should see appropriate results for the trinkets now - at least I did in my limited tests. I didn't notice this when I was testing before since it doesn't show up if you don't run with more than 2 trinkets. Regardless it is now fixed and the OP and my previous post today both point to fixed versions.

Let me know if this fixes it for you.

#22 Canadianloaf

Canadianloaf

    Glass Joe

  • Members
  • 18 posts

Posted 25 August 2009 - 12:19 AM

I encountered a similar issue with ArP trinkets. I used Grim Toll and Pyrite Infuser for my test since they bring the same non-proc stat (+Hit). To simplify the test, I chose only a single option in each slot except the trinkets. With my gear list both trinkets push Hit to the Alliance hit cap so the difference in the non-proc value of the two is negligible. The gear is also well under the soft ArP cap. In this case, the proc value of Grim Toll (612*.2*2.67) should well exceed that of Pyrite Infuser (1234*.173). However, the optimizer selects Pyrite Infuser.

The text log is available at: http://files.me.com/...dianloaf/peuun6

Edit: Didn't refresh before posting.. will test whether the most recent patch fixes this.
Edit2: Awesome, Zerack. Confirmed that this test case is fixed.

#23 norg

norg

    King Hippo

  • Members
  • 589 posts

Posted 25 August 2009 - 10:20 AM

Asphyxia - I've both updated trinkets to be proc-based (with uptime as an entered value) and added consideration for caps being reached from the procs for partial value. This should help Blood optimizations immensely.


I'm in the process of double checking the Blood sets and I'm a little confused by the results I'm getting.

For the 'No restrictions' set, it is recommending a set with Mjolnir Runestone and 414 static ArPen, which obviously makes perfect sense when looking at the soft/hard caps.

However when I go to run the 'No heroic hard modes', (ie remove all items which are ilvl 258 or higher, and add in a couple more 245 ones to suit), it recommends a set with DC:G instead of Mjolnir and then starts adding in as much static ArPen as it can (526 with the items I made available, though presumably if I had made more available in the way of gems, for instance, it would have used more).

Can anyone think of a good reason why it would stay at softcap w/ Mjolnir but then ignore Mjolnir and shoot for the hard cap with slightly different items? It seems strange because I can't see any examples where I've removed an ilvl 258 item without having a direct ilvl 245 replacement available (Garrosh's Rage for Might of the Nerub, for example).

I just want to make sure it's working properly regarding the ArPen caps before I go and re-do all the BiS Blood sets for the twentieth time!

#24 Asphyxialol

Asphyxialol

    TEH DEEPZ!!!

  • Allied Members
  • 1,768 posts

Posted 25 August 2009 - 11:55 AM

Because the other stats on the items outweigh the 18.3% uptime on for an effective 122 Armor Pen, thus making Mjolnir excessive and rather than aiming for soft cap they push on for a passive armor pen of 526 (which is only ~100 more, and thus equivalent to Mjolnir, so when you think about it... it's pretty close, you just need enough itemization on the other items to push the equivalent of 102 crit and 20 armor pen, which realistically isn't hard).

#25 norg

norg

    King Hippo

  • Members
  • 589 posts

Posted 25 August 2009 - 01:09 PM

Because the other stats on the items outweigh the 18.3% uptime on for an effective 122 Armor Pen, thus making Mjolnir excessive and rather than aiming for soft cap they push on for a passive armor pen of 526 (which is only ~100 more, and thus equivalent to Mjolnir, so when you think about it... it's pretty close, you just need enough itemization on the other items to push the equivalent of 102 crit and 20 armor pen, which realistically isn't hard).

Actually, I can only assume I didn't add DC:G to the equation when I initially did that 'No restrictions' set, as when I double checked it just now, the optimizer eschewed Mjolnir in favour of DC:G, even when static ArPen was under the softcap.

The overall scores were 14688 when using Mjolnir and 14713 when using DC:G. All other items were the same.

From that I can only assume that the final verdict is that DC:G > Mjolnir for Blood, even if you are able to utilise the entirety of Mjolnir's proc. Does that sound right?

#26 Zerack

Zerack

    Piston Honda

  • Members
  • 219 posts

Posted 25 August 2009 - 01:24 PM

From that I can only assume that the final verdict is that DC:G > Mjolnir for Blood, even if you are able to utilise the entirety of Mjolnir's proc. Does that sound right?


Yes, it does. To do some quick math:

I'm assuming you're using the Blood weights currently in the simulator, in which case Str = 2.82, Crit = 1.82, and ArP = 2.67.

For , you have a static 90 Strength. The proc is then worth 300 * 0.333, or 100 Strength. Total value of the DC:G is then 190 * 2.82, or 535.8.

For , you have 102 static crit. The proc is then worth 665 * 0.183 ArP, which is 121.695 ArP. Total value of the trinket is 102 * 1.82 + 121.695 * 2.67, or 510.56565.

The conclusion is that DC:G is better than Mjolnir Runestone even for Blood with our current weights. If we weight ArP more highly then Mjolnir will pull ahead in fairly short order, since they are already close.

#27 norg

norg

    King Hippo

  • Members
  • 589 posts

Posted 25 August 2009 - 01:53 PM

Thanks for clearing that up, Zerack.

There used to be a consensus that ArPen increased in value the more it was stacked. Is this still the case, and if so does the optimizer accurately reflect this?

Doc touches on the issue in this post and seems to imply that ArPen is still getting better the more it is stacked.

Thanks again for your work on the optimizer, btw. It's an absolute godsend for that BiS thread!

#28 Zerack

Zerack

    Piston Honda

  • Members
  • 219 posts

Posted 25 August 2009 - 02:48 PM

The optimizer does not increase or decrease the value of any stat as the stat is stacked more or less (caps for stats excluded, of course). The nature of ArP means that it works on something of a sliding scale - the best solution is most likely to generate a set, run that set through Kahorie's to get weights at that point, and then re-generate a best set using the new weights.

If you iterate like this you should be able to cope with the varying ArP weights. That said though, I did see the post with the ArP values and my gut tells me that the weight in our Blood set is a bit low for the amount of ArP that the sets have. That's outside my area of expertise, though - someone else will have to clear that up.

#29 dr_AllCOM3

dr_AllCOM3

    Great Tiger

  • Members
  • 985 posts

Posted 25 August 2009 - 03:28 PM

Arp is tricky. It can't really be included in stat weights, since it's never the same. Unfortunately Arp hasn't been discussed any further, even though it's the most interesting topic for Blood (and maybe Frost) right now.
Maybe I'll run a test series, but now right now.

#30 Zerack

Zerack

    Piston Honda

  • Members
  • 219 posts

Posted 25 August 2009 - 04:43 PM

Alright - given that ArP is a difficult problem to solve, what would be a good (accurate) way to calculate its value in an arbitrary set short of running a huge simulation in Kahorie's simulator? Could the people who develop weights run tests at different ArP rating levels (0, 100, 200, 300, etc.) and develop EP at each of those levels that could be used to variably weight ArP in sets?

Could we develop an equation that was a gross estimation for the value of ArP based on factors like Strength, ArP, and other stats (per spec)? I realize that these seem like hefty tasks, but as has been pointed out, the single point weight for ArP really isn't a good way to estimate value for these sets. I'm more than willing to dig into this problem a bit more (it wouldn't even be that complicated to implement something like this, really) since it will provide better results for everyone using the sim.

To be clear, what I'm suggesting is perhaps something like this. Instead of a single ArP weight, you would have something like this:

Total ArP %|Weight
0|1.50
10|1.90
20|2.30
30|2.50
40|2.75
50|2.90
60|3.10
70|3.23
80|3.38
90|3.50


You could use a table like that (with more precision and real numbers) to evaluate ArP at different levels during the same run. The equation alternative could end up looking something like (obviously again made up completely):

ArP_weight = 1.5 + ArP_Percent / 50

This would scale the weight from 1.50 to 3.50 the more ArP you had. Does either of these solutions seem like something we could come up with? My thought is that the tabular method is just a matter of enough work with Kahorie's simulator and some static gearsets with ArP changing, while the equation might be a bit more complicated to deal with.

Let me know what you guys think.

#31 Asphyxialol

Asphyxialol

    TEH DEEPZ!!!

  • Allied Members
  • 1,768 posts

Posted 25 August 2009 - 06:01 PM

The issue with only changing Armor Pen is that it is impossible in-game to increase armor pen to those levels without gaining and losing other stats. I am not sure simply editing the single stat would result in proper test results. Is there a way we can come upon an agreed upon set of generalized stats for the different brackets or something?

I can say that through my personal gear upgrades the more armor penetration I had the higher its weight became, but I have to keep in mind that with the upgrades I received with armor penetration on them also generally came with upgrades to other areas of my character as well...

Obviously 100% armor penetration would make a larger difference on someone in 2000 str, 750 crit rating, hit/exp capped gear than it would on someone with 850 str, 500 crit, hit / exp capped, or even my current stats, which are only about 1790 str, 620 crit, 148 haste, spell hit / exp cap, and about 290 arm pen without Runestone factored in.

#32 Zerack

Zerack

    Piston Honda

  • Members
  • 219 posts

Posted 25 August 2009 - 07:00 PM

Obviously 100% armor penetration would make a larger difference on someone in 2000 str, 750 crit rating, hit/exp capped gear than it would on someone with 850 str, 500 crit, hit / exp capped, or even my current stats, which are only about 1790 str, 620 crit, 148 haste, spell hit / exp cap, and about 290 arm pen without Runestone factored in.


Obviously the person with higher stats will see a higher absolute DPS increase from 100% ArP than someone with worse stats, but on a point for point EAP scale, how much difference will there be in the 1108th point of Armor Penetration for the two players? I'm really asking here, since I'm not that familiar with the math. DPS gain being not equal to EAP change in this case could mean that (with some approximation) we could arrive at something not perfect but more accurate than the current system.

On the other hand, I could be completely wrong in the above paragraph. You never know.

#33 norg

norg

    King Hippo

  • Members
  • 589 posts

Posted 26 August 2009 - 10:29 AM

Diverting the topic away briefly: the rings section contains entries for Carnivorous Band and Gormok's Band. These are actually just the Alliance/Horde versions of the same item, so to avoid confusion it would probably be best to change them to the same name and give one a Heroic tag.

Ultimately I suppose it'd be best to give each item both its Horde and Alliance name, but that depends on whether you can be bothered there! It's not so bad for items where both versions have virtually the same name (ie. Planestalker Band vs Planestalker Signet), but as you have proved, it can get confusing when the equivalents have completely different names.

e: incidentally does anyone know if the separate item names thing is a trend they are planning to carry forward beyond the Coliseum loot? Because I must say it's made compiling these tables much more of a pain in the ass. :mad:

e2: also the Armbands of Dark Determination have the heroic and non-heroic stats the wrong way round.

#34 Zerack

Zerack

    Piston Honda

  • Members
  • 219 posts

Posted 26 August 2009 - 02:43 PM

also the Armbands of Dark Determination have the heroic and non-heroic stats the wrong way round.


Are you sure? It matches mmo-champion's loot list on my end. The Heroic version has two sockets which drive down the stats in other places.

I'll fix the ring, though - good catch.

#35 norg

norg

    King Hippo

  • Members
  • 589 posts

Posted 26 August 2009 - 03:11 PM

Are you sure? It matches mmo-champion's loot list on my end. The Heroic version has two sockets which drive down the stats in other places.

I'll fix the ring, though - good catch.


Ah yeah, my bad.

#36 Calgar

Calgar

    Piston Honda

  • Members
  • 162 posts

Posted 27 August 2009 - 02:21 PM

Yes, it does. To do some quick math:

I'm assuming you're using the Blood weights currently in the simulator, in which case Str = 2.82, Crit = 1.82, and ArP = 2.67.

For , you have a static 90 Strength. The proc is then worth 300 * 0.333, or 100 Strength. Total value of the DC:G is then 190 * 2.82, or 535.8.

For , you have 102 static crit. The proc is then worth 665 * 0.183 ArP, which is 121.695 ArP. Total value of the trinket is 102 * 1.82 + 121.695 * 2.67, or 510.56565.

The conclusion is that DC:G is better than Mjolnir Runestone even for Blood with our current weights. If we weight ArP more highly then Mjolnir will pull ahead in fairly short order, since they are already close.


Alright. I wanna just double check something here, the bolded part in particular.

Stat x stat weight = value, adjusting for uptime obviously.
Base stats: 102 crit x 1.82 = 185.64 - Easy enough
Proc: 665 Arp x 2.67 = 1775.55, which now has to be adjusted by the uptime of 10 seconds every 45 seconds. So 10/45, or .2222 = 394.56

394.56 + 185.64 = 580.2. Which is clearly better then DC:G's value of 535.8.

I know the difference in our numbers is the uptime, you are using 10 seconds every 60, but it's 10 seconds with a 45 second ICD AFAIK. Am I not calculating this right? Does the ICD start from the end of the proc and you factored in another 5 seconds of padding for actually getting it to proc?

Especially considering how little ArP the BiS blood suit has, I'd imagine the runestone would be great.

#37 Zerack

Zerack

    Piston Honda

  • Members
  • 219 posts

Posted 27 August 2009 - 03:47 PM

The detail of this is in the proc chance of . It's listed at only 15% after its 45 second cooldown on Wowhead, so it's not actually up for 10 seconds every 45. As described in this post (link), the uptime is only going to be around 18.3%, which is what is used in the simulator. At that proc rate and the weights currently in the Blood set, DC:G still turns out to be better.

Now I'm not discounting that the higher value of ArP as more is accumulated doesn't cause this to be a little bit skewed. However, if you both assume that the weights are correct and trust the math on the proc uptime there (which I've seen supported by anecdotal reports within 1-2% of that number), then the comparison of the two trinkets is valid.

#38 frmorrison

frmorrison

    Protector

  • Allied Members
  • 11,427 posts

Posted 27 August 2009 - 09:25 PM

incidentally does anyone know if the separate item names thing is a trend they are planning to carry forward beyond the Coliseum loot? Because I must say it's made compiling these tables much more of a pain in the ass. :mad:


It is annoying, but I think they did it for flavor and to see how people reacted to it. I agree that makes things more complicated and hope it isn't used again.


Regarding the stat weight of ArP (that does not increase linearly and has a cap), a stepwise scale is better than a single value at all points.

#39 norg

norg

    King Hippo

  • Members
  • 589 posts

Posted 28 August 2009 - 08:18 AM

It is annoying, but I think they did it for flavor and to see how people reacted to it. I agree that makes things more complicated and hope it isn't used again.


Regarding the stat weight of ArP (that does not increase linearly and has a cap), a stepwise scale is better than a single value at all points.


It's fine for the set items or items that have a visual difference to differentiate between Horde and Alliance, but it seems fairly pointless applying it to rings and trinkets.

Incidentally I am still having issues with extremely slow calculation times. I always select a red gem, orange hit/str gem, and a Nightmare Tear. If I try and add a yellow hit gem to the mix the calc time increases from a few minutes to a few hours. This is having already changed the 'Hit (to melee cap)' weight to be below the strength weight, and with between one and three options per item slot.

#40 Zerack

Zerack

    Piston Honda

  • Members
  • 219 posts

Posted 28 August 2009 - 12:42 PM

Incidentally I am still having issues with extremely slow calculation times. I always select a red gem, orange hit/str gem, and a Nightmare Tear. If I try and add a yellow hit gem to the mix the calc time increases from a few minutes to a few hours. This is having already changed the 'Hit (to melee cap)' weight to be below the strength weight, and with between one and three options per item slot.


I'm in the process of re-writing the gemming to be hopefully more efficient and to deal with higher hit values gracefully, but in the meantime while we're still putting Hit To Melee Cap < Str, it's worth noting that there is no situation in which a Rigid cut gem (+hit) will be used over an Etched cut gem (+hit/+str) by the optimizer. When I run optimizations for myself with hit < str, I use Bold, Sovereign, Nightmare Tear, Etched, and Inscribed / Fierce (depending on crit > haste or haste > crit). I'm fairly certain that these five gems cover every gem that would ever be used in that situation.

Those calculations finish in around 2 hours on my resurrected four year old laptop with 5-6 items in most of the slots, so I would call it fairly reasonable. Let me know if you continue to have problems, though, and I'll keep you posted on my progress.




0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users