Simple Item Ratings Model

Topics: Retired
May 16, 2008 at 3:59 PM
Sometimes building a realistic model would be a ton of work (consider PvP issues), but we are still willing to come up with item values.  As an example, these are some of the values I use for choosing Moonkin Battleground gear:

Stat Value
SpDam 1.00
Stam 1.00
Resil 0.48
SpellCR 0.41
+Heal 0.23
Armor 0.11

This is along the lines of the famous AEP models for rogues.  It seems like it would be straightforward to build a "model" where the options tab would just be a two-column table, corresponding to this spreadsheet.

This would also provide a tool for people leveling.  A level 40 feral leveling druid might decide (rightly or wrongly) to use weights of

Str = 1
Agi = .85
Stam = .75
Int = Spi = .4

and have a useful tool for finding upgrades (although this really needs better support for importing "of the Eagle" type items).

In my spreadsheet, I actually don't enter these raw values.  Instead I have fairly simple models that produced

1) Burst DPS (1 +spell = .0532% improvement, spell crit = .0201%)
2) Sustained DPS (+spell = .0648%, crit = .0289%)
3) Moonkin Survivability (stam = .0706%, resil = .0304%, Armor (gear) = .0079%)
4) Burst healing (heal = .0259%, crit =.0205%)
5) Sustained healing (heal = .0518%, crit = .0029%)

Where "Sustained" means that mana conservation/regeneration is a significant concern.  I then decided that for battlegrounds I'd use a weighting of

35% Burst DPS
5% Sustained DPS
40% Survival
15% Burst healing
5% Sustained healing

and then normalized the results so that +spell/heal had a value of 1.00.  For other activities (say healing a 5-man PuG instance,), I might use the same base tables, but with substantially different weights.

So, in addition to the ability to create AEP-style tables, it would be nice to be able use weighted-combinations of such tables.

May 16, 2008 at 4:22 PM
Weights are inherently wrong, by design, and will not be used in Rawr.
May 16, 2008 at 7:01 PM

Astrylian wrote:
Weights are inherently wrong, by design, and will not be used in Rawr.

Actually weights are very prominent in Rawr.

calculatedStats.MitigationPoints = (7000f * (1f / (calculatedStats.DamageTaken / 100f)));


That 7000 is a weight on the relative value of Survival and Mitigation.  In many (most) cases Rawr doesn't let the end-user select the weights, but they are there.

Of course this example proves the first part of your statement.  It is very unlikely that 7000 is correct for all gear levels and encounters, and not_correct == wrong.  Even if it is a pretty good value for both Kara and BT, it would almost certainly need to be changed for level 80 raids, or for a tank in ZF.

Likewise, the Moonkin DPS model uses (I believe) an implicit weight of zero for Stamina.  It will tell me to trade 4000 health for 25 +spell, which is correct under certain assumptions (ie I won't take any incoming damage), but those assumptions won't always apply.

In fact if I enter my Bear stats in Rawr, I can pull up "Chart: Custom > Relative Stat Values".  That looks pretty much like a list of weights to me.  I don't see how they are any more "wrong" than the gear comparisons.

Weights are "wrong" when they aren't adjusted based on your current gear and encounter.  It looks like that 7000 is wrong.  My inclination would be to let the user change it, but in any case it seems to serve a useful purpose in Rawr.

Rawr is a great tool.  I understand a major motivation for producing Rawr was disappointment with tools that used fixed weights.

Three major components of Rawr are:

- User Interface
- Equipment Database
- Evaluation Models

I have a non-Rawr evaluation model that tells me that with my current gear, and an assumed incoming stream of PvP attacks (magical/physical/direct/dot/crit-rate) and heals, the relative value (time-to-live) to me of Stamina/Resil/Armor/Defense/Agility ... .  I can tie that to a DPS model, perhaps by making the assumption that I can do 1% more damage by either
  a) 1% higher DPS
  b) 1% longer DPS 
It would be nice if I could use the output from that Model, with Rawr's User Interface and Equipment Database.

May 16, 2008 at 7:03 PM
Actually, weights are perfectly valid, but change frequently, and they are used in Rawr. For example, if you look at the Prot Warrior, you have weighting between Threat, Mitigation, and Survival. In the Mage model, you can lookup the DPS value of 10 more points in damage, hit rating, etc (near the bottom of the list of charts).

What is wrong is thinking that weights don't change. Weights are different from fight to fight, and the value of stats also varies with every piece of gear you change.

But it can be valuable to get the weignting of stats with your current gear, for use in roughly estimating the value of a new piece of gear in game. There are many mods that will 'rate' gear using a set of values you give it. When having to decide quickly in a raid whether to spend points on an item or not, having a single value assigned to gear, even if it's just a rough estimate, is very convenient. After I replace a couple pieces of gear I go back to Rawr.Mage and figure out the new stat values based on my current gear... typically they change only a little bit. Spell crit is slowly climbing up to be just as good as +spell damage and haste.

Point is that some models already provide weightings between different factors, and other models already provide ratings for stats.

However, the request above is too ambitious. You can't model healing and DPS in the same model with any kind of sanity, nor is Rawr designed to model lowbies. It's a raiding tool (or at least instancing tool) for level 70 characters filling a specific role in PvE. It's not designed to model PvP, lowbies, or multi-role armor.     
May 16, 2008 at 7:10 PM
"Valid" or not, weights *are*, by their very definition, inaccurate. You said it yourself: "...for use in roughly estimating the value of..."

Rawr will not, and does not, use stat weights for determining any of its ratings. Why would you use an inaccurate calculation, when you could have an accurate one, in the same time (or a few milliseconds longer)?
May 17, 2008 at 12:11 AM
I think what he's suggesting is that in absence of accurate models writing a weight-based evaluation model is several orders faster to implement and can serve to fill the gap of non-existing models.
May 19, 2008 at 4:22 PM
Oh I see... he wants a model that just lets you plug in AEP values (or equivalent for other classes). I guess I could see that as a mildly useful stopgap... but eh. There are websites that do that, Rawr doesn't have to.