I’m experimenting with a gem report for each class. It is easy enough to generate a list of the most popular gem choices. Like a lot of the other player-configurable options in WoW, you get the infamous power law distribution: a few choices that many players make and a long tail of left-of-centre choices each of which is made by only a few players.

For priests (level 70, pre-patch-3.0.2) we get something like this:

Gem Colour Popularity
Discipline Builds
Steady Talasite green 75%
Teardrop Living Ruby red 64%
Royal Nightseye purple 54%
Luminous Noble Topaz orange 45%
Mystic Dawnstone yellow 39%
Teardrop Blood Garnet red 32%
Powerful Earthstorm Diamond meta 28%
Solid Star of Elune blue 24%
Teardrop Crimson Spinel red 22%
Insightful Earthstorm Diamond meta 21%
Holy Builds
Royal Nightseye purple 80%
Luminous Noble Topaz orange 76%
Teardrop Living Ruby red 75%
Purified Shadow Pearl purple 55%
Purified Shadowsong Amethyst purple 42%
Teardrop Crimson Spinel red 35%
Teardrop Blood Garnet red 34%
Teardrop Crimson Spinel_35489 red 27%
Luminous Pyrestone orange 24%
Royal Tanzanite purple 23%
Bracing Earthstorm Diamond meta 23%
Luminous Flame Spessarite orange 22%
Shadow Builds
Runed Living Ruby red 115%
Glowing Nightseye purple 71%
Runed Crimson Spinel_32196 red 67%
Veiled Noble Topaz orange 37%
Glowing Shadow Draenite purple 30%
Runed Blood Garnet red 29%
Potent Noble Topaz orange 27%

As usual, the gem names link to Wowhead.

Percentages can be greater than 100 because the same gem can be used in more than one socket per item or per character. That doesn’t matter really. The percentage value just forms a sort of popularity index which can be used to order and rate the options.

It would probably make sense to group the gems by socket colour and/or by the type of stat granted since these are the starting points for deciding what gem to acquire. I’ll add that to the next version of the report.

It is possible to generate other tables that would show:

  • popular gem/gear combinations
  • popular sets of gems per character

I’m playing with queries for both of these but the data doesn’t seem to show a lot of clear patterns. Any suggestions for other ways of analysing gem useage would be welcome.

gearing up

November 5, 2008

I’m reasonably happy with my gear report now and I’m starting to crank out the pages. There are two pages up for priests – one for level 70 and one for level 69. The level 69 analysis is intended to be a guide for leveling players, giving suggestions for the gear to aim for as they pass through the 60-69 bracket. My plan is to post a set of lists for each class: one list for each x9 level and one at the level cap.

I haven’t broken out the PVP players yet, but that is still on the to-do list. Aside from anything else that will get us the x9 twink PVP gear choices which would otherwise get lost amongst the majority PVErs’ stuff.

One problem with x9 lists is that the sample size is very small. My test database has about 380,000 characters in it, but by the time we get to a specific class/level/tree (eg a 69 holy priest) we are down to double figures. For that reason, these x9 lists should just be seen as general shopping lists and the percentage values should not be taken too seriously.

Of course the WotLK event horizon is a potential problem even with gear. Based on what happened with the Burning Crusade, all those lovely purple thingamajigs in the level 70 lists are going to be yesterday’s epics very soon. For that reason I’m guessing that a level 69 list is going to be more useful than a level 70 list for the next couple of months.

Azeroth’s got talent

October 25, 2008

I finally have a way of presenting talent builds that I’m reasonably happy with. But the presentation is a bit… um… complex… sigh… Maybe too complex; I’m not sure. Perhaps you could help me out by having a look at the tables here and telling me whether studying them has lead you to:

  1. enlightenment
  2. befuddlement
  3. a migrane
  4. all three

Anyway here we go…

(Caution: the examples here are from pre-patch-3.0.2 data.)

We start with the basic x/y/z builds for a class and their relative popularity. For example the most popular level 70 priest build is 20/41/0 at about 14%. So we want to know what actual talents are used in that build. And we want to know how many points the average 20/41/0 priest puts into each talent (what I will call the points “spend”). That is, not everyone puts 5 points into a 5-point talent.

So for every popular build, we want a table with these columns:

Tree Tier Talent Popularity Spends

There will be some variation across these columns. 20/41/0 is just three sums and within the 20 and the 41, priests have made various decisions around a basic group of talents.

The most simple case is when there is consensus around a talent. For example, the vast majority of 20/41/0 priests take Unbreakable Will as a tier 1 discipline talent. So we get this:

Tree Tier Talent Popularity Spends
Discipline 1 Unbreakable Will 97% 5 of 5: 98%

Only 3% of 20/41/0 priests don’t take Unbreakable Will. Of the 97% who do take it, only 2% of them don’t put the full 5 points into it. Capische?

Then we have talents where there is less of a consensus.

For example the tier 5 Holy talent Healing Prayers is taken by only 39% of 20/41/0 priests. And those that do take it are split about how many points it deserves: just over half think it is worth the full two points and just under half think that it is only worth 1 point. So we get a line like this:

Tree Tier Talent Popularity Spends
Holy 5 Healing Prayers 39% 2 of 2: 59% 1 of 2: 41%

These lines by themselves are not overwhelming; it’s when they all come together that we get information overload. Note that only talents and spends taken by at least 10% of the sample are shown, otherwise the table would be beyond comprehension.

Here is the full analysis for the 20/41/0 pre-3.0.2 priest: Behold!

Tree Tier Talent Popularity Spends
Discipline 1 Unbreakable Will 97% 5 of 5: 98%
Discipline 2 Improved Power Word: Fortitude 94% 2 of 2: 99%
Discipline 2 Silent Resolve 88% 4 of 5: 46% 5 of 5: 34% 1 of 5: 12%
Discipline 2 Improved Power Word: Shield 34% 3 of 3: 90% 1 of 3: 10%
Discipline 3 Meditation 99% 3 of 3: 99%
Discipline 3 Inner Focus 94% 1 of 1: 100%
Discipline 3 Absolution 21% 1 of 3: 80% 2 of 3: 13%
Discipline 4 Mental Agility 94% 5 of 5: 78% 4 of 5: 13%
Holy 1 Improved Renew 99% 3 of 3: 99%
Holy 1 Holy Specialization 91% 5 of 5: 59% 3 of 5: 16% 2 of 5: 15%
Holy 1 Healing Focus 89% 2 of 2: 97%
Holy 2 Divine Fury 95% 5 of 5: 97%
Holy 2 Spell Warding 11% 5 of 5: 63% 2 of 5: 19%
Holy 3 Inspiration 87% 3 of 3: 89%
Holy 3 Holy Nova 23% 1 of 1: 100%
Holy 4 Improved Healing 99% 3 of 3: 99%
Holy 4 Holy Reach 75% 2 of 2: 93%
Holy 5 Spiritual Guidance 98% 5 of 5: 94%
Holy 5 Spirit of Redemption 94% 1 of 1: 100%
Holy 5 Healing Prayers 39% 2 of 2: 59% 1 of 2: 41%
Holy 6 Spiritual Healing 100% 5 of 5: 99%
Holy 7 Holy Concentration 92% 3 of 3: 91%
Holy 7 Lightwell 15% 1 of 1: 100%
Holy 8 Empowered Healing 100% 5 of 5: 96%
Holy 9 Circle of Healing 99% 1 of 1: 100%

Still with me? Please do take a second and tell me if a table like that is genuinely useful. If it isn’t, I’m happy to have another go at the problem.

Progress is being made here, slowly but surely. I’ve got my class statistics report generator working reasonably well now. There are a few um… “issues…” with WordPress and HTML, but the charts and tables do seem to be readable.

For the priest class, I’ve created separate reports for holy-specced and shadow-specced. For characters below level 70, that just means that I divide them up by the tree into which they put the majority of the talent points they have spent at each level. Most classes will need to be treated like this – splitting them into pages for the key tank/dps/healz specs – for those classes that are not purely one or t’other.

There are tables for priest stats:

  • Priest Health
  • Priest Mana
  • Priest Stamina
  • Priest Intellect
  • Priest Spirit
  • Priest Spell Damage
  • Priest Spell Crit Chance
  • Priest Mana Regeneration (MP5)
  • and… not forgetting… Priest Healing

I’ve included tables for both shadow and holy priest healing, if you want to see what the differences are at each level.

I’ve also started generating reports on talent builds. I’m fully aware that the information in these reports has a sell-by date… perhaps as early as next week… At the moment I’m just interested in finding ways to present the information so that the tables are not too big and boring. There are a lot of builds, but generally only a relatively small number of popular choices. I’m trying to keep the signal-to-noise ratio in my talent reports as high as I can without losing important information. Not sure I’ve got it exactly right yet, but I am getting closer.

holy priest, batman!

October 9, 2008

So here I am, cranking out my nice charts for the class statistics reports when I spot this huge spike in the maximum mana for holy priests. 36101 mana for a level 60 priest!?!? Check this bloke out here.

Does anybody know whether this is even possible? There do seem to be bugs in the Armoury code, but I would have hoped that the numbers were reliable at least. But it seems like at least a fraction of the data is b..s.. This character’s gear and other stats don’t combine to produce 36K mana; he must have doubled up on his Wheaties for breakfast.

Unfortunately, the bottom line is that data mining results can only be as good as the data we’re mining.

Update 18 November 2008. It looks like the armoury data has been reset. There was a suggestion that the WotLK release would be accompanied by a purge of the older toons from the armoury; it looks like Blizz has done that.