All-Star Rankings

Greetings Kihra,

Could you please reconsider how all stars points counting works?

Right now #1 is always at 100 while everyone else’s score is being calculated from that player. If that #1 player uses super bursty comp (20 seconds kills for example) the DPS difference is gonna be so huge everybody score will be lowered down to the ground. It just doesn’t feel right.

Is it possible to change the formula to calculate the base from something like the average DPS of the top 10 all-stars instead? And you’d probably wonder “What happens to players above average?”, well you could use about the same formula you’re using right now, calculate the difference between average and their higher DPS, and just assign the value as extra point above 100.

I feel that way everyone wins. The players that took the time to setup and use super-bursty comp/spec (#1) gets extra point, while players using normal comp gets their score normalized. Also with legion coming the numbers are gonna go higher, this could make for big differences in #1-100 DPS depending on setup.

This thread from the old forums is relevant.

http://forums.warcraftlogs.com/viewtopic.php?f=15&t=4787

TLDR: I recognize there is a problem (mainly caused by how much we outgear the content in question) and am brainstorming how to address it.

Here are some ideas (thanks, Theck!) for transformations that could handle outliers more gracefully. The sample graphs show what happens to the distribution of points when the top outlier is 2x the next highest score.

Log distribution smushes but also affects middle of the pack fidelity:

Putting a coefficient in front compresses the distribution:

This changes the formula to apply log directly to your dps and max dps, skipping the normalization beforehand. It squishes the distribution while still punishing parses of 0 with 0 points.

One that looks really promising is to raise the normalized score to a fractional power. This compresses the distribution, but not as much. It’s still bounded by 0 and 1, and I can pick a power that is appropriate depending on how much I want to make DPS work to raise their score.

More thoughts:

A problem with approaches that only manipulate a DPS/maxdps distribution is that the transformation of any score doesn’t really care whether or not there was an outlier or not. For example 0.5 in the “root-1/4” approach will become 0.84 regardless of whether or not there was an outlier at the top. In other words, if I perform badly and do 50% of the top dps (which is not cheesed) and my rank position is around 50% as well, but someone else performed really well and got 99% rank position (e.g., they are rank 2 out of 100), but only did 50% of rank 1, both of those people would get a score of 0.84.

I’ve tried two approaches in the past.
(1) DPS % - This formula was 100 * (your dps) / (rank 1 dps)
(2) Rank Position - This formula was 100 * (total ranks - your rank + 1) / (total ranks)

The first one suffers from the “outlier” problem we’ve been discussing, where cheesing a fight and getting way ahead on DPS can push rank 2 way down.

The second one suffers from two problems: not enough differentiation at the high end (all competitive players have 99 points), and penalizing too much when DPS amounts are close together, e.g., if the top 50% are within 10% DPS of one another, i.e., there can be a “bunching” problem with DPS being too close to one another.

So maybe a hybrid approach that combines (1) and (2), e.g., 0.5 * X + 0.5 * Y, where X = the formula in (1) and Y = the formula in (2) would work well.

1 Like

All-star score points feels esthetic, I’m attributed points depending on my DPS based on max DPS, adding them together and given a final rank number. If the maxDPS one week suddenly becomes a 2x or even 3x outlier, my rank number most likely won’t change, the value of my score will. Where I was rank 1 with 1000 points I still stay rank 1 with 920 points. It’s just a little frustrating to see a big drop.

The suggestion about changing the maxDPS from first to average was from the top of my head including a little discussion with some guildmates but yeah we’re no Theck. We just wish we were able to compare ourselves against other class/spec in all star score value but outlier difference makes it impossible right now (same rank with a score difference of over 250 points). I just hope whichever solution you’re gonna decide to implement will allow us to have fun doing that.

Thank you!

So here’s another brainstorm idea designed to address all the concerns so far. People seem to use All Stars for two things.

(1) As a way to simply see a single “aggregate” score to get a feel for how well someone is doing across all bosses. This person doesn’t care about all the top end cheesing, and really doesn’t want to see a score suddenly get cut in half just because rank #1 did something crazy. This person wants to know “Is this hunter good?”, so they’d be interested in combined class scores as well.

(2) For the super top end players, it’s a leaderboard for the best of the best. To this person, anything you can do to score more DPS is relevant. This person wants rankings to be combined even further, by role (tank, healer, dps), and the goal is to allow the top end to really break away from the rest of the pack.

So I got to thinking, what if I simply have two measurements instead of one?

(1) would use rank percentiles for points, so e.g., 99th percentile gets 99 points out of 100. A player would still have two scores as they do now, the combined score by class and role (e.g., hunters, rogues, mages) and the spec score for class and spec.Instead of being “All Stars” this would just have a more boring name like “All Bosses” or “Aggregate Score” or something.

(2) would use the scoring system that is in place now, but it would be based off absolute DPS by role. So for example, Arcane Mages would dominate a WoD list on the damage dealing role for this system. This would be the new “All Stars”.

The other thing I would fix with both (1) and (2) is never to combine unique roles. Damage dealers, tanks and healer would all be kept separate.

Another brainstorm:

(1) Make everyone get points based off rank %, but with 99+ getting 100 points. 50th % would get 50 points etc.
(2) Award bonus points only to those who achieve 99% or above. 20 possible bonus points. The # of bonus points is computed as a % of rank 1’s DPS.

This allows someone to distinguish themselves without blowing out the All Star Score just because they did some kind of great cheesing on a single fight.