|
|
| Author |
Thread Statistics | Show CCP posts - 0 post(s) |
Boot Booter
Titans of Phoenix
Damage LLC
1
   |
Posted - 2015.10.21 14:50:00 -
[1] - Quote
What are the different factors going into determining a players MU? Is it a raw value or is it a ranking system? How rapidly can a players MU change?
Statistics below... Beware
I have a theory about why matchmaking is so poor. Here's my logic. A players actual skill and their MU ranking will not always be proportional. Some players will be better than their ranking shows and some will be worse creating a scenario of a distribution of actual player skill at a given MU ranking. Now this is OK, but my theory is that this variance increases as we move towards newer players. Conceptually, vets have played many games (higher sample size for MU) and therefore their MU will be more accurate (smaller variance of actual player skill at a given MU value). Meanwhile, newbs who have less games under their belt (smaller sample size) will experience a greater variation in actual skill at a given MU value. Combining this together creates a scenario of less confidence in actual skill as we move towards newer players (lower MU rankings).
Ok, hope that made sense, but why should we care? Ever say to yourself or your squad "we could've won if the blues were better" or "they had better blues" and scrolled down the leader boards to see the bottom five players on your team achieve less than 5 total combined kills? Well my theory is that this lower confidence in actual skill at lower MU ranks contributes to lower confidence in creating balanced matches. In order for matches to be balanced we must have more or less equal confidence in actual player skill at any given MU rank.
Now I have a potential solution to this in mind but I want to check to see what is actually going into MU. This is just a theory after all. I could be completely wrong, or for all I know CCP has already considered this. All I know is that statically speaking there's some truth to the theory.
Thanks for reading |
Boot Booter
Titans of Phoenix
Damage LLC
1
|
Posted - 2015.10.22 02:47:00 -
[2] - Quote
Thanks for the clarification. First of all for anything I said to be true, MU would have to be a lifetime value. Each match a player played would count. Also I find win/loss to be a very poor metric because of its inherent uncertainty (you could be a very good solo player yet lose a lot due to the poor matchmaking). I like kill/death and wp/death the most, I think those are pretty clear cut metrics for players actual skill. Next I would normalise each players kill/death and wp/death by their respective percentiles. For example if you're a beast logi and a mediocre slayer you might be in the upper 95 percentile for wp/death and the upper 50 percentile in kill/death. Meaning that your wp/death is better than 95% of the player base etc. In this example you'd have a combined MU score of 145. The best possible MU score would be close to 200.
Cool, now we could add some other metrics in. However like I said previously, the newer you are the less confidence we should have in you're MU score due to a smaller sample size. I guess you could think of this as standard deviation in kill/death and wp/death. Newbs in theory should have a higher standard deviation. So based on the standard deviation in your kill/death and wp/death you'd have a final MU score with a standard deviation. For our example, perhaps our beast logi is a 145 MU with a standard deviation of +/- 2 MU. Conversely a newb might be a 50 +/- 20 MU.
Now imagine the scenario with a bunch of Newbs with the MU score 50 +/- 20. Some could be 60s or 70s in actual skill while others could actually be 10s, yet to the matchmaker they all appear to be 50. I believe this is what causes poor matchmaking. The question is how to incorporate a standard deviation or variance into MU and the matchmaker.
Well i haven't ironed anything out yet but my initial thought to solve this problem is to build matches in a way that...
1) minimises total match MU standard deviation (we don't want to build matches with extremely high uncertainty in their outcome)
2) matches each teams total MU standard deviation (we want to build teams that we are equally confident in)
3) matches each teams total MU (obviously what we currently do)
It's tough to really say more without any data but I think this is the right way to think about balancing matches.
Thanks again for reading 
|
Boot Booter
Titans of Phoenix
Damage LLC
1
|
Posted - 2015.10.22 02:53:00 -
[3] - Quote
I'd like to summon ratatti to this thread to see what he thinks about all this MU talk |
Boot Booter
Titans of Phoenix
Damage LLC
1
|
Posted - 2015.10.22 02:56:00 -
[4] - Quote
Shaun Iwairo wrote:Boot Booter wrote:I'd like to summon ratatti to this thread to see what he thinks about all this MU talk To be fair he can't really talk about it.
Well poo, hope he reads at least. I think im onto something here. |
Boot Booter
Titans of Phoenix
Damage LLC
1
|
Posted - 2015.10.22 13:11:00 -
[5] - Quote
HOWDIDHEKILLME wrote:Mu won't work as long as people leave battle... Maybe leaving battle should increase your mu for being a cherry picking puss?
Maybe people wouldn't leave battle if they weren't so consistently lopsided. Maybe there's a flaw in MU |
|
|
| |
|