Play
Multi-Player
Coins
Community
Settings
Help
Community   Maps   Forum   Mail   Ladders   Clans   Recent Games
Sign In | Sign Up
<< Back to General Forum   

Posts 1 - 11 of 11   
Rating system questionable?: 1/9/2016 23:51:17


Psychonaut
Level 57
Report
Just beat a Poon Squad member in my last RT game. My rating went down. I understand getting less increase when playing lower players but to lose rating points win or lose? That kinda sucks :P
Rating system questionable?: 1/10/2016 03:32:04


Psychonaut
Level 57
Report
Bump because I shall not be ignored .-. Will nobody explain this?
Rating system questionable?: 1/10/2016 03:35:12


Cata Cauda 
Level 57
Report
You know stats can be out of date, up to 30 mins... So maybe you got beaten before but the server didnt show your rating change fast enough?
Rating system questionable?: 1/10/2016 03:35:18

MasterOfWarZone
Level 42
Report
It's based on the total # of games, and average rating of said opponents, and win %. Thus playing bad people lowers average rating of your opponents. It is even more prevalent on the elo-based ladders like 1v1/2v2 than on RT.
Rating system questionable?: 1/10/2016 04:10:33


Psychonaut
Level 57
Report
I understand that but still why do I lose point when winning against certain players haha. I can understand only gaining like 1 point or even 0. But not negative :/
Rating system questionable?: 1/10/2016 10:43:07


Buns157 
Level 66
Report
It's because they're so poorly rated with a small standard deviation. It's a flaw of the system, but at least at your rating it doesn't happen as much.
Rating system questionable?: 1/12/2016 06:42:51

wct
Level 56
Report
My guess, without actually checking the details, is that your rating's *mean* increased slightly, but your *standard deviation* had a bigger increase. And Buns' explanation fits the same conditions I'm describing. If their rating's mean is much lower than yours, then after the rating update, your rating mean will only increase a little bit. The adjustment to your rating's standard deviation is a little more complicated, and I can't even remember exactly how it works, but if your opponent's standard deviation is 'extreme' (maybe quite big, maybe quite small), that might cause your standard deviation to actually increase, even though you beat them. This is especially true if your standard deviation is not very small to begin with. This seems to be the case for you currently.

The TrueSkill *estimated* rating is calculated as: mean - (3 * deviation)
This is to make sure the estimate of your 'real rating' is 99% likely *not* to be an over-estimate. But by subtracting three times the standard deviation, it only takes a relatively small increase in your s.d. to reduce your rating even if your mean has simultaneously increased a small amount from a win.

The more games you play, the tighter/narrower your s.d. will become, so that slight reduction in the *estimated* rating will be soon overcome.

Example: Before the game, your rating might have had a mean of 1000 and a s.d. of 150, making your estimated rating = 1000 - 3 * 150 = 1000 - 450 = 550.

If, after you win the game, your mean increases slightly to 1005, but your s.d. increases slightly to 155, then your estimated rating will actually decrease to a lower rating = 1005 - 3 * 155 = 1005 - 465 = 540. Which is down from 550, even though the mean 1005 has actually increased from 1000.

If, after several more games, your rating's mean is still 1005, but your s.d. decreases even just a little bit from 155 down to 151, that would make your estimated rating = 1005 - 3 * 151 = 1005 - 453 = 552, which is now finally higher than your initial estimated rating of 550.
Rating system questionable?: 1/12/2016 07:26:16


Sułtan Kosmitów 
Level 63
Report
@wtc great! i was always annayed when losing points due to winning but that means that it could still be profitable. It depends on how your mean is calculated, so this is my question to you, is it possible it doesn't rise after a won game or even decrease?

I understand you probably don't know that and would have to check just as me but I am a lazy man and I believe in your desire to help and would really apreciate it if you answer. :)
Rating system questionable?: 1/13/2016 02:42:08

wct
Level 56
Report
It depends on how your mean is calculated, so this is my question to you, is it possible it doesn't rise after a won game or even decrease?

If we're talking about 'basic TrueSkill', which is what I will call the version described by the whitepaper on it, then I'm pretty sure that a win always increases your mean.

However, I'm not sure if WL uses 'basic TrueSkill' or not. And it may be the case that the RT ladder and the 1v1 ladder use different tweaks to TrueSkill.

For example, it's common for your TS rating to change in the 1v1 ladder even if you haven't finished any games recently. I believe (not sure) that this is because the 1v1 rating system recalculates everyone's ratings all at once periodically (like once a day or so, it seems). So, if someone you played recently had their rating changed by a recent game of theirs, then that might affect your rating indirectly. Also, since old games can expire, this can also affect your rating.

I know much less about the RT rating system. I've heard that it doesn't expire old games. That makes me suspect that it *could* use a more simple TrueSkill system, perhaps 'basic TrueSkill'. If that's the case, then I think a win would always increase your mean. But I don't know for sure.

I could be wrong on any of this stuff, so take it with a grain of salt.

I will say this, though: If there is a version of a TrueSkill system where a win can actually lower your mean, I would call that system 'broken'.
Rating system questionable?: 1/13/2016 02:48:44

Omniscient 
Level 54
Report
RT ladder uses Trueskill. 1v1 ladder uses bayeselo.
Rating system questionable?: 1/13/2016 10:44:37


Math Wolf 
Level 62
Report
What wct says is exactly correct. The short summary:

WL uses TS (for the RT ladder), so your mean should never decrease after a win (it may stay the same). The variance will increase if you play against someone whose rating is very different from yours. So even though your rating may go down slightly after 1 game, it still has a lasting positive effect due to the increased mean.

1v1 uses BayesElo, which means that if any of your opponent's ratings changes, your rating will change as well until the game expires. Winning a game always increases your rating if you keep everything else fixed, but other games that finish or expire around the same time may influence your rating as well and may cause a decrease in rating.
Posts 1 - 11 of 11   

Contact | About WarLight | Play Risk Online | Multiplayer Strategy Game | Challenge Friends, Win Money | Skill Game | Terms of Service