Free Allegiance Forums

Posted: **Sat Sep 06, 2014 8:16 pm**

Phantom032 wrote:QUOTE (Phantom032 @ Sep 6 2014, 10:43 AM) Seriously, am I the only one who actually bothered to learn how AllegSkill works and everyone else is just guessing most of it?

Welcome to allegiance

Posted: **Sun Sep 07, 2014 1:25 am**

Of course you are right Cashto and EXACTLY what I told the group regarding doing such as putting it back through the algorithm. However you can, theoretically and possibly practically achieve 90% accuracy in predictive outcomes. So TS implimentation showed .80% plus when you filtered for game size, stack, and clunkers. If we wouldn't have filtered for stack ( I'm not sure my memory is right but I,think it was 2.5 differential) it probably would be higher... maybe

The work including results was presented to management who said it had to be rewritten in some flavor of SQL rather than C++. It was at this point I bowed out and give no more man in the street details.

The intention was to implement ranks, buff stats and add rewards, and finally i
I can also tell you the settings, assumptions, and such related to Allegiance where always always always meant to be vetted and tweaked. The b also needs care and feeding too. My point is the system works and well but the current implementation probably needs an over haul.

Thanks for reading

Posted: **Sun Sep 07, 2014 4:04 pm**

Bayesian method demands that the players being ranked are the sole responsible party for game outcome. Until that demand is met, the application of that method is inappropriate.

Besides, ranking players in a 3 v 3 game seems ridiculous at this point.

Posted: **Sun Sep 07, 2014 6:02 pm**

Wasp wrote:QUOTE (Wasp @ Sep 7 2014, 09:04 AM) Bayesian method demands that the players being ranked are the sole responsible party for game outcome.

Uh, no. Citation needed.

Posted: **Sun Sep 07, 2014 9:18 pm**

cashto wrote:QUOTE (cashto @ Sep 7 2014, 02:02 PM) Uh, no. Citation needed.

Uh, yes. Citation Given

"The TrueSkill ranking system assumes that the performance in a single match is VARYING AROUND THE SKILL OF THE PLAYER, and that the game outcome (relative ranking of all players participating in a game) is determined by THEIR performance."

...You will never predict the outcome of tomorrows weather in England by measuring the amount of ice on Mars.

You will never predict a players contribution to game outcome by observing the COMMANDER wins/loss or games won or lost due to rock placement or other variables which cannot be isolated (thru larger samples). The games being measured are more greatly influenced by other factors that completely dwarf the players contribution to such a degree that you will never, ever, get enough samples to make an accurate prediction. Many many years now we've seen each and every ranking system fail miserably because of this very problem.

The only way it will work is to remove the commanders overwhelming god-like influence on game outcome, make all rock placement consistent and remove development.

Posted: **Sun Sep 07, 2014 9:54 pm**

Wasp wrote:QUOTE (Wasp @ Sep 7 2014, 11:18 PM) Uh, yes. Citation Given

"The TrueSkill ranking system assumes that the performance in a single match is VARYING AROUND THE SKILL OF THE PLAYER, and that the game outcome (relative ranking of all players participating in a game) is determined by THEIR performance."

...You will never predict the outcome of tomorrows weather in England by measuring the amount of ice on Mars.

You will never predict a players contribution to game outcome by observing the COMMANDER wins/loss or games won or lost due to rock placement or other variables which cannot be isolated (thru larger samples). The games being measured are more greatly influenced by other factors that completely dwarf the players contribution to such a degree that you will never, ever, get enough samples to make an accurate prediction. Many many years now we've seen each and every ranking system fail miserably because of this very problem.

The only way it will work is to remove the commanders overwhelming god-like influence on game outcome, make all rock placement consistent and remove development.

That does not say it needs to be the sole responsible party, just that it needs to have an influence. The more influence the faster the system can find accurate rankings, but there can be other influences. As long as those other influences are random, they only prolong the time until an accurate ranking has been found. Any nonrandom influence (players picking which commanders to fly for) obviously affects the ranking.

EDIT: To clarify, over a lot of games you will get lucky/unlucky roughly half the time, because the influence is random. So with more games its less likely you were especially unlucky or especially lucky. If you fly for random comms, this also applies. For example if you, before you join the server decide "I will play blue now" that is (probably sufficiently) random. If you decide "I will join the team that is down a player" that is not random, as that is greatly affected by nonrandom influences.

Posted: **Sun Sep 07, 2014 9:59 pm**

Wasp wrote:QUOTE (Wasp @ Sep 7 2014, 02:18 PM) Uh, yes. Citation Given

"The TrueSkill ranking system assumes that the performance in a single match is VARYING AROUND THE SKILL OF THE PLAYER, and that the game outcome (relative ranking of all players participating in a game) is determined by THEIR performance."

...You will never predict the outcome of tomorrows weather in England by measuring the amount of ice on Mars.

You will never predict a players contribution to game outcome by observing the COMMANDER wins/loss or games won or lost due to rock placement or other variables which cannot be isolated (thru larger samples). The games being measured are more greatly influenced by other factors that completely dwarf the players contribution to such a degree that you will never, ever, get enough samples to make an accurate prediction. Many many years now we've seen each and every ranking system fail miserably because of this very problem.

The only way it will work is to remove the commanders overwhelming god-like influence on game outcome, make all rock placement consistent and remove development.

There is no connection between the weather in England and the climate on Mars. But there is a connection between player skill and the outcome of the game.

You're right that it isn't 100% determined by player skill, but it doesn't have to be. As long as there is SOME correlation, it can be measured. As P32 mentioned, there is no requirement for the outcome to be COMPLETELY free from exogenous, random factors. There are elements of chance in, say, backgammon, but that doesn't mean the same approach can't be used.

And the fraction of skill contribution is not small. The number of games that can't be won due to miserable rocks is vanishingly small. A good team can, by and large, make it work.

The current system does assume that every player contributes an equal proportion to the outcome of the game, but that can be modeled. It's not impossible to fix.

Posted: **Sun Sep 07, 2014 10:35 pm**

Phantom032 wrote:QUOTE (Phantom032 @ Sep 7 2014, 05:54 PM) That does not say it needs to be the sole responsible party, just that it needs to have an influence.

That does not say it needs to "have an influence". It says that the outcome must vary around the skill of the player...ie...the player must be the responsible party. You make assumptions that a player will somehow provide consistency to those random events you listed by evenly applying those events across his games. This is a flawed assumption because even if that player chose to play for commanders evenly,..the player does not remove or isolate those commanders influence such that the commanders influence can be discarded to reveal the players skill. The game consistently varies around those things which more greatly influence outcome and the player thus becomes more of an unpredictable "random event".

cashto wrote:QUOTE (cashto)You're right that it isn't 100% determined by player skill, but it doesn't have to be. As long as there is SOME correlation, it can be measured.

THAT is the problem. The correlation / dependency cannot be assured or even assumed.

Posted: **Sun Sep 07, 2014 11:10 pm**

Wasp I want to thank you for helping me years ago in the helpline.I never got to thank you because you disappeared. You and Freeb were kick-ass guys. Thanks so much.

Posted: **Sun Sep 07, 2014 11:28 pm**

kerosene wrote:QUOTE (kerosene @ Sep 7 2014, 07:10 PM) Wasp I want to thank you for helping me years ago in the helpline.I never got to thank you because you disappeared. You and Freeb were kick-ass guys. Thanks so much.

Heya Kero,

Always a pleasure to help a fellow pilot....and a great pleasure to see you're still a part of the active community.

Free Allegiance Forums

2 easy things that will make more people play more often