Since the forums are restarted I'll just ask the basics.
I "get" the ELO concept. I would like to help with the autobalance and ELO reboot if you can use it. I can muddle through C++, and VB but am absolutely no coding wizard. I wrote many hundreds of pages of printed out code but it was all self taught and it been a few years too. Also Ive got the full blown MS Dev Studio but it's got to be at least a version old (6.0).
I do understand numerical methods, statistics and such which may be an area I can be of use.
Before I start shooting from the hip can someone give me the low down of the details? That's if you need/want/like my help
MrChaos
ELO
*nods head*
OK missed it.
Have you considered using something more like MS Research's approach? Using something which uses mean and variant to adjust scores or at a minimum adjust teams.
The number of overall wins is the basis of the scoring. The confidence interval is 4 sigma (99.99%). It takes everyone's stats into account before handing out points to each individual independent of each other. Nasty stacks wins can actually cause a very minor overall score decrease since the minisicule point award is countered by the larger sample size.
It's inherent weakness is that they start a newbie in the middle of ranking (25 in this case) assuming a skill level but this quickly changes after a few games by the ginormous variances.
A bit of a think to solve the newbie gets points until 8 is needed. Since Im not sure how this can skew things. This maybe where always assume a newbie player's skill upon popping his number. Let me chew on if you want.
The math is straight forward as is the methods used. My final concern is the time to reach reasonable variance numbers.
What you think?
MrChaos
OK missed it.
Have you considered using something more like MS Research's approach? Using something which uses mean and variant to adjust scores or at a minimum adjust teams.
The number of overall wins is the basis of the scoring. The confidence interval is 4 sigma (99.99%). It takes everyone's stats into account before handing out points to each individual independent of each other. Nasty stacks wins can actually cause a very minor overall score decrease since the minisicule point award is countered by the larger sample size.
It's inherent weakness is that they start a newbie in the middle of ranking (25 in this case) assuming a skill level but this quickly changes after a few games by the ginormous variances.
A bit of a think to solve the newbie gets points until 8 is needed. Since Im not sure how this can skew things. This maybe where always assume a newbie player's skill upon popping his number. Let me chew on if you want.
The math is straight forward as is the methods used. My final concern is the time to reach reasonable variance numbers.
What you think?
MrChaos
Ssssh
" border="0" alt="cool.gif" /> PookPook wrote:QUOTE (Pook @ Jul 2 2006, 10:39 PM) Well if you would like to put together something based on the inputs I show above, feel free.
It doesn't have to be T-SQL, pseudo-code is just fine if T-SQL isn't your forte.
Ab-so-lut-ely in T-SQL (got vanilla SQL reference book so should be fine), using your variable names too, and/or with appropriate "hooks" for implementation. *nervous gulp* /mrgreen.gif" style="vertical-align:middle" emoid=":D" border="0" alt="mrgreen.gif" />
It would be better to test first and tweak before publishing. Being an engineer data is near and dear to me. This would be statisically based and we could check assumptions before implementation. I'm pretty sure over two team games will require integration, matrix algebra, and further research so maybe a group effort would be in order for multiple teams stat updating. Also groups work better so a math guy like ClayPigeon and maybe a real honest to God programmer (my assumption is your just that L-A-Z-Y that opening a new forum, being benevolant co-dictator, working on the core, and FAZ is all you greedy, selfish ass will give us /mrgreen.gif" style="vertical-align:middle" emoid=":D" border="0" alt="mrgreen.gif" /> < I $#@!ing hope he got the joke hmmm I'll add another one just to be sure > /mrgreen.gif" style="vertical-align:middle" emoid=":D" border="0" alt="mrgreen.gif" /> < and a j/k > j/k ) would be nice. If not I'll proceed and make (hopefully not actual) @#(! happen. I noticed Tigereye has an algorithm which will compare as well.
How soon Pook? Two Weeks© D-U-H
BELOW IS DETAILED DISCUSSION OF CONCEPT (YOU'VE BEEN WARNED)
Looking for the answer to a cadet test question I stumbled across a six page discussion on the rebalance issue which got the gears turning a bunch.
Skill to time played i.e. returning vet (Bacon concern) gets addressed by "TrueSkills" in that a fresh non8 being on the winning team errr 8 times in a row (assuming normal distribution i.e. no stack either vet or rookie) has a 0.4% possiblity of happening by chance. So returning UberTeamWhore Vet (Viscur) very quickly gets adjusted mean (ELO < which I strongly suggest we rename >) but has hellish variant (uncertainity < chaos factor /blush.gif" style="vertical-align:middle" emoid="
" border="0" alt="blush.gif" /> >). In order to do a fair balance you'll need both.Some serious thought needs to be given to how a player can make choices. Im beginning to think a AZ/FZ type deal were in order for stats to be counted (both personal and ELO) have only a limited number of possible choices based on ELO and time in global NOAT.
If you will unbalance by X amount OR don't bring balance closer by Y then you wait with X and Y growing larger and smaller respectively every Z increment of wait time.
This would also include a confidence interval of OHMYPOOELITTTLEHEAD based on the variance (maybe that 8 wins in a row newbie IS just one lucky Mofo) too but I'll refrain for now other then to mention the higher the confidence generally the longer the wait.
Yes eager Alice in the back row it's still possible to not play for an asshat commander or with that jerk MrChaos all you got to do is waaaait your options will open including the ability to propose a game.
Can't wait because you Ritilan dealer is in the Bahamas? Then go ahead and do a no stats game. Vet stack, rookie stack, unstack, restack, cap ship scum, rerererererebase cap scum, pod eject for distance, or maybe just don't give a $#@! and plaaaay. Just remember each of those newbie flying a scout kills while sitting in a Cap turret won't be recorded for history.
Interestingly enough small games are a quicker indicator of actual skill in team based games according to theory. The thinking is that your influence on outcome is larger since (not to get too technical) there are fewer of you on the team. So some thought could be given to the ten player minimum all game requirement for stats counting. Making NOAT waiting shorter.
Other items needing to be addressed would be; what happens when the unthinkable occurs and someone drops either by accident and on purpose (penalize/partially credit loses/wins on purpose and hold slot for z time when accidently). Developing commander stats (snot hard at all), dealing with ridiculous level of childish booting (not easy).
My thoughts Eager Alice? You boot ANYONE for ANY reason you get an automatic lose with automatic review by Booting Board (Senate?) which decision is based on RoC. Disagree? Don't play stats based. You get general boycott to playing stats and the Senate reacts, you get rules change. Easy.
The two main items holding back growth is game experience and well game experience. Better known as stacking and booting. Please note Im all for individual stats but, at least initially, their not contributing to rank. Eventually who knows. It may turn out some strange stat will be the banana pricing for Allegiance.
Huh? Oh right sorry.
A while back when data mining concept was being sold the public I heard a story that the lower the price of bananas the more stuff people bought at supermarket. No matter if they wanted, needed, or even bought the darn things. See? *shakes head* *sigh*
MrChaos