http://www.moserware.com/2010/03/computing-your-skill.html
Le me say this:
I spent a bunch of time way way back when trying mightly to understand, to me, the infamous eight page paper from MSR regarding TrueSkill that the author above mentioned and which is the basics of AllegSkill boiled to the most basic of basic parts. Much of my struggle with said paper was while I have training in the field of statistics and probability the level was not sufficient to follow the ninja and demigods of the field's jargom so I did not know that I already know a good deal of the underpinnings. In essence my ignorance of the nuances had me doubting the basics which I did in fact really understand and I had to reconfirm stuff I already knew.
Step in the way back machine with me but first two of many many many faults: Im tenacious when I get motivated and lazy to a fault when I am not.
Way way back when I felt quite certain that we could do a better job of predicting outcomes than the way things were being done at that time. It wasn't the ranking system being used at that time but how Noir was balancing his core based on win/loss percentages between factions. I knew at a fundamental level Noir's approach would only be accurate by pure luck. So I put out the word that we were going to approach this in a systematic manner, check a number of systems, find the best for us (if any), present it to management, and if they were so inclned they could impliment it.
Welp about eight people showed up all with shouter hats and pointers sticks other than one person who brought a shovel. Translation: Everyone wanted to be in charge, give their opinion, and check everyone else's efforts except one who actually was willing to do the work. Being in RL a rain maker (translation: here is a task, here is a date, damn the torpedoes, make it so) I knew how to get'er done but this was different, way different. There were no handles, levers, buttons, motivators, or cattle prods to move things forward when you had roadblocks.
So I learned on the fly that I had made a HUGE GINORMOUS KING-G-G-G sized blunder... we were knee deep in opinion, passion, and drama which was rising fast and honestly no one had a single actual fact to back it up. So I got sneaky and handed out tasks based on each subgroups favorite idea and as the dates kept being missed and no work was availible to review many left and those that stayed became more helpful... anyhoo
After learning something outside of wiki I realized that of the ahhh four-five different approaches we had agreed to look at only one was different and the others building off a common start point, Elo. The lone wolf was an attempt to use various in game actions to develop predictive models via what is/was popularly termed back in the day "Fuzzy Logic". It was my initial sweetheart btw and right in my wheel house as I had a long romance with this kind of approach while doing research in college. People regularly advocate this as "better" which I think they mean will be more accurate. After my efforts I'm not so sure this is the case. Let me blah blah blah.
As far as I know most new vehicle's transmission use some variation of Fuzzy Logic to attempt to tune the car to the individual's personal driving habits in an effort to please them so they keep buying more of the same. The transmission will slowly adjust its behavior to the driver within predetermined limits. *shrug* They have been doing this for some time now so it may be as complicated as one of Jimmy's posts but my guess is they try to keep simple as my pea brain. So they lean on trillions and trillions of data points of expereience predetermine the transmission algorith's knobs and switches response points. What I am incoherently trying to say is to "teach" this kind of relatively simple system (from an inputs used POV) requires a very solid foundation of what is and is not an important input OR an amazingly long curve fit that almost certainly is convoluted by the endless combinations and permutations of the inputs you chose. Even ten inputs is a right proper mother $#@!er and no guarantee that your ten choices are THE approach that works or the many and numerous interactions between the ten points. I leave it to others to attempt to give the number as the detials matter. That is why using in game actions to predetermine the game's outcome seems such a herculean task even Drizzler^Spidey would go "meh whats on TV" if attempted.
There is a bunch more stuff after this but Im just going to stop here for now. I may continue or not but I hope you enjoyed the post more than I did writing it... which won't be to hard
tl;dr monkey and monkey cheese were used in creating this post
