Lot’s of small little projects happened over the last few months. As stated previously, I’m taking a sabbatical from large pinball projects. I added some functionality to the MPF interface which Jan merged into the MPF project. I can’t even remember what that functionality was at this point. (Maybe support for switch matrices? Yeah, I’m going to go with that.)
In the last month, I’ve been writing a Qt application to optimize picking groups during pinball league nights. I wrote up a whole article on why the method is better than what is currently out there including academic papers on the math behind it. I then went through the math to figure out the margin of error of determining the best player. Yeah, I’m nearly certain only Bowen would have been able to follow it, or all those statistics heads that read the blog (well, I doubt there are really any statistics majors reading this blog). I was using the New England Pinball League (NEPL) attendance at Pinball Wizard Arcade (PWA) to see if the new method really made a difference. After doing the math, it became evident that enough games weren’t being played during the season guarantee that the best players were placed in A division. The method did improve the quality of the player that was placed in division A, but with the limited number of meets, it couldn’t guarantee that the absolute best players were placed in division A.
An easy way to understand this would be to think about the worst case scenarios. Let’s say you have a league with 80 players and they attend every meet. (That makes it easy because it is 20 groups of 4 players each). Let’s say that 40 of the players are the Bowens of the world (excellent players), and 40 of the players are the Hughs of the world (sucky players). Even though it is unlikely, when randomly assigning groups, all the Bowens could always be placed with other Bowens. When the groups are split into A, B, C and D divisions, all divisions would be made up of half Bowens and half Hughs.
So the algorithm I worked on reduced the number of times that a single player will replay another player within a season. It makes the above situation less likely, but it does not eliminate the possibility. The only thing that can really eliminate the possibility is to insure that there are at least 20 meets in the season to guarantee that everybody gets to play everybody else. As per the current NEPL rules, only the first 5 weeks determine which division you will be placed in. So at the most, you will only play against 15 different opponents. (With randomly assigning groups, it can be fewer players than this).
Just to state that more succinctly, if there are more than 16 people playing at a location, even with the best group picking algorithm in the world (but not relying on past wins/losses), the break down of the divisions cannot quantitatively put the correct people into the correct divisions.
The algorithm that I created reduces affects of the randomization, but it can’t overcome it because of the limited number of meets. So basically, the algorithm is better, but there is no way that it can be perfect.
I guess it is like baseball to me. I hate the whole idea that an umpire can call pitches, and “open up” or “restrict” the strike zone. It drives me insane. A computer can do a much better job, and there wouldn’t be the fear of umpire bias, or bad calls. My wife likes the “human” aspect of the umpire calling strikes and balls, but I say, bah, humbug.
Anyway, the Qt project is now up in the repository. With all of the above issues, I don’t know whether it is really valuable, but it was fun to write.