Sorry it took me a few days to post in here. I've been insanely busy and didn't want to post in here without reading everything so far. So anyways, thanks for bringing me in and here are my thoughts on several issues.
Ratings of pools:
Pools should absolutely be rated. I understand that players will drop sets or not try hard and that could skew complete accuracy. However the one thing elo depends on the most for accurate ratings is quantity. By not rating pools, we end up with pro players that tend to stay in brackets longer having pretty accurate elo's, but everybody else all clumped together and only getting 2-3 rated matches per tournament. I wouldn't be entirely against using a smaller K value for pools though. Even then though I think a half K value might be too drastic. Maybe 75% would work. Only way to know on that one is to try it. That's something that could be changed down the road if we felt necessary.
Another thing to consider in pools was mentioned earlier. There's no way to tell the order of the matches. When I do the elo's for my monthlies, I use two different calculators. For pools, I use the USCF Ratings Estimator
and base an individuals matches off his opponents rating from the beginning of a tournament. Basically the same as using a formula that Tirno mentioned earlier: R_A' = R_A + K(S_A - E_A). This way you don't have to know the order of the matches and you still get accurate results. The nice thing about pools that you don't get from brackets is an exact set result, like 2-1, 2-0, 1-2, 0-2. I experimented with the option of giving a player in pools partial credit for a 1-2 loss, but no penalty to the player that won. What I did was give a player with 2 wins in a set one full point, a player with one win a 1/3rd point, and a player with 0 zero points. It's an interesting idea and seemed to help make pools just a little more accurate. The downside? Players catch wind of it and let a friend get that 1 point off and somebody gets a boost. Also players already do that to help friends get out of pools. It has it's ups and downs. What do you guys think about that idea? Here's an example:
Player rated 1200 plays five guys in pools. Their ratings are 1000, 1100, 1200, 1300, 1400. He goes 2-0, 2-1, 2-0, 1-2, 1-2 (3-2 or 3.66-2). By using the USCF Ratings Estimator and a zero for a 1-2 loss the player ends up at 1227. If you give the player a 1/3rd win each for his two 1-2 losses, he ends up at 1280. There's a pretty big difference there and what it helps us to accomplish is recognizing a close set vs one that a player was dominated in. Any thoughts?
For brackets, I chose to use UTDZaks software and manually enter each result. This updates a players rating after each individual match. It is extremely important to calculate these results in chronological order to ensure accuracy. Somebody early explained how to do that: do each of the winners' bracket sets in ascending order, then do each of the losers' bracket sets in ascending order, then do the grand finals. Due to how the brackets work, this ensures correctness time-wise for all players involved in any ratings changes for a given match anyway. So whatever we do the software has to know how to do that.
Another thing I would like to recommend is not using a smaller K value for losers bracket. I understand a player sent the the losers bracket early on plays more matches, but the fact is the players lasting longer in the winners brackets are playing higher rated players for the most part. That should balance that issue out. Here is the problem with using smaller K values for losers bracket: Player A and Player B have the same rating. Player A loses round 1 winners, then wins 5 games in losers before dropping a final match. He gets 5th and ends up 5-2. Player B wins two games in winners and then loses two in a row, one in winners and one in losers, getting 9th with 2-2. By using a smaller K value in losers, Player B could actually end up with a higher rating even if they played opponents of the same average elo. This is just something that should not happen.
Somebody earlier mentioned potentially rating money matches. They should definitely not be rated. I know earlier I said more matches = more accuracy. The problem with money matches though is that a lot of the time it's "I'll MM your Pichu vs my Bowser." Those kinds of limitations on a players choices will do no good in keeping accurate ratings.
I like this idea a lot. We do have a problem though with either using XXYY0000 codes that we create and players have to remember, or using smashboards IDs in which players without smashboards accounts cause a problem (and there's lots of them). What about using the 0000 section to allow the player to chose their particular unique id code? Or using last 4 of SSN? Or phone number suffix? Basically my thought is to use something that the player WILL remember. Just a thought.
As far as backlogging tournaments goes, I personally don't think there should be a limit as to how old a tournament can be. I know with that there are going to be players like Ken who is rated just as high as HungryBox. The fact is though, that elo will reflect metagame. Ken was the best then, and his rating would reflect that. If you look at chess, Bobby Fischer is still the 3rd highest rated USCF player of all time. But what happens if you look up the top 100
? He's not on there. What they do when making these lists is only show active players. USCF keeps a database of all players, active or not. But on their lists they only show active. They use a 12 month system. We could adopt that or use something like 6 months. For my monthlies, I was using 4 months. For our online database, we could have it sortable by "All" or "Active Only." Thoughts?
I will however state that I'm not completely against using a lower K value for tournaments past a certain age. But I wouldn't want to make that point be too soon, like 2011. I'd say Dec 2009 at the earliest would make sense in my opinion.
Another idea I am opposed to. Players come and go but what they have earned, they have earned. If somebody reaches a high elo of say 2000 and disappears for a long time, of course there is a good chance they'll come back weaker. The fact is though that elo can handle it and it'll adjust itself. Just let the math do it's thing. Besides there's no telling if a player will come back out of nowhere stronger than ever.
Just a random thought. Starcraft: Brood War has a program called BWChart that can keep a database of players and their "AKA's." This way, Chillin, ChillinDude, and chillindude829 all get recognized as the same person. Is that something we'd be able to do? Here
is what it looks like.
A Few things that should or should not affect K values:
Of course, these are my own opinions. But I believe K values should not changed based off:
- location in bracket (whether it's winners/losers or finals or round 1)
- tournament location
However, K value could or should be changed based off:
A Few More Random Thoughts:
- Number of previously rated games by a player
- Bracket vs Pools
- Age of event
- Rating of a player (higher rating = lower K value)
If anybody has TioPro, I will need some assistance converting a TioPro file to regular tio. I got the file for Zenith and I can't open it with regular tio.
For TOs that don't use Tio, I'd be happy to convert their files or bracket images.
Elo's for doubles: Shouldn't be any. Chess has a 2v2 variant called "Bug House" and in BH tournaments, teams are seeded by the average elo of their two players. Makes sense to me. But elo's should not be affected by the results of these events, and thus elo is for seeding only in 2v2. Maybe the plugin can have the option to automatically rank and seed players based off the average elo of the two players.
So yeah, those are my thoughts so far. Sorry it's such a long post but I had a lot to catch up on. There was one more thing I was thinking that shouldn't affect K value but I forgot. It's really late so please excuse me if anything I typed up makes no sense at all. Point it out and I'll explain it better lol.