Masters events - What???
Posted By: Henrik Bukkjaer In Response To: Masters events - What??? (rew)
Date: Saturday, 28 December 2013, at 1:11 a.m.
In Response To: Masters events - What??? (rew)
Let me clarify:
In DBgF we doesn't rate everything. But we rate close to everything.
At the beginning (25 years ago), we had two lists: one where EVERYTHING was rated (it was called the "All matches list"), and one for official tournament matches only.
The "All matches list" got out of hand (private matches was allowed here, some players even cherry-picked when they got away with it!).
And the requirements for the tournament list, caused that list to suffer a little bit of "too few matches".
Therefore we "melted" the two lists together, and ended up at the following criteria for getting a match rated:
- Official tournament matches only. (This includes the cafe/weeklies listed in the DBgF directory, and side events at tournaments).
- Minimum 4 players participating. (Now a requirement by law in Denmark)
- Minimum 5 point matches. (Now a requirement by law in Denmark)
- All "singles" matches are rated (ie, no consulting doubles matches, but matches played individually as part of doubles or team events are OK).
- Speedgammon are rated, if clearly stated on the invitation AND there's a minimum of 20 sec/point + 10 sec/move.
- Blitz (faster than the time control mentioned above) cannot be rated. - "Closed" club tournaments can be rated if approved in advanced by the tournament committee - typically clubs are allowed to get a couple of closed tournaments rated each year, such as their championships.
- In the future (it has been announced), a separate "Doubles" rating list will be available.
- In the future (yet to be announced), a separate "Speed Gammon" rating list will become available - probably to include both speed and blitz play.
In my opponion, ELO is the best system available to do useable gammon ratinglists. But it requires some things to be in place for it to work, and it has a couple of downsides that can be addressed. It also comes in different flavours:
First of all, it's important to get as much play rated as possible. You need a lot of data in there for gammon, because of the volatile nature of our game. You also need a lot of data in there, to get the lists filled with many players on different levels, and get them spread out over the spectrum (avoiding the master/open/intermediate closed population issues). Rating club play, cafe/weeklies, and minor side-events help this spread.
Secondly, you need to get new players into the system in a good way. And that all depends on how mature your system is. We know the "accelerator" from FIBS, which is fine in a scenario where your system is inmature AND you have top players playing beginners/intermediates in one big mix. But unsuitable for a more mature system like DBgF rankings and for systems where people tend to play players at their own level. I'm a big proponent for inserting new players on estimated start ratings depending on what they sign up for (or previous history), but unfortunately I haven't convinced DBgF to do this. If can be done in many ways, but the "optimal" again depends on your populations, the nature of new players entering the system, and the system maturity.
Scaling match levels. This is also something that should be considered. Are you going to use squareroot functions or linear functions for this, and what are the minimal matchlenght you will allow. I think the DBgF implementation works fine, but again it couls depend on what mix of matchlengths you're going to feed into the system.
Age/decay. ELO isn't good at handling inactive players. But how should a period of inactivity be handled? Well, it depends on how thorough the system is at recording "everything". If you have a system that really covers all competitive play, and a player is inactive, you can assume that he is indeed inactive. On the other hand, if your system only covers big tournaments (and maybe only covers main flights and not and side action), then the player could indeed have been active at some level, but just not competed in anything deemed admissible for the list. The nature of backgammon players coming from a period of inactivity to become active again, is that they are a bit rusty to begin with, but typically not too shabby, and at the same time, they fairly quickly gain their near full strength again. I have a proposal of a "temporary ELO reduction system", that I hope to get incorporated into the DBgF system - or at least, I hope to get it tested with our data and evaluated. The idea is that after a period of X days of inactivity, you start building up an amount of ELO points to be subtracted from your ELO rating, this build-up increases over time, thus your effective ELO is reduced. And should you start to play, the ELO would reflect your current strength. However, when you then start to play, your reduction is then "played away" at some rate, until you have clocked enough experience so you have no reduction left, and thus is back at your "pure" ELO rating (simulating that you have "played your rust off"). Such a system would (if tuned correctly, which I think we can do with the vast amount of data available in the DBgF database) solve the two-fold problem of inactive players (not having them occupy very high rated spots for long periods of inactivity, and not having them rated correctly once they start playing again).
Same solution can be used for newcomers who enters the system, having them start lower than the system average (by way of temporary reduction), and then play away this reduction through experience, allowing their learning curve to have another slope in the early matches (where you typically learn relatively faster) - without all the drawbacks of the FIBS accelerator.
There are so many advantages of using ELO like systems as opposed to tournament results based points systems. And if you can get enough data for each player, there are even advantages over using PR evaluations (I think the optimal assessment of a player considers both PR input and ELO rating).
Again, if the ELO input is thoroughly founded in plenty of data, I would seriously consider to prefer that over PR input, in predicting success at a tournament with a mix of players (strong and not-so-strong). ELO should clearly be best at parameters such as efficiency vs. less skilled opponents, clutch performance, and "conservative" error patterns / giving opponent tough decisions. Whereas PR input will be better to predict the chances of beating a world class opponent in a single game (eg. the final).
Ken has some other aspects of doing his list, that should be kept in mind. Giving players something to strive for, and celebrating tournament achievements. An ELO system doesn't really do that. Eg. you don't really get extra points for winning "important" matches, etc. In the DBgF we've satisfied that need, by creating a results database. For each player you can see the number of results they have achieved, click the link, and see the entire list of results. I think this is a good combination with ELO ratings.
In addition to that, you have the option of scoring "bot-norms", which are also listed on the player profile, so in case you are a really strong PR player, but simply not capable of turning that into tournament wins (for whatever reason that might be), you have the chance to get recognized for your skills any way.
It's then up to the "reader" to determine if they will give more weight to ELO, results or bot-norms (PR).
Messages In This Thread
BGonline.org Forums is maintained by Stick with WebBBS 5.12.