KGS Issue - Rated Games of Any Size
This is a wish that has been rejected; I moved the discussion out of /GameHandling? to clear up that page.
- (Sebastian:) Allow rated games of any size. (Well, at least 6x6). Use a weighing factor proportional to the size. It just doesn't make sense that you don't get a rating as a beginner when you play as recommended. This would be an important incentive for those newbies who currently play only guest accounts because they wouldn't get ratings anyway, to switch to a real account.
- wms: Not a bad idea, but the problem is that nobody knows how other sizewins/losses relate statistically to 19x19 wins/losses. KGS is based on a statistical system, you can't just weight them differently, then you will still have wrong data, just at a different weight. If somebody can find data on % wins for players who are x ranks apart (where 1 rank = 1 stone on 19x19), then I'll be able to include off-sized games, but until then, it would screw up the rating system.
- Neil: If ratings are made easier to get, then they are made less meaningful and reliable. Leave it alone, I say.
- (Sebastian:) This is a specious argument. On the contrary - any statistician will tell you that reliability increases with the number of samples taken.
- Neil: Yes, what I just wrote was very bad. It didn't communicate what I meant at all. What I mean is, right now to earn a rating you have to play full 19x19 games (usually with decent time, but that's another wish). To be able to gain a rating playing 9x9 allows you to get by with fewer skills. That's what I mean by making a rating easier to get.
- (Sebastian:) OK, I see your point. But I don't think this is an unsolvable problem. One way around it could be to set the weight dependent on the rank: Phase it out smoothly between 20 and 10k. (If someone really manages to beat 15k players consistently on a small board then she/he probably is at least somewhere around 16k. This player may be in for some surprises when he/she starts playing on a 19x19, but I don't think this harms the system.) Another could be to apply this only if a player has a ranking with a "?" (and it would not suffice to remove it). In this case, well he may appear a rank or two stronger, but it wouldn't do any harm.
- phenomene: 9x9 or 13x13 go is just another game. Rated games on small boards sizes for 30k-20k players may be a good idea, as they often rush into playing 19x19 without enough knowledge of the game in order to get a rating. But allowing 9x9 to be rated between say dan players is a very bad idea in my opinion ; it is another game, that shouldn't influence one's real go rating.
- (Sebastian:) I agree. That's exactly what I meant by "Phase it out smoothly between 20 and 10k": Use some weighting factor under 20k, and let it be 0 above 10k, with a nice transition in between. However, I'm not sure how much effort it would be to implement it. If it is more than wms can presently afford then I still would hope he finds time for the basic feature. As I said above, I don't see this difference as an unbridgeable problem.
- DrStraw: How feasible would it be to implement separate ratings for each of 9x9, 13x13, 19x19? The mechanics of rating should be easy but displaying the ratings would require some manipulation of the user info screen. The biggest problem would be determining the difference in strength for each handicap stone but there are commonly used standards which would be used initially (3k/stone for 13 and 5k/stone for 9). KGS could blaze the trail for comfirming (or contradicting) these numbers by permitting these ratings and periodically adjusting the factors to keep these ratings in sync with 19x19 ratings.
- (Sebastian:) I like this idea. Internally and on a user's home page, keep separate statistics. In the "ranking tag" (the short "[25k?]" added to a name and used for sorting) and the graph just display one value (This can be a weighted average or the most secure and significant of the three, with bias for 19x19. Conversion takes place at current "exchange rate".)
(Additionally, it may be a good idea to show in the ranking tag if is based on a small board rank. This can be either with an "s", similar to the proposed "g" for guest accounts, or simply with a "?").
- Migeru: The user could select which rating (s)he wants displayed by simply changing the sort criterion for the player list ("sort by rank" would be replaced by three different "sort by size rank" options). When sorting alphabetically, display no rank. Color-code the different ratings on the rating graph.
- wms: As Dr. Straw pointed out, the biggest problem with multiple rating systems for different sizes is displaying the information. Having one rating system, which is for standard sized boards, is much simpler and fits the needs of 99% of the people, so adding something more complex (whether it is making the rating system some bizarre 19x19 and 9x9 hybrid or adding multiple rating systems) doesn't seem like such a great idea to me. Rather than having multiple rating systems, I'd be more tempeted to have a tournament-like ladder system for 9x9 or whatever size you wanted.
- joelr: On the other hand, it sounds like KGS contains all the data and most of the code to do this off-line. You just run the rating routine on only the 9x9 games, and then see what scale factor (rank per stone) gives the best fit to the 19x19 ranks. Even if the results aren't shoehorned into the current display, a summary could be published as a service to the community.
- Chris Hayashida: The problem with extrapolating statistics from free games (9x9 or otherwise) is that the play doesn't necessarily indicate serious play. Just looking through 9x9 games and results, you'll often see B+R simply so that one player doesn't have unfinished games. It's a free game, so what does resigning matter? Trying to shoehorn that into a ratings correalation between 9x9 and 19x19 is difficult, as there are way more variables. Handicap stones will end up being some weird fractional value. The system just becomes more inaccurate as more variables are introduced. It's like trying to shave while on rollerblades on a moving train. Besides, I'd rather have wms's time be spent on something more useful.
- What about giving a not established ranking to beginners based on boards of every size. Once you have played enough games on a 19x19 board to get an established ranking. From then on your ranking is only calculated by those games. In that way beginners would have a non-established ranking to have an idea of their strength. On the other hand established rankings would not at all be affected by games on smaller boards. Would that be a way?
- what about this approach: allow rated games of size 9, 13 and 19, and calculate separate ratings for these. java client displays only the 19x19 rating (but size 9 and 13 ratings are used internally for pairings in these sizes!). 9x9 and 13x13 rating graphs and info are only accessible through the webpage. (and for the ratings, number of games they depend on should also be shown there: e.g. "9x9 rating: 4.4k (7 games)"). (if only one additional rating is added in a first step, i vote for size 9)
Does a realtime Go server with separate rating(s) for non-19 size(s) exist? (does anybody know?)