631
Announcements / Re: Write leagues system from scratch
« on: February 02, 2011, 03:57 AM »In your opinion both players should gain/lose 0 points and stay at 1050, because this is the expected result.
Yes, IF the actual result of games is the same as expected result, then rating change is exactly 0. If a player played according to his skill/rating, then his actual skill didn't change. Sounds logical?In my opinion both players' real skill level is 900. When they play a draw against each other they 'confirm' that they are equally good, namely '900'-good and not the 1050 that happens to be their season rating. Thus, it doesn't seem illogical to me, that both players drop a little in season rating (to resemble more closely their real skill-level of 900).
It is illogical. Season rating shows us how well a person did IN A GIVEN SEASON. Overall rating is supposed to show player's skill in the long run, over multiple seasons, over his entire league 'career'. These are two different things, don't you think? So why do you want player's seasonal rating to DROP DOWN when he is doing like expected?
I repeat again, if you do better than expected, your rating should rise. If you do worse, your rating should drop down. If you do as expected, it should stay the same. I hope you now understand why I have problems with your proposal.
I think all of your plans for changes come from the idea that seasonal rating should eventually approach overall rating. Did I get you right? But as I said, seasonal rating (why shouldn't even call it rating in the first place, but rather "performance", "score, or smth) should show us how players perform in a given season, not show the skill, that's what we have overall rating for.
I agree that seasonal ratings should be kept completely, 100% separate from overall ratings. If overall rating is used at all in the calculation of seasonal rating, then seasonal rating loses its importance, and becomes something more like quarterly performance. And this would again impact activity, which is about the highest concern we should have for a league.
If, as franz implies, assigning more importance to seasonal ratings diminishes the importance of overall ratings, then so be it; it's still a boon for activity and competition. Overall ratings will still be useful to look at, for those who want to look at them.
The essence of zero-sum is that the sum of points gained by winner and loser is zero. This is not a necessity IMO.
Think again what is rating supposed to show you - the skill. So after a series of games, say 10, the result is as expected (according to player's ratings), for example 8-2. The result tells us that both players played ok, according to their skill, so the sum of rating changes is 0. If it wouldn't be, the ratings would change whereas we said players performed as expected. Don't you think this distorts the picture a bit?
I think a zero-sum system isn't ideal, for worms at least. It seems like your problem is that non-zero-sum systems wouldn't be quite as objective, which is slightly true, right now you can have a good feel for a player's skill if you know they're at 1800, or 1600, or 1400. However, as a counter example, table tennis ratings are wonderfully accurate and employ a non-zero-sum system, as shown here: http://www.ratingscentral.com/HowItWorks.php. And I always understand almost exactly how good a table tennis player is just by their rating there. This system allows young unranked prodigies to gain say, 400 points for beating a high-ranked player, without that player going down too dramatically.
And over at NNNL, although everyone's points inflate as the season goes on, top players end up with roughly the same upper limit on their seasonal rating. You can definitely create non-zero-sum systems that yield ratings that correspond to intuition.