<< Back to Ladder Forum   Search

Posts 21 - 23 of 23   <<Prev   1  2  
Season II matchmaking: 3/30/2012 16:59:37

Dr. TypeSomething 
Level 3
Report
What about adding a random component to the matching algorithm? Right now it tries to pair you with the person most closely ranked near you. What if there was a random number generated, for example, between -20 and 20 that determined if you would play anybody between 20 above you and 20 below you. Sure some people would get lucky, but over 20 games or so the luck should approximately even out. But basically it would still solve the problem of the very best playing the very worst that happened in Season 1, while adding some chance for mobility. I think there is something like that right now because some of my games have been with players ranked a few spots above me, but the variability could be increased.
Season II matchmaking: 3/30/2012 16:59:49

[16] Jasper 
Level 52
Report
Let me start off by saying that I am not and have not been playing in this ladder, but I was reading I feel I know what some of the problems are here.

First off, the system used to decide ranks is an ELO system. This is a very sound system, however, it isn't exactly used well here.
What the ELO system does is to quantify someone's skill level by giving it a number. Then, what it does is give an approximation of how much chance a player has to win against another player based on the relationship between their ratings and based on that, the points will be given and lost after the game. As such, if your skill level stays the same, so should your rating, and though you lose some games against worse opponents and win some against better opponents, your rating should stay the same.

This system is based on the assumption that the rating of players is accurate. Sure, there may be a slight difference and that will compensated over time by the system, and people's skill levels is supposed to change over time, but if the ratings aren't approximately correct, the system doesn't function.
An example of this would be when in chess a number of players was advancing faster than the ELO system was able to reflect, so they played against opponents that were weaker than them, but because their rating hadn't caught up with their actual skill level the system viewed them as weaker than their opponents. This worked out well for the players who were advancing so fast, as their ratnig went in the right direction. However, their opponents were punished much harder for the loss than they should have been, so their ratings were no longer accurate compared to their skill level. As you may be able to see, the ELO system is a very delicate system that can be tipped out of balance by simple things like that.

Looking at things that tip the balance, we should also look at new players. You will be unable to represent their skill level accurately. They are given a certain entry points in rating. This won't be an accurate representation. However, because relatively few players are new and most players have established ratings that are correct (more or less, anyway) they will settle in at a correct level eventually. Oftentimes the system is also manipulated into recognizing that new player's ratings aren't accurate, by for example, calculating their opponent's losses and gains differently (a different k-factor) for their first games.

With the way the ELO system works in mind, it quickly becomes clear that starting everyone up at the entry score at the same time is going to cause some problems. Yes, you would be getting a decent system eventually, but it would take time before everything is settled down. And that doesn't happen in two months, I am afraid. As such, the rating system doesn't function as it is supposed to when used for the seasonal ladder, which lasts only 2 months. For such short periods, it would be a much better idea to look into rating systems which are meant to be used in single competitions (rather than lifetime-systems, like ELO).

(As an aside, Fizzer also makes a point about inflation on the [wiki page in question](http://wiki.warlight.net/index.php/Ladders). While point inflation is something that happens in ELO rating systems, so is deflation. The thing Fizzer is talking is about something else, though. It is all about the system not having settled in yet. Ratings aren't accurate yet, which means that the entry points given to new players also don't match the average skill level of players. Not yet, anyway. Over time the system will settle and that anchor point that is defined (the points given to new players) will be lined up neatly with whatever the average skill level of new players happens to be. Until that time, there will be inflation to get to that point. Once settled, this inflation will stop and you will have a stable system. However, this does not happen in the warlight rating system, because it has a point drain to no longer taking into account older games. As such, you will have a system that is not very likely to settle at all, and in fact, the change that Fizzer suggests is needed, in fact breaks the ELO system. However, that is all about the other two ladders, as the problem in the seasonal ladder is even more funcamental as I explained before.)

Getting back to the problem here, we aren't done yet. The way the ELO system is used is a big problem, but it is not the only one. The other problem is indeed in what most people are complaining about: the pairing system.

If I understand things correctly, it is based on the swiss pairing system. Again, we are dealing with a very solid system.
The problem, then, is in the fact that it is more than simply a pairing system. It is in a way a rating system. The Swiss system pairs players in such a way that their number of wins is a measure of how well they have been playing at that tournament (or ladder, or whatever). I don't use the word skill here, as that's not what it is about here, skill is a quality of a player, while being ill does not change your overall skill level, it may well mean you play more poorly during a certain tournament (which a skill level should see as a small problem you will be able to get back from in time, while in a tournament you may just finish last). I don't know how well that translates to the system that is used here, but I would imagine something like the number of wins saying something about a player.

The real second problem is that these two system are in one another's hairs. ELO says it doesn't matter if you play against much stronger opponents, as it is already taking that in account. It actually even assumes you play players of varying skill levels to make your rating stay correct. Actually the moment at which it needs this the most is when the system is still settling in. In the mean time, the swiss system makes sure that this playing against player of different skill levels doesn't happen and as such, the settling of the system is actually slowed down. Basically, because they try to do the same thing in different ways, they are ending up getting in one another's ways.
Season II matchmaking: 3/31/2012 04:09:59


Chaos 
Level 54
Report
we could have divisions with round robin system where the top x promote to the next division (next season) and the bottom x demote to the lower one. It could take a few seasons before some players reach the top of division 1, but at least they can win on their way up.
Posts 21 - 23 of 23   <<Prev   1  2