Elo Decay

TheEloIsALie ? ;)

The only thing that would be reasonable (on top of the current system where inactive players aren't shown on the ladder rankings) is to increase their k-factor for the first few games (degree of uncertainty, determines how quickly their rank changes), since if they lose games due to being rusty they will drop to their correct (rusty) elo rating quickly. My understanding is that Zero-K's elo system uses a constant k-factor for everyone, so that might be problematic.

BlueTemplar: Yup, this topic is exactly the reason for the name. If I wasn't alluding to certain pastry-related scrawlings, it would've been "My" instead of "The" I guess.

My idea for this was to introduce elo inflation. Losing takes normal elo, but winning gives +1 or +5% to the normal amount. This way active players gain elo quicker, and people who are inactive "fall behind" and slide down the ladder, but can quickly regain their position due to how elo works.

Im not a math person, so I don't know how retarded this is, but it does seem a simpler method of thing things than coming up with arcane algorithms to calculate how much skill decays over time and how to subtract elo with it.

Elo decay/inflation get really messy real quick.

Decay treats everyone as if they had the same resilience in acquired skills which is blatantly untrue. For me it takes around three games to be able to use the interface without feeling like I'm bending over backwards. I'm disoriented and uncomfortable. Then after that it'll take about another 20 games to get to where I was. Then I often experience the benefits of a 'perceptual reset' where I'm able to take all the things I knew before and improve on them by freeing my mind from the same old stale neural paths, becoming a better player than I ever was before. This process is not the same for everyone, many players seem to return and be exactly where they were withing five games. Some seem in a perpetual state of rust where they're never able to play quite enough to reach their old peak. There is no formula that predicts this unless the formula adapts based on patterns found at an individual level - which afaik is a helluva lot more difficult to make than the benefits of such a model is worth.

Elo inflation disallows comparisons between players across time. Even comparing the same player's position between one week and the next becomes a complex affair. StarcraftII did this when I was playing, and it was insanely frustrating, and resulted in inaccessible means of measurement. There were divisions, but you could be easily #1 in your division while being lower than #50 in a neighboring one within the same league. There were leagues (bronze, silver etc.), but in order to get to the next league once your MMR was settled you needed to perform better than 70% of the players in that division for an undefined period of times. Every adjustment to the system complicates it, removing it from the grasp of the common player, while also often creating further unforeseen problems. Eventually you get to where SCII was, where even an extensive googling won't let a player know wtf is going on.

Perhaps we should further split the scoreboards while periodically resetting ladder? We could have a hall of fame for lifetime scores, while also having the biannual(?) laddering. It's a tradeoff between the issues created by equalising elo (effectively turning every player on the ladder into a smurf), and the benefit of having continued activity incentivised rather than instilling the fear of losing elo on return. It wouldn't have to reset teams elo either.

In fact, you could go even further and separate elo as above while retaining old elo ladder separately. All-time elo, continuous elo, seasonal elo, teams elo. Everyone's happy (except the guy that has to code it). :)

A lot of reasons not to impliment elo decay have been given. What do you think about a frequently reset elo ladder?

Kshatriya, why throw away the current - usually very accurate - skill estimation (also known as "elo") of a player? If you want a different ladder just create a super complex custom ladder system that is independent of elo.

Logikfreak, it already kind of works like that. The problem is that apparently the requirements are set too low. See 1v1 ladder: I'm there at rank 21, yet I haven't played 1v1 competitive for years. PRO_rANDY's last 1v1 was 7 months ago. Hence my two suggestion:

quote: Hiding people who don't 1v1 from the 1v1 ladder would be a good start (and vice versa for teams). This could also be expanded to "hide players who haven't played at least x games in the last y days" (x could be 5, y 30) to make activity a bit more important for the ladder. Or just create a super complex custom ladder system that is independent of elo

Sprung, no need for that. 30 days is enough, no need to lower it. The real problem is that this is not checked properly. See my previous post.

There's always no-elo. Anyone can choose to go no-elo for a few games until they feel they're in the same ballpark as they were.

There should be some kind of decay, albeit very small. Not only are you getting worse over time, but the game is also changing. I would expect it to drop from 1800 to at least 1600 if I don't play it in a year. It should only start decaying after about two weeks of inactivity, or the decay should be exponential over time.