Rating sigma should increase over time

A lot of people play less FAF because they reach a high shown rating that isn't necessarily representative of their current skill:

  • Some people manage to get overrated and then don't play many/any rated games afterwards because they don't want to lose that high rating
  • Similarly, some ladder players reach a peak ladder rating and just hold it indefinitely without having as much incentive to keep playing to be considered a top ladder player by rating
  • Many other players reach a proper rating, but then stop playing FAF for months/years for other reasons, and when they come back to FAF, they are overrated and forced to lose many unbalanced games to get to a rating that is reflective of their current skill level (many of them end up re-quitting rather than going through this process, as having to lose like 30 games in a row can be quite unpleasant).

Increasing rating sigma slowly over time should help each of these problems and should encourage people to play FAF more.

So, to explain the rating system; shown rating can basically be thought of as "base rating (mu) - uncertainty (3 * sigma) = shown rating". The higher the sigma value, the lower the shown rating and the more uncertain a rating is. So, new players start at "1500 base rating - 1500 uncertainty = 0 shown rating". A typical player with a lot of games has around 250-300 rating uncertainty. Playing rated games lowers uncertainty to around 100-350 uncertainty as a stable level whose exact value largely depends on whether the player plays a lot of 1v1's (less uncertainty), 3v3's, 8v8's (more uncertainty), etc.

If we increased uncertainty slowly over time (such as by 1 point per day), then the shown rating of inactive players (and those who avoid playing rated games) would slowly decrease over time. The base rating would stay the same, but it would add incentive for players to keep playing rated games rather than reaching some peak rating and then barely playing afterwards. It would also help with properly rating players who stop playing FAF for a while, as their rating would slowly become gray/uncertain again if they don't play for a very long time. That would make it much easier/faster for returning players to get an accurate rating upon returning, and if they haven't played for years, their rating would even be gray, which would make the transition easier while being a fair representation of "rust".

pfp credit to gieb

Stopped playing for 2 years, fell like 150 points in rating, 1 point of deviation a day is too much and will lead to rating manipulation. No one is 100 shown rating worse because they didn’t play for a month. You don’t lose so much capacity as a player that a 3 month vacation causes you to lose 90% of your games against your old self.

Who are these dudes that reach some peak ladder rating and specifically stop because of that btw

Also don’t mess with a player distribution system to tinker with player activity, terrible misuse of trueskill. Games make leagues on top of their ELO system to incentivize playing the game.

how to turn the person that plays 1 ladder game a week into a person that plays 0 ladder games a week:

Funny enough with a lot of my experience watching some of the old GPG players that came back to FAF after 5, 6, or 7 years it tends to be that they perform incredibly badly for their first half dozen or dozen games. After that, they quickly recover a lot of the idiosyncrasies of the game (since you never really forget the basic rts principles that carry across all games) and so have a much easier time getting back to their original high rating with minimal rating plateaus. All this change would do is make it actually take longer for that situation to correctly resolve as those initial "bad games" are given an undue influence.

It also doesn't actually make their first half dozen or dozen games any more "contested" because trueskill still utilizes your mu to actually select opponents and so they are still going to be playing the same sort of stomps that they are playing in the current system.

If people are playing for rating instead of fun then they should be stopping playing. There's no reason to try and bring these players back through rating manipulation and other psychological tricks.

This whole game is a psychological trick to get you to release dopamine and serotonin, you should instead be campaigning to make opioids over the counter.

ftxcommando said in Rating sigma should increase over time:

Stopped playing for 2 years, fell like 150 points in rating, 1 point of deviation a day is too much and will lead to rating manipulation. No one is 100 shown rating worse because they didn’t play for a month

It doesn't have to be 1 point per day; people are welcome to suggest how much the decay should be. Also, 1 point per day would be 30 points in a month; not 100... Regardless, since it would only be the uncertainty that is changing over time, the player isn't losing any base rating and would still be matched fairly.

If players want to manipulate their rating, they can do that now in the span of a few hours. To get the equivalent rating manipulation by taking advantage of decay, they'd save about an hour of their time per year that they waited. Don't complain about a couple of drops in the ocean when ice caps are melting.

ftxcommando said in Rating sigma should increase over time:

Who are these dudes that reach some peak ladder rating and specifically stop because of that btw

You're welcome to look over the data yourself. Also, here's a relevant thread by arch: https://forum.faforever.com/topic/651/ladder-needs-some-form-of-decay

ftxcommando said in Rating sigma should increase over time:

Also don’t mess with a player distribution system to tinker with player activity, terrible misuse of trueskill. Games make leagues on top of their ELO system to incentivize playing the game.

Increasing the uncertainty of the skill level for players who haven't played in a while is a very reasonable thing to do that isn't misusing the system. If anything, it makes the system more accurate. If someone hasn't played the game in 2 years, their rating should be more "uncertain".

Here's a relevant thread by a returning player: https://forum.faforever.com/topic/207/rating-adjustments-for-returning-players

CaptainKlutz said in Rating sigma should increase over time:

how to turn the person that plays 1 ladder game a week into a person that plays 0 ladder games a week

Perhaps you misunderstand. You are welcome to pm me about this. This shouldn't be a negative for people who play 1 ladder game a week.

pfp credit to gieb

@thomashiatt said in Rating sigma should increase over time:

If people are playing for rating instead of fun then they should be stopping playing.

This is precisely for people who want to have fun. The current system preserves your rating, this new system suggested by @Emperor_Penguin will increase the chance a returning player has fun.

Having lost 50 rating means you have to lose like 4 fewer games, possibly none, before you play at your level again?

30 deviation = 90 shown rating decrease. This is incredibly inaccurate and probably destroys the system’s accuracy entirely especially for top players that may play a single game a month if that. Rating decay (which is still a bad idea) but if it were implemented, would need to be something like .1 sigma a day with a cap at +350 sigma total or so. Would take 7 years to reach that cap but whatever. Possibly coming into play once 3 months of inactivity is hit.

“ If players want to manipulate their rating, they can do that now in the span of a few hours” ok and you can go commit murder if you wanted to does that mean it doesn’t matter if I mandate every person in the US open carries? Rating manipulation is always bad. Don’t make more happen.

Honestly no idea how linking to a thread where I explained the exact same solution I explained here does anything. Are you a robot that does the same bad idea again and again in the hopes that this time I don’t post?

“ Increasing the uncertainty of the skill level for players who haven't played in a while is a very reasonable thing to do that isn't misusing the system. If anything, it makes the system more accurate”

Prove it. The scenario you linked is exactly the worse case scenario I mentioned which is the only person reasonably benefiting from a decay system. And even then he came back after 7 years and lost around 200 rating. That’s practically an order of magnitude leas extreme than your proposed decay rate.

This will become irrelevant with the division system as the divisions reset after a certain amount of time.

askaholic said in Rating sigma should increase over time:

This will become irrelevant with the division system as the divisions reset after a certain amount of time.

That's a fair point regarding ladder. Rating sigma slowly increasing over time would still be relevant for global rating though. So, I believe it should still be considered.

pfp credit to gieb

If it doesnt happen already (I dont really understand how the rating system works) what about increasing the amount that your rating changes based on consecutive wins/losses? I.e. your 3rd loss in a row loses you more than the 2nd loss, and so on (and if it already does this then possibly increasing the extent to which it does it slightly).
That way you dont have to sit through as many unbalanced matches where your team gets mad at you because you're far worse than your rank suggests, hopefully improving the experience for everyone.

I'm assuming from the screenshots in the linked thread that the league system is just intended for matchmakers not custom/global rank, in which case while it would help to some extent it wouldn't solve the issue for global (or are there plans for a league system for global as well?)

@ftxcommando said in Rating sigma should increase over time:

This whole game is a psychological trick to get you to release dopamine and serotonin, you should instead be campaigning to make opioids over the counter.

Your take amuses me on a scientific level.

~
People seem to be focused on the arbitrary 1/day rate, and should not be. We could dial in some equation later.
~
How would sigma go back down? If I recall that is handled by normal trueskill, but could you highlight the process and if any changes would be needed for this alteration?

You must deceive the enemy, sometimes your allies, but you must always deceive yourself!

@kalethequick said in Rating sigma should increase over time:

How would sigma go back down? If I recall that is handled by normal trueskill, but could you highlight the process and if any changes would be needed for this alteration?

Sigma going back down is handled by trueskill and shouldn't need any changes.

Playing rated games lowers sigma to a normal range relatively quickly (every rated game has an impact); 1v1's lower sigma very quickly, while larger games do it at a more moderate pace.

If you mouse over a player's rating in the lobby, it will show something like 1421 +/- 210. The 1421 is the base rating. The 210 is the uncertainty. The shown rating is 1421 - 210 = 1211, which is rounded to 1200.

If a sigma increase over time is implemented, and if that player didn't play for a long time and their uncertainty increases to 500, then their rating would be 1421 +/- 500. At that point, if they started playing again, their shown rating would be 900, and their base rating would change more quickly for their next few games (to match their current skill level), and the uncertainty value would quickly lower itself back into the normal range.

pfp credit to gieb

I keep hearing about people who used to play FAF a lot, but stopped for months to years, and would want to try playing again, but don't because they're now overrated at their old rating and don't want to have to lose so many games just to be able to play balanced games again. I think my above points are still quite relevant, and it would make sense for FAF to implement some sort of sigma increase over time.

For example, FAF could do an increase of 0.4 sigma per day for users with less than 150 sigma. That would equate to an increase of ~36 rating uncertainty per month with a max of 450 rating uncertainty. For perspective, most players with high game counts have between 150 and 300 rating uncertainty, while new players start with 1500 rating uncertainty. So, that would cap a regular player at losing a maximum of about 216 shown rating after roughly 6 months of inactivity, with no additional rating decay after that. Note that that would just change the uncertainty and not the base rating, so it would basically make the player a little more 'gray', and their base rating would consequently get adjusted a bit more quickly to match their current skill level, while their shown rating would temporarily be a little lower in the interim (since their rating is a bit more uncertain).

This will become irrelevant with the division system as the divisions reset after a certain amount of time.

Does division resetting increase sigma or have some material impact behind the scenes? Like, I understand that it then won't display someone's division until they play X more games or smth, but given that divisions largely seem like a cosmetic change that displays division instead of rating, that division reset sounds more like a cosmetic change that doesn't remove sigma relevance.

Regardless, this still has relevance for global rating either way.

pfp credit to gieb

You could implement that but unless you start running timed scripts over the database, its going to get a bit messy.

The easiest way to implement this would be to apply the rating decay right before a player's game is rated. So before the Trueskill calculation of a finished game runs, it checks for all players involved in the game when their last game was played and if their rating has decayed since the last game.

The problem is that this means that the rating decay exists but is not actually visible in the client until people actually restart playing, which they won't if they think their rating is too high.

So in addition to the server change, we would need to effectively let the client recalculate the rating decay of the logged in player at any point in time. Unless every client does this for ALL players ALL the time, the players own decayed rating will then only show up in that players client, while other players still see the non-decayed rating, adding to the general rating confusion.

I'm not sure the complexity is worth it.

@katharsas Just having it apply right before the game is rated would still help - e.g. while some people might just not try playing again due to being overrated, others might try a game, and then see how little their rank dropped and decide it's not worth the effort of playing more games to get to a more appropriate rank. So such a change would help with the latter case.

It's not like we'd need to add decay in every single day. For example, perhaps we could just do it upon player login, or just run a script over the database once a month, or do some more creative idea : )

pfp credit to gieb