[Suggestion] Elo decay and Elo from AI

When you look at the beta leaderboards, you notice that certain top ranked players have not played for weeks, maybe even months, yet are still present in the top 10.

players should have to play a few matches regularly to be allowed to keep their high rating.

i think if a player has not played for over a week, that player should lose a couple of elo points each day until he plays another non AI-opponent match or until he drops to a certain elo level (below top100).

Additionally i also think that AI opponents should not affect your elo rating once you reach a certain point (roughly 1400 elo).
Players can abuse this by only joining the matchmaking queue when there are no other players of similar high rating in lobby. AI only gives 2-3 elo at that level, but if you are careful and only match against AI, your rating can only get higher.

 

1. Why is this needed at all?
2.-II-
3. Playing against the ai isn't 100% win chance and if you lose you crash hard. Also the chance to have any time not to have people to play against goes down as more people join.
4. 2-3 rating is as good as nothing besides losing to them would then take a few hundred points out of your team meaning that unless you have a 100% win rate against them its moot.

 

why do i think decay is needed?
I just think that inactive players should not stay at rank 1. if you cannot play against them, you cannot take their rating from them.
Champions need to defend their title. They might have had a very long win streak in the past and then decided to stop playing because they did not want to lose rating.
A ranking system should not encourage staying inactive.

against the AI you might not be able to guarantee a 100% winrate, but you can certainly get close to it.
It's just my personal opinion, but i don't really think AI belongs in a PVP rating system.
this probably won't be a big deal after release anyway, when more players start playing again. Currently, when your elo is very high, you can almost guarantee an AI match.
if you can guarantee AI matches and are confident that your winrate is close to 100%, then you can certainly abuse it to get an even higher rating and keep staying there, as there is no decay.

 

I agree with both of these points. ELO decay encourages players to stay active, and creates a more accurate leader board. This is a throwback game, but it isn't an arcade style leader board that shows how many points you can accumulate in a single play-through. With ELO decay, you can check to see who the top players are at the time you check, not who was able to sprint 100-200 points ahead and sit there.

AI is pretty much a 100% win rate above 1300 ELO or so (I am not a dev, I don't have the numbers, I am guessing). Personally I don't think that rating should be gained off of an AI opponent since it artificially increases a competent players rating.

A rating should be proportional to your competition, and if your competition isn't playing any more or is a computer that always loses then it isn't much competition at all.

 

Agree with both of those. Make ELO decay exponential and on an increasing frequency.
Don't play for a week, lose a point.
6 days later lose two points.
5 days later lose 4 points.
4 days later lose 8 points.
3 days later lose 16 points.
2 days later 32 points.
Next day lose 64.
Next day lose 128.
Next day lose 256.
etc.
No matter where you started, approx 1 month of inactivity would drop you to 0. That said, i'd recommend having the loss stop at half their max rank or at 1000, whichever is higher so that if they decide to jump back in they don't have to squash new players for a few hours to do it.

But...keep their lifetime highest rank on record and on display so that people who play them can tell that they're not just facing a newbie.

And yes, stop the gain from AI after 1500 or so. Make them earn it.

 

thats unreasonably harsh. I can think of lots of legit reasons for a 1 month hiatus. Forcing them to spend hours to grind back to their previous ratings so that they can actually fight people on par with them is detrimental to everyone involved. It wastes the player's time (it takes hours to get back to the higher ratings), allows opportunities for lower ranked players to get curb stomped by far superior opponents who have the same elo as them due to decay, makes it more difficult for higher rated players to matchmake and even benefits the elo decay-ed player by giving them a bunch of easy wins (more items/chests).

Elo as it is currently implemented in the game is not a zero sum game. You don't need to fight against #1 to raise you're elo above him. ELO should decay for other reasons, but nowhere near as fast as your proposing. the right way to do it is to have a increasing elo deviation with time. Elo accuracy in reflecting skill decrease with time, if you log in after a long hiatus, you win/loss should create a bigger change in rating for the initial few matches.

On a side note, why is CH's elo system not a zero sum game with 1200 or 1500 representing average and having people start there?

 

I don't think it needs to be that complicated and i think my rate of decay is fine. Sure, there are plenty of reasons for taking some time off from the game, but you don't even see a significant amount of decay until 21 days have passed. That makes it pretty easy for a casual player to maintain rank. Occasional stompings will happen, but i don't think ELO decay is going to be responsible for a significant number of them. Intentional deranking and just plain bad luck are a worse threat.

At any rate, dropping rank is really not that big a deal - it is just a game after all, and keeping the leader boards fresh so that newer players feel like they actually have a chance to move up is more important than keeping lazy players happy.

As to why it's not zero-sum, i'd say that because all-in-all it's a pretty casual game, and it's pretty discouraging for a new player to go below their starting score. A constant sense of progress keeps players coming back. The traditional ELO encourages high-ranked players to not take risks - a model that encourages increased participation is better for a game like this.

 

It makes zero sense to force a player that went on a hiatus to spend 10+ (4 game an hour aprox) hrs grinding joyless games against opponents too weak for them to get back to their original rank. People are more likely then not to just never come back at all.

 

If somebody does go on hiatus, it makes sense for them to actually have to... play games to stay in the top spot. It's pretty ridiculous when I see that the top players in a game are inactive, having gotten super high scores back when the meta allowed for almost unstoppable builds. This makes the top spot almost unobtainable for the players who are actually active, which is unfair. I think that if somebody is inactive for a week, they should start to slowly give points away to the AIs, which would then be spread out among the active community.

 

Or you could just have everyone drop a few points per day, regardless of activity. That way, you can't avoid dropping by playing a single game each month or so.

 

I'm not interested in allowing high-ranked players to keep stroking their e-peen without winning one game in three weeks. If they want to take time off then they should actually have to play when they come back to play. If they don't want to play then good riddance.

 

Is the point of the ranking system to give glory to the top players? Or is just to estimate a player's ability so that they can be matched against other players with similar ability? If it's the first, then it's pretty clear that player's should be penalized for not playing - but if it's the second, then it probably doesn't make a lot of difference.

In any case though, even if the sole purpose is to match people of similar ability, it would probably be fair to assume that the ability of a play will atrophy a little while they are inactive. (And even if they are still a sharp player, they might not be as in-touch with the metagame.) So perhaps we could satisfy both goals with a rule something like this:

For every week a player with above-average rate is inactive, reduce their rating by 10% of the difference between their rating and the average rating.

For example, if the average rating is 1000 and an inactive player has a rating of 1600, that player would lose around 60 points per week. (The points penalty would drop as their rating dropped. After 4 weeks of inactivity, their rating would be 1394. -- That's for 10% each week. Perhaps even just 5% would be good.)

As for AI players, I think it's great that they have their own rating which can go up and down. And I think it's fine that human players can increase their rating by playing AI. However, there is one rule that doesn't apply to AI player's in match-making, and that rule may cause some ratings distortion. The rule is that players are not normally matched against the same opponent in a short amount of time. If two players are allowed to face each other repeatedly many times in a row without facing anyone else - this can obviously have an distorting effect on the rating system - and since the rule doesn't apply to AI players, matching involving AI players might be distorting the ratings. (Many people have commented that Mom's rating is lower than it should 'really' be because she has to face top players over and over again. I think this is true.)

To address this problem, my suggestion is this:

Rated matches involving AI players should only contribute to ratings if the match pairing doesn't violate the usual rules which prevent human players from repeatedly being paired with one another; or if the result of the match is not the same as last time.

With that suggestion, winning a heap of matches in a row against Mom would only boost the player's rating by 1 match worth (and only reduce Mom's rating by 1 match worth); but if the player then lost a match against Mom, that loss would affect the rating of both players. Also, if the player had a few matches against other human players, then their next win against Mom would once again count toward their rating.

 

Of course you do or you wouldn't have said it. Your solution causes other (unanticipated?) problems however. You've effectively introduced into the game something that people are trying to remove from the game -- low ranked players playing against high ranked players who have dropped their ranking to get wins. Who is being punished here?

If you are #ranked 1 for a bit, you go away for a month, come back and will be able to quickly get the epic chest. At this point you will likely realize that you are actually better off playing in short spurts than being continually active at which point ranking becomes meaningless because it's more a measure of how active you are rather than skill levels.


My suggestion would be to have two actual ranking mechanisms which contribute to the overall ranking. The first mechanism would be long term, historical, skill based focus based on the ranking of the opponents you win or lose to. The second contributing ranking mechanism would be the short term version of the first. The second one would fluctuate based on your games only within the given time period, whatever that time period might be, let's say a 1-3 month period. The combination of those two mechanisms would result in your current ranking.

The first of the mechanisms would illustrate your general skill level while the second would allow you to always be fighting for the top scores at any given point. Combined, any inactivity would reduce your overall ranking, but it could only fall as far as your overall skill ranking.

 

I think spectromancer solved this problem nicely. It is an elo system, but if you lose you always lose one point less than under the normal elo system. This leads to a bit of point inflation over time, so inactive players get slowly pushed off the rankings without penalizing anyone. Also you get a (fake) sense of progress, because even if you get stuck at 50% win/loss your elo number will slowly increase. The downside is that it is not that 100% accurate since a better player can land below one who has simply played more matches in the rankings, but I doubt anyone cares all that much.

 

This is just number inflation though. I don't see how this would work long term. Our rankings will get to be in the millions. How is a new player intended to view and digest such huge differences between their ranking and older players?

 

No need to exaggerate. The effect would be modest. If you play the ridicolous amount of 2000 games and lose 1000 of them you only get 1000 points extra. The point is to pump points into the system so you can eventually overtake inactive players.

 

1) To get a million point advantage, you will need to win a million games.

2) they don't. The whole point of this system is that older players who have been away too long should be ignored. The huge difference (whichiwll only happen after a long period of hiatus) accomplishes by making it incomparable (which reflects reality in that it is incomparable by that point). With point gains of up to 25, it wont take much to climb back up when you're active again.

I'm not saying its good (I'm kinda pensive about something arbitary like adding a +1 to the equation for elo movements), but your concerns are not the issue.

 

I'd prefer that the rating system focus on estimating the relative ability of players, rather than focusing on how active they are. Hypothetically, if two players have equal ability, but one plays 6 times per day and the other only plays 3 times per day - then I think they should still have equal rating. I don't like the idea of letting scores creep upwards just by playing more games.

 

I think Kardoic said it best. The elo system is not made for epeen its so you get matched fairly. The only thing about going inactive is that you might play people who are actually better than you when you come back. I know this for a fact as it happened to me with lance. Don't go ruining a rank system because you see it as epeen and have epeen envy its simply not a good way to look at it.

That said i see no reason not to have something simple pointing out if people are inactive like a tiny badge or a note next to them in any scoreboards if they haven't played for say a month or two.

 

Another (sort of) related ELO question...does the rating change take into account only win/loss, or also include the number of victory points gained per player? What I mean is, if I lose with 5 VP or lose with 0 VP against a given opponent, will I lose the same amount on my rating?