Hera’s take :
I have been raising this issue from some time that just using ELO for match making is not very balanced considering the fact that there are several players who have played thousands of games and a beginner is matched against them. They result in very one sided match. I would not mind that if these matches make you loose less elo weighted by number of games. WDUT?
I agree with Hera on the fact that the starting Elo is too high for new players, so much that it is meaningful to look for a solution to make new players start lower.
But I agree with Grubby in https://m.youtube.com/watch?v=EN_iUhaaFV0 (he is a famous Warcraft 3 player who played aoe2 with TheViper) that ladder reset is a bad idea, which Hera proposes in the video.
Namely, I trust that a 1000 Elo player will consistently stomp a 800 Elo player. If there is a Elo reset, this 800 player will still get stomped by this 1000 Elo player along with all > 1000 Elo players. As happy as the 800 player would initially be to be back at 1000, he still isnt on that level and will quickly go back to 800. So the Elo reset will make a mess for everyone.
This may be different for Team games, where a “good player” may need good teamates to be effective.
I disagree with you that the number of played games makes any noticeable impact on an Elo & based ladder, because players improve at very different rate. Someone who is
(1) watching Hera videos,
(2) watching his replays,
(3) trains build orders,
(4) has high level experience from another RTS,
(5) use aoe2database/aoesimulation/mapEditor to check units behavior
will be drastically better after 50 hours in game compared to a “veteran” of 5000 hours who just play for fun with wacky sytrategies and never checked anything online.
For me, ideally the ladder should do the following (I am aware some ideas may be too much effort to implement or controversial):
I have seen the vid from Hera. His solution is kinda the most terrible thing to do. How to fix slaughters? Reset the Elo every year, so not only the new players have to deal with unbalanced game, but every player has to deal with these kind of games. That will only increase the issue.
If you make the starting Elo lower, then we see Elo deflation. After a team the average 1000 Elo player now will be 800 Elo in the future and the same issue arises. So just lowering the starting Elo doesnt work. This is just a flow in the Elo rating system. A new player is considered as average player. That is the bases of an Elo system.
Based on the vid i first thought Hera would make the same mistake as you, but in the end he mentioned this fix. He think it is a decent fix, but i disagree with this.
I do agree with this. Playing time doesnt say much. So adjusting Elo for the number of games you played is bad. So that idea should not be implemented.
I do like the idea of having a few AIs in the ranked queue just to see the Elo of the AI. This can be used for tweaking the AI as well. Not sure on adapting the ranking based on this. Elo ratings can be volatile. As result also players ratings will get volatile. Personally i have huge swings in Elo. Plus or minus 100 Elo happens a lot. If you are continuesly shifting players Elo to compensate, then it will become a mess.
That is not really how Elo should work. It result in a inflated of deflated system. Also not everyone plays regularly against the AI. So this isnt viable for everyone.
Having different ratings for each map type seems a great idea, but can also back fire: If you need to get a stable rating on all ladders, then it will take even more time to get to your Elo. So that 20 losses before you win might become even 30 losses. So this isnt really a fix for this issue.
Not sure how this works against the issue raised here. Also who decides the +50 Elo or -150 Elo? Personally also not a fan of having to pick a civ for every map if i dont want to play them. Also for team games: If i queue up as solo, then i dont want to have the same civ as my ally. But i dont know my allies before getting matched.
My personal solution:
Do not use Elo. It is old and outdated and have some flaws. You are much better off with a system like TrueSkill. It learns your skill level much faster and is much more adaptable for things like having different ratings on different map types. You can even have a link between 1v1 and team games to improve the ratings for team games. Team game ratings are still a mess because people can easily being carried by an ally and then have to drop down their rating.
TrueSkill 2 can even use some ingame statistics, but i dont really know if we need to use them. Many statistics can be misleading. Score for example doesnt tell the full story always.
To be honest, Elo 1v1 is one of the best thing the game has.
It’s difficult for new players, but is great for all the players after 10 ranked games.
If you ask me, the game has to build confidence to the players, kinda a tutorial, for players that want to try ranked for the first time.
The ladder does not need to be a strict Elo Formula.
Nothing (besides implementation effort and the willingness of devs & the community) prevents the delta rating after a game to be tweaked by a few point to ensure a stable Elo. It is not foreseen by the standard Elo system (who aims at keeping the average rating to a constant value), but it is not forbidden to change.
Yes, it is a very complex problem to solve as not everyone has the same profile:
It is very difficult to create a rating the correct profile fir each player, even more so if the rating should stabilize quickly and if maps are rotating.
It is my ideal system, not one I feel is easy to attain.
It would be something based in the civs win rates. This would evolve every patch and after each map.
Again, not easy to implement and to diffetenciate how much of the win is due to the players and how much due to the civ. It can be done with some machine learning models for instance, but I am not sure what would be the best way to do it.
This wouldnt work for the match making in Team games indeed. But at least it can impact the rating updates after the game (the team with good civs having a higher posteriori rating)
Maybe one could use the civ from the previous game on the map, might be an option.
This sounds like a great system. I will have to read about it.
Well the current system is not a desaster either.
But there is no harm in looking for a better system, because it is defenitely possible and would be meaningful in many ways. Anyways, whether the devs want to change or not is not up to us, as long as they are not explicitely asking.
The ranked lobby should have a minidraft, very simply. If you don’t want it, there is the Quick Play feature, which after being implemented has been abandoned and even put in a hidden corner.
There is nothing wrong with the elo and derivative system, like Glicko, it is fine in 1v1 matches.
Is a system designed to assess personal performance in a team game.
If we want to fix the system once and for all, we must first take out all temporary smurf accounts, as they continue to generate ELO points out of thin air.
In the first 10 games, a new player has a much higher K modifier than his opponent. Which means that if he wins he gets more ELO than his opponent loses. This also happens in reverse, but in practice much less; in fact the average ELO is currently not 1000 but something like 1040 (last time I checked).
Yes, but you can compensate for that. If the average ELO is 1000 and the new player is 800 it means that somehow you have to reintroduce 200 points into the system. One idea could be that whoever beats him gets 50% more points until the bonus is exhausted.
I just want to throw in, the starting rating is actually 800, not 1000. In your first games, it’s more likely that you’re up against 800’s (or even lower).
The problem is that losing is always going to be at least somewhat negative. I would imagine that winning your first game probably makes a massive difference in whether you stick around, and this problem only balloons as time passes and less and less new players win.
Many players just like playing against the AI. If you could design some sort of system that could accurately judge player skill and automatically assign them an AI that will challenge them, but still be weak enough to let them win, you could really encourage people to get better and better.
Then you can mix in coop pve play. Multiple players against AI, that will get people in the door to multiplayer and show them how fun it can be.
Then you can use these games and experiences to accurately assess their actual skill and assign them an elo that starts them off at the right place preemptively.