The new system just flips whose elo gets inflated. Now it will be the lower ranked players who will get a boost to their elo for no good reason. So you can easily drop your elo by queueing up for some games with weaker players.
The current system is way better than that.
1 Like
Not if you disable family accounts from playing ranked games. If you want to play with friends that are way beyond your level, you can just play lobby games or QMs.
That’s a suggestion I’ve made many times before.
This though, is a terrible idea. Unranked has no semblance of actual ranking, and is totally infested with smurfs. People shouldn’t be forced to just play lobbies with their friends if they want to play together.
1 Like
Can please someone link this crucial information to the devs about Elo:
Elo is a logarithmic projection of the strength distribution. That means, that if you want to add elos, you have to use the correct addition method for logarithmic functions.
How this works is described here .
And how you can practically implement is described here.
The basic concept was proposed by @Mercy9545 years ago here
Every attempt to add up elos in different ways like just directly adding them up is doomed. It won’t work, cause it ignores the basic mathematical properties of the elo system.
7 Likes
14 months… I doubt if devs ever spent 30 minutes investigating this issue. It’s obviously not zero-sum anymore!
The example in their update news shows no inflation in the match 2000+1000 vs. 1500+1500. How about 2000+1000 wins 1600+1200? avg elo are 1500 and 1400:
2000elo has 96.93% win rate to 1400elo → +3.07%
1000elo has 9.99% win rate to 1400elo → +90.91%
1600elo has 64.01% win rate to 1500elo → -64.01%
1200elo has 15.1% win rate to 1500elo → -15.1%
Sum of elo delta is 14.07% which means the elo sum of 4 players increase 14.07% * 32(or 16. Depends on the multiplier)
Of course you can argue that in the case of 2000+1000 loses to 1600+1200, the entire sum loses 14.07%, however, we all know that some pro/semi-pro players simply get paid and carry a few low-elo players. They do it EVERY SINGLE DAY!!!
There are so many simple ways to solve this zero-sum problem. For example, calculate individually and enumerate all 4players*4players cases.
Anyway, I still give a thumb-up to the change today because the previous one really sucks.
@ConvertThis4915, considering you were talking about the TG elo thing earlier, I’m now going to drag you into this. Can you please pass one of these ideas along to the people who can implement it? The devs don’t even have to say it came from the community as long as it actually happens. People have worked out what you need, the dev team just has to take some of the ideas.
Very nice change to TG Elo system!
Of course it isnt perfect and could have some adjustments. But it is way way better than the old system, that was just a total clusterf…
-
If Team 1 wins:
- Player A on Team 1 will gain 2 Elo (as if a 2000 Elo player beat a 1500 Elo opponent)
- Player B on Team 1 will gain 30 Elo (as if a 1000 Elo player beat a 1500 Elo opponent)
- Player C on Team 2 will lose 16 Elo (as if a 1500 Elo player beat a 1500 Elo opponent)
- Player D on Team 2 will lose 16 Elo (since their rating is the same their teammate, they lose the same number of points in this scenario)
btw, what if team 1 loses? does the high elo player then lose more or also only 2 elo?
This is weird, as I understand if team 1 loses, player a loses 30 elo and player b loses 2.
What in the aftermath means with a 50 % winrate player a loses 14 elo per game on average and player b wins 14 on average… that is NONSENSE
Besides because of the wrong team elo calculation (again!!!) the winrate will actually not be 50 %. I will calculate the expected winrate of team 1 vs team 2 now.
2 Likes
If player A elo loses to lower TG avg and player b elo loses to higher TG avg, ofcourse player A should lose more than player B, exactly the same with winning.
| Player A | Player B | Player C | Player D | |||
|---|---|---|---|---|---|---|
| Elo | 2000 | 1000 | 1500 | 1500 | ||
| Real Elo | 2000 | 1000 | 1500 | 1500 | ||
| Shown Strength | 8.77118932 | 2.961619375 | 5.096756246 | 5.096756246 | ||
| Real Strength | 8.77118932 | 2.961619375 | 5.096756246 | 5.096756246 | ||
| Current Team Elo | 1500 | 1500 | ||||
| New Team Elo (as it should be) | 2267.944155 | 2138.412146 | ||||
| Winrate vs 1500 elo Team | 0.678229285 | 0.5 | ||||
| Net average elo gain | -8.296662885 | 19.70333711 |
Team 1 wins with 67.8 % probability vs team 2 but player a loses 8.3 elo on average and player b wins 19.7 elo on average.
And that’s on games where the devs think “these teams are equally matched”.
fyi: the calc assumes that player a loses 30 elo when losing vs team 2 and player b loses 2 elo. That’s how I interpreted the dev’s post. But devs are cryptic.
2 Likes
Makes totally sense. If 2k player plays only against lower ELO TG avg, he should lose ELO with 50% winrate. As long as player A keeps playing avainst teams where his personal ELO is higher than TG avg, player A needs way more than 50% winrate to climb.
Makes perfect sense.
But team 1 doesn’t has a winrate of 50 %, the winrate is way higher.
And if teams are evenly matched so a 50 % winrate is the result all players should exactly keep their elo - as it would be proven accurate.
It doesn’t makes any sense to me if in a team some players lose elo over time and others win… what is this nonsense? If a team wins more together all players have the right to gain elo and if a team loses more together all players have to lose elo aswell.
1 Like
Ofc players in the team should gain/lose ELO differently over a period of time. Why should they gain the same amount. Higher player looses more or gains less than low elo. You lose games to lower players, you lose more. You win against higher players, you win more. Makes perfect sense imo.
LOL.
I think you have no idea what this means. It means that every player with higher than average elo will be vastly underrated as he has to win way more games to keep his elo up.
And the low elos will have vastly overrated elos but lose most of their games.
That’s the result of that nonsense system. Team elo will be completely broken. If you are kinda bad player you will lose basically every game, making it absolutely horrendous to even join team games, cause nobody wants to play with you and blame the loss on you (despite it is to blame the devs they made this system that pushes low elo players team elo towards the average).
And after the low-elos are all completely demoralized the mid elos will be blamed until nobody wants to play ranked TG anymore.
4 Likes
I totally agree with casusincorrabil, I think maybe we need to put the formula here for developer to just directly implement it to the system, which are
S = e ^ (elo / 400 / ln(10))
And then in reverse:
Team Elo = ln ( SUM(S)) * 400 * ln(10)
1 Like
I understand the feeling of casusincorrabil that most of us play this game from 20 years ago and many people in the community are now PHD or master in math or statistics but the elo system in the game still using the wrong calculation method really upset us
1 Like
The bad thing will be for 2k+ players who premade team up with their friends who are sub 1k. Solo quers wont feel a thing, as long as solo que do not make teams with too big ELO differences. Also premades with similar ELOs wont feel a thing either.
Basicly it will only affect badly those top players who like to play with they newbie friends where top player solos 3 enemies while their friend are building their second TC…
Say it correctly please These are smurfs. It’s basically impossible to get 500 tg elo if you don’t intentionally throw.
2 Likes
Did not mean those under 1k “small dudes”. But new system should also inhibit setting up those 300 ELO accounts for the purpose to artifically deflate avg TG elo. Its still possible, but would take way longer time to set up those accounts, also the longetivity is way lower for those accounts.
Way more work for “small dudes” for same outcome with new system.
For the price of completely destroying TG Ranked cause nobody wants to play with lower elos anymore.
Some “fixes” just don’t work.
1 Like