Loading...
  OR  Zero-K Name:    Password:   

1v1 games should not alter team balance

14 posts, 501 views
Post comment
Filter:    Player:  
sort

33 days ago
Here's an example of how 1v1s are abused to lower ones rank to get better teams in casual team games: https://zero-k.info/Battles/Detail/1401061

I'm sure it's not the only one, it's only the most blatant one I could find. There's no reason to have 1v1 games change the balance of team games, as they play completely differently. There's no balance involved in casual 1v1 games, no matchmaking of any sort.

Team games should be balanced based on performance in team games.
+7 / -0
The current ladder separating being Competitive/Casual doesn't really fit its descriptors in practice, it would be more accurate to describe the separation as MM 1v1/everything else.

1v1 players tend to care about their Competitive rating. Team game/FFA players care about their Casual rating. This is because its the one tied to the main variant they play.

I actively avoid playing Sortie/Battle as then I would have my tough-to-get 1v1 MM points at stake.

In current day if people want to play a real "casual game" it means they turn noelo=1 setting on

I propose separation in ranks as such:
1v1
small teams (2v2-4v4)
large teams (5v5+)
FFA

with either MM or player hosted/autohost games all counting to their type, and if people want to play an unranked game they can do what they do now and noelo=1

edit: Also fixing games to alter ranks should not be something going on.
+3 / -0
This keeps coming up, but people still don't do the work to verify whether separating out things improves balance or worsens it to the point bad games would be common.

The giant CSVs of doom (with the data on battles) are available if people want them (e.g. can just ask me or Anarchid or someone). Personally I just don't know how to implement WHR in python or I'd have already run the experiment.

EDIT: Ok, the linked game is blatant elo dumping.
+0 / -0
33 days ago
GBrankdyth68

When doing my FFA ladder shenanigans I found this:

https://github.com/pfmonville/whole_history_rating

However, I think DeinFreund had a specific way to integrate team/free for all games. Could probably just ask him how he handled those situations.
+1 / -0
I'm not going to claim some higher understanding of statistics, but how can you claim any sort of improvement/regression based on data that did not implement the discussed change? All previous team game compositions were determined from ranking that was based partially on 1v1 games, and some even FFA games. I've never seen this addressed.

This is not some weather data where the underlying theory does not change the data itself, rather how it is interpreted. In case of balancing the theory of how teams are determined creates said data, it makes no sense to me to retroactively try to deduce how effective it would be without a trial.
+0 / -0

33 days ago
quote:
I'm not going to claim some higher understanding of statistics, but how can you claim any sort of improvement/regression based on data that did not implement the discussed change?


If you can prove that your algorithm can predict the outcome of a match better than the existing one, then there's reason to believe (but yes no proof) that it can also produce a more balanced game.

Potentially, a perfectly balanced game each time doesn't produce the maximum amount of fun, or hours played, or player retention, or whatever it is we actually want to optimize. In that regard, it would be interesting to collect thumbs up / down from each player at the end of the battle and potentially feed that into balance.

I don't know how current system works, but it would probably be sensible to throw some noise in there to discover surprisingly interesting balances and also provide useful data for analysing the results.
+0 / -0

33 days ago
GBrankfiendicus_prime I'm not talking about the specifics of how it is implemented, but rather how using old data doesn't provide a valid point for effectiveness of new implementations. Since the data is not related to it.
+0 / -0
33 days ago
It is fine to test a rating system on data that has not been balanced with it because a rating system can also evaluate unbalanced games. You only have to replicate the rating history of the tested system instead of using a rating history of a different system. As GBrankfiendicus_prime said, there is a strong correspondence between predictiveness and balance.

I think that it would make sense to have a 1v1 and a team rating instead of a MM and casual rating. Further separation like small and big teams would probably not be worth the probably worse predictiveness and therefore probably worse balance.
+0 / -0


32 days ago
I think MM should not have rating intersection with battle rooms, and that FFA should not be rated. Other than that, go wild. I agree that
quote:
It is fine to test a rating system on data that has not been balanced with it because a rating system can also evaluate unbalanced games
but consider it to be a minor point, not worth arguing about. The aim is to make a rating system that people enjoy using, not to predict games with the most accuracy. I've seen the data that says a single unified rating has the greatest predictive accuracy.

Someone needs to answer the question of how ranks are displayed though. In a way that doesn't annoy a bunch of people. And someone would have to dive into the ratings code and change it, without crashing the server under the extra load.
+1 / -0

32 days ago
quote:
FFA should not be rated

If implementing / changing / adding a new / existing ladder is not so easy to implement I think this is the best idea.
Well you could change the autohost to "no-elo" but most of the players are in a non-autohost lobby and you don't go to the autohost to wait for players while there are several players in the non-autohost lobby - so you can wait yourself to death...

quote:
-All FFA Game automatically as "no-elo" to count no matter if it was played in the Autohost or in a "custom" Lobby.

Is this easy to implement?
+0 / -0
32 days ago
To be fair, players that are good in 1v1 tend to be good in pobpot as well. Maybe they are overrated in lobpot in the sense that 1v1 tactics don't always work out when a good players is swarmed by a bunch of noobs, but generally speaking, a Hoppili or a Slaab on the team usually has a big impact.
+0 / -0
quote:
Well you could change the autohost to "no-elo" but most of the players are in a non-autohost lobby and you don't go to the autohost to wait for players while there are several players in the non-autohost lobby - so you can wait yourself to death...

This has already been done, and since then there has been a definite shift towards seeding non-autohost lobbies. We are aware that making FFA truly no-elo will require more than modifying the autohost.

I think that there are two different questions here.

I think that abuses of the system like throwing away casual elo to other players in 1v1 or FFA games is a problem, and in the past it's been a problem addressed by moderator action.

I do not think it is obvious that, outside of deliberate abuse cases, 1v1 and FFA elo being linked to the rating used to balance casual teams games is a problem.

I am coming around to the opinion that the best solution to people abusing FFA to manipulate rating is to make FFA always no-elo. (Or stick it in its own rating category, but nobody has mana for that.) I am not as convinced that forcing custom room 1v1 to be no-elo is a better solution than moderator action in the rare cases that it is significantly abused.

quote:
Is [forcing FFA to be no-elo] easy to implement?

In theory, I believe it is not too difficult. In practice, I am not sure what is going on with making any server updates at the moment.
+0 / -0
The problem isn't the algorithm doesn't work in the case of players conducting normal gameplay, the problem is that the player base is actively manipulating the (any) system for their purposes. Ultimately there are players that want to increase their elo, and there are players that wants to win (more) games, they are different.

Any system change will have to take into account of players adapting to it and finding new ways to achieve their goals. The only way is to model player utility function distributions and nash equilibrium strategy given.....ahhhh the meta meta game

https://en.wikipedia.org/wiki/Goodhart%27s_law
+2 / -0
29 days ago
Good 1v1 players are probably solid team players. Solid team players aren't necessarily great 1v1 players. (me)
+1 / -0