Board Game Arena

Yes. I missed the post too. It is well explained though. You can find it in announcements.

Juliar86 wrote: ↑17 December 2023, 12:03 Yes. I missed the post too. It is well explained though. You can find it in announcements.

Is it, though? Do you accept their explanation that invisible ELO is better for the Hanabi community because we shouldn't be so competitive, so they unilaterally decided to remove it without consulting said community?

First off, they have no clue how important ELO is for making in-game decisions about clue complexity.

Second, as Geraldine put it:

GeraldineMerida wrote: ↑16 December 2023, 19:04 And when you do announce changes, please don't be so patronising: who are you to decide what I should be focusing on when I play a challenging co-op game like Hanabi?

I disagree that the developers and BGA team do not understand the importance of ELO... These are the same folks that have finally looked at the extraordinarily high rate of RT's and negativity from the BGA hanabi community. Their intentions for changing the system were clearly as a result of this in order to stem the overflowing tide of vitriol.

But I'm also not a big fan of what was chosen. Either disable ELO completely and allow the community to self-select their desired level of playstyle, through more match-making options, or make the ELO system more robust to do it automatically. Impossible to live in this middle-ground.

IMO, it should show the players total number of games and they opt for what kind conventions they want to play with, so all confusion is cleared up at the onset. 90% of people losing their cool in this community is some assumption about a playstyle or move choice that isn't universally understood or ubiquitous. Very rarely are players intentionally making the first 'mistake'.

I don't really understand this move to be perfectly honest. By hiding away players specific ELO's makes it much harder to tell what skill level they actually are. An average ranked player could be someone who is really bad, got over the 100 line and now can't go below due to the rules or they could be someone nearly at the 200 mark working their way up. Also expert ranking can mean anywhere between 500-700 which is a huge leap.

Earthboundia wrote: ↑19 December 2023, 14:35 I don't really understand this move to be perfectly honest. By hiding away players specific ELO's makes it much harder to tell what skill level they actually are. An average ranked player could be someone who is really bad, got over the 100 line and now can't go below due to the rules or they could be someone nearly at the 200 mark working their way up. Also expert ranking can mean anywhere between 500-700 which is a huge leap.

Yes, exactly!

Master ranking is even worse -- are you 700 or 1150? There's a huge gap in knowledge, understanding and trust across 450 ELO. I still play with low ranked Masters, but their ELO informs my in-game decisions about clue complexity and whether or not I have to save a promised card when it reaches the chop.

I think seeing the tiers is enough for Hanabi. It's unreliable anyway unless the most serious calculation-flaws are fixed:

* lower ELO gain in 2p games
* lower ELO gain in games using casual mode (aka flamboyants) - by a lot. This makes such a big difference
* increase ELO gain in games with black

Additional tiers would be nice though (best would be all multiples of 100 but I guess it's difficult to come up with more names for them)

Silene wrote: ↑20 December 2023, 10:49 I think seeing the tiers is enough for Hanabi. It's unreliable anyway unless the most serious calculation-flaws are fixed:

* lower ELO gain in 2p games
* lower ELO gain in games using casual mode (aka flamboyants) - by a lot. This makes such a big difference
* increase ELO gain in games with black

Additional tiers would be nice though (best would be all multiples of 100 but I guess it's difficult to come up with more names for them)

There is no way to correctly balance elo this way. Who says that 2p is easier than 3p or 4p? How much easier? Same for black vs flamboyant... I'd say that 2p+Multi+Black is much harder than 4p+Multi+Black because you have less room in players hands to save cards so you more likely have to let something go.

So the best way would be to just hold a separate elo based on specific game settings.

Stroom wrote: ↑20 December 2023, 17:17 There is no way to correctly balance elo this way. Who says that 2p is easier than 3p or 4p? How much easier?

Theoretically, you really could correctly balance the “bot” ratings to calculate a single ELO rating to measure overall Hanabi skill. I don’t think BGA would ever implement such a system, but it’s possible. Data would say which formats is easier, and by how much.

You do it this way… Ignore all the existing ratings on the site entirely, and start fresh. Instead of starting players at 0, pick some mid-point so that there’s room to move up and down. (0 can be a mid-point, but only if you want to be working with negative ratings, which nobody does.) You also create a set of “bots” that the real players are playing against; each bot is assumed to always score a particular value, and players beat the bot of they score higher than the bot, and lose if lower. This is similar to what they do on BGA right now.

(There’s two major variations of this approach I can think of. You can set the bots to score integer values from -5 to 30, or a subset of the range, as is the case now. Or you can set the bots to score half points 0.5, 1.5, 2.5, …, 24.5. If you use half-point bots, you don’t have to worry about ties against a bot, because player either win or lose against each bot.)

Now here’s the key difference from what BGA does now. Don’t try to eyeball the bot ratings. Just set them all to the same starting point as the players, but when players win against a bot, the bot loses the points the players gain. When players lose, the bot gains the points too. The same way that the ELO system allows player ratings to reach an equilibrium point, and fluctuate around that equilibrium as games are played, the bot ratings would reach their own equilibriums.

Don’t just compare scores against the highest bot the players beat, compare against all of them. That’s how players get more points for higher scores.

And then you have an actual measure of just how difficult it is to obtain each possible score in each variant of the game.

You’d set the k-value to something pretty low, to take into account that each player is being compared to each bots. Otherwise you could wind up awarding a couple of hundred points to players when they score 30.

And you could pick a point in time and run that algorithm forward using game data BGA already has, if you don’t want to hard reset everyone on the site.

You could slice the games up any way you like - a set of bots each for 2-, 3-, 4-, and 5-player, for example. But you can get finer - a set of bots for 2-player 5-colour, another for 2-player w/6th suit multi, another for 2-player w/5-card 6th suit multi, another for 2-player w/Avalanche, then all of those with Black Powder added. And so on through all variations and combinations. Taking into account all the variations I can think of, and slicing as finely as possible, you would have 1362 bots, which players play against in sets of 26, 31, or 36, depending on the number of suits in use in the game. (You may want to make a minor adjustment to k-value to account for the number of bots in a set.)

Again, I don’t think BGA ever will do something like this… but in theory, you could.

Stroom wrote: ↑20 December 2023, 17:17 There is no way to correctly balance elo this way. Who says that 2p is easier than 3p or 4p? How much easier? Same for black vs flamboyant... I'd say that 2p+Multi+Black is much harder than 4p+Multi+Black because you have less room in players hands to save cards so you more likely have to let something go.

So the best way would be to just hold a separate elo based on specific game settings.

If the point is to increase community & cooperation, then 5p games should get the highest return on ELO, then 4p and so on. 2p should be the lowest. It's far easier to start a 2p game, it plays faster, and it's so much easier to find 1 other player at the same convention level as you than 4 others.

But I would be happier with a separation between 2p and Multi-player instead. I dislike playing with Masters who climbed the ranks only playing 2p -- not because I'm a 5p elitist, but because the conventions are vastly different, and they take a lot of time to learn. If I'm playing a Master-restricted 5p game, I expect Masters of 5p games, otherwise we're simply not speaking the same language.

All that being said, it still makes no sense to have invisible ELO. There are far better solutions for negativity at Hanabi tables, and it starts with helping players make their table's conventions more clearly known.

Travis Hall wrote: ↑20 December 2023, 23:50

Stroom wrote: ↑20 December 2023, 17:17 There is no way to correctly balance elo this way. Who says that 2p is easier than 3p or 4p? How much easier?

Theoretically, you really could correctly balance the “bot” ratings to calculate a single ELO rating to measure overall Hanabi skill. I don’t think BGA would ever implement such a system, but it’s possible. Data would say which formats is easier, and by how much.

You do it this way…

Again, I don’t think BGA ever will do something like this… but in theory, you could.

This does not seem to work. It assumes that all players play to the fullest of their skill. That there are no intentional quitters. And seems like it averages all the players. Overall, it still seems like a subjective way to do it.

Let me rephrase it - there is no practical way to do this. It needs far more effort than anyone willing to put into it, and even then, you might not see the benefits of it.