Suggestions to improve point allocation

[HanabiGrandMaster]

(... a) win rate is a bad system as you do not really grasp the actual skill.

If more skill doesn't equal more win rate, then what does skill even mean?

A player who accidentally wins the first game would be at 100%.

No it wouldn't. If you take the average of (at least) 17 games, a player who accidentally wins the first game would be at 5.9% (or 58.8 ELO points).

Someone who plays thousands of games decently would not be able to get rid of the history of the initial games. And if you only count x latest games, the score might start jumping a lot if the player is trying out different strategies. But elo is actually a good system because you actually have to consistently play well to reach a higher score. With a low K value, it is harder to climb in score. Sure it's hard to fall too but that's actually good as your elo fluctuates less. Over time elo will adjust itself.

I don't think we need to discuss this further, since you clearly don't understand how ELO works. And if you think you do, then think of it this way: ELO is meant for 1v1. So for cooperative games it should reflect TEAM vs GAME. And team is expected to win 100% of the time.

What the actual problem at the moment is that all the game modes you can play are aggregated into one general elo score. So a 2p player and 5p player have the same number but they mean different things. Similarly whether you use 5 colors or multicolors or black powder or flamboyants.

That is not a problem. Pretty much all variants have similar theoretical winrates if you play perfectly (no cheating). The biggest game changers are "unofficial variant" (which balances 2p and 5p to be more similar to 3p and 4p games (higher winrates for 2p and 5p), and flamboyants who reduce your loss rates from ~5% to ~0.5%.

I'm sure your winrate proposal would not address that issue either. And it would require EACH game to be thoroughly analyzed for a specific scoring rule... while elo works well enough for every game with minimal simplistic asjustments.

But how about we just abandon the question if my maths is correct or not. With all the things you bring up, it feels like you're trying to make it personal.

I will give you a different perspective, by using these three assumptions:
- The majority of posts is about ELO doesn't reflect (Hanabi) skills.
- ELO is ranked too high.
- We can only filter on ELO's in steps of 100, or 200 points. (0+, 100+, 200+, 300+, 500+, and 700+ filters)

So can we please agree that ELO for cooperative games should be roughly between 0 and 900 points? Since the ELO can be an arbitrary number (ELO's are relative, not absolute), we should adjust the numbers to reflect the filters we can apply on BGA. This wasn't possible before, because people could boost ELO by abandoning. Since that feature was removed, it is now possible to change the ELO reward tables for ALL cooperative games.

If you don't agree, then what is the point of being so skilled that you are worth 1400 ELO points, but can only filter playing with people under 700 ELO points?

[Stroom]

If more skill doesn't equal more win rate, then what does skill even mean?

Win rate is NOT equal to skill. For Hanabi it does not make sense. You might have different partners who play differently. Some worse, some better. The other players would take down your win rate.

Also, what you seem to consider a win is only max score. But for elo calculations you can also "win" at 28 or 29 or even less if your average elo for the team is low enough. It is not as black and white as you make it seem. A win for 5 color is not the same as a win for black powder + multicolors.

A player who accidentally wins the first game would be at 100%.

No it wouldn't. If you take the average of (at least) 17 games, a player who accidentally wins the first game would be at 5.9% (or 58.8 ELO points).

Oh, OK. Sure, in this case it might work... but it still only applies to max score. This is just bad. You could have a string of bad luck or bad partners who bring down your score. If it's only 17 games, it will influence your stats greatly. And it stops representing your skill if 5-10 games can drop your score significantly. With current elo it does not do that.

Someone who plays thousands of games decently would not be able to get rid of the history of the initial games. And if you only count x latest games, the score might start jumping a lot if the player is trying out different strategies. But elo is actually a good system because you actually have to consistently play well to reach a higher score. With a low K value, it is harder to climb in score. Sure it's hard to fall too but that's actually good as your elo fluctuates less. Over time elo will adjust itself.

I don't think we need to discuss this further, since you clearly don't understand how ELO works. And if you think you do, then think of it this way: ELO is meant for 1v1. So for cooperative games it should reflect TEAM vs GAME. And team is expected to win 100% of the time.

I know how elo works. I have programmed an implementation of it in a personal project. For team games it is also a similar implementation where you just take an average to represent the team score. Sure, it was originally for 1v1 but it does work for 1vX and team vs environment as well. For 1vX you just calculate as if for every X you played a 1v1 game. For team vs environment you have an arbitrary opponent strength that you as a team have to beat. You don't seem to grasp how it works and you aren't happy with your current score, it seems.

Not sure where you take it that a team is expected to win 100% of the time. Bad decks can happen. A hand can become completely full and you have 0 clues so you have to discard blindly.

BTW the correct way it to use lower case for elo.

What the actual problem at the moment is that all the game modes you can play are aggregated into one general elo score. So a 2p player and 5p player have the same number but they mean different things. Similarly whether you use 5 colors or multicolors or black powder or flamboyants.

That is not a problem. Pretty much all variants have similar theoretical winrates if you play perfectly (no cheating). The biggest game changers are "unofficial variant" (which balances 2p and 5p to be more similar to 3p and 4p games (higher winrates for 2p and 5p), and flamboyants who reduce your loss rates from ~5% to ~0.5%.

You completely do not understand the situation. Don't try to fix it if you do not understand Hanabi or the scoring system. You can't balance different game modes under one elo. There are so many more variables and varying skills can influence the outcome differently so the calculation would never work to fairly address it. Apparently your analysis of the theoretical win rates is off too. You do not take into account that player hands can get clogged and how often it can happen. I'd say that flamboyant can reduce "loss rate" more than that if you use it correctly. Still, a 29 score is not a loss for elo calculations if your team average is low enough. Your entire basis of win and loss is incorrect.

Still, don't try to create a complicated new system. The current system is good enough. At the most, it just needs to lower everyones elo by 200 points so the current 700s and 800s would not be masters anymore.

But how about we just abandon the question if my maths is correct or not. With all the things you bring up, it feels like you're trying to make it personal.

I will give you a different perspective, by using these three assumptions:
- The majority of posts is about ELO doesn't reflect (Hanabi) skills.
- ELO is ranked too high.
- We can only filter on ELO's in steps of 100, or 200 points. (0+, 100+, 200+, 300+, 500+, and 700+ filters)

So can we please agree that ELO for cooperative games should be roughly between 0 and 900 points? Since the ELO can be an arbitrary number (ELO's are relative, not absolute), we should adjust the numbers to reflect the filters we can apply on BGA. This wasn't possible before, because people could boost ELO by abandoning. Since that feature was removed, it is now possible to change the ELO reward tables for ALL cooperative games.

If you don't agree, then what is the point of being so skilled that you are worth 1400 ELO points, but can only filter playing with people under 700 ELO points?

Not personal.

1. Elo roughly does represent skill. But it does not explain whether it is for 2 player games or 3 player or 5 player... and what game mode you play. I only play the same game mode and I feel that I have mastered it. I don't say that I would play as well with other modes.
2. It does not matter how high the elo score is. Everyone is judged on the same basis and everyone can get a very high score. In chess, some grandmasters are at 2000+ in real life too.
3. Yes, the filters do not let us go higher and that is an issue for me. But I can play quite well with most 700s too. Winning with them increases my score while playing with 900s would at best keep the score the same.

No. Elo does not need t be only up to 900. Some people can master the game better and elo does not have an upper limit. Elo is an arbitrary number just as well as skill is an arbitrary and subjective thing. Elo seems to give us a somewhat consistent basis for it. If only each game mode had a different score...

If someone has climbed a lot in elo then if he filters by that high elo, he might not find anyone to play with for hours. By the current system, seems like at about 1000 team average elo, if you win, you gain 0. So playing with lower levels is actually more beneficial.

Seriously, it works well enough. Maybe the calculation might need to set a "stronger" virtual opponent for the elo baseline.

[ChiefPointThief]

That is not a problem. Pretty much all variants have similar theoretical winrates if you play perfectly (no cheating). The biggest game changers are "unofficial variant" (which balances 2p and 5p to be more similar to 3p and 4p games (higher winrates for 2p and 5p), and flamboyants who reduce your loss rates from ~5% to ~0.5%.

How did you come up with the flams number? Also loss is anything that isn't a perfect score?

With all the things you bring up, it feels like you're trying to make it personal.

Outside of your convo I have seen some comments on the subject that seemed personal. Apparently talking hanabi elo is a serious matter

[ChiefPointThief]

You can't balance different game modes under one elo. There are so many more variables and varying skills can influence the outcome differently so the calculation would never work to fairly address it. Apparently your analysis of the theoretical win rates is off too. You do not take into account that player hands can get clogged and how often it can happen. I'd say that flamboyant can reduce "loss rate" more than that if you use it correctly. Still, a 29 score is not a loss for elo calculations if your team average is low enough. Your entire basis of win and loss is incorrect.

Still, don't try to create a complicated new system. The current system is good enough. At the most, it just needs to lower everyones elo by 200 points so the current 700s and 800s would not be masters anymore.

So you agree w/ my initial analysis about flams and that the current system isn't up to par. How much do you think they increase ones elo? Also w/ all the variables you mentioned why still are people taking hanabi elo so seriously?

I know how elo works. I have programmed an implementation of it in a personal project. For team games it is also a similar implementation where you just take an average to represent the team score. Sure, it was originally for 1v1 but it does work for 1vX and team vs environment as well.

Did you leave out team vs team on purpose? That is really bad and hopefully bga considers changing it one day.

[Stroom]

You can't balance different game modes under one elo. There are so many more variables and varying skills can influence the outcome differently so the calculation would never work to fairly address it. Apparently your analysis of the theoretical win rates is off too. You do not take into account that player hands can get clogged and how often it can happen. I'd say that flamboyant can reduce "loss rate" more than that if you use it correctly. Still, a 29 score is not a loss for elo calculations if your team average is low enough. Your entire basis of win and loss is incorrect.

Still, don't try to create a complicated new system. The current system is good enough. At the most, it just needs to lower everyones elo by 200 points so the current 700s and 800s would not be masters anymore.

So you agree w/ my initial analysis about flams and that the current system isn't up to par. How much do you think they increase ones elo? Also w/ all the variables you mentioned why still are people taking hanabi elo so seriously?

Flams make it easier somewhat... but in 2p you need more clues at times and can easily run out which causes worse situations when you don't necessarily get a clue back for playing the 5. And often the flam you get has no use at that time either. You have to adjust your strategy in a way that you only play 5s when it is (or seems based on your missing knowledge) unavoidable. So it is not 100% better but might be at a higher skill level.

Still, I don't see that the system needs to be changed drastically. You just need to understand that the skill level is just expressed 100-200 points higher than "in the old days". Still, something that you can easily get used to rather than wasting time on trying to make a new system which will definitely have it's own flaws. Overall, there is no objective way to calculate a balanced elo over several game settings. Base games are inherently different from black powder, multicolor and flams games. Creating a different elo score for each is overkill.

Just accept the fact that there is one number that is a rough estimate about a players skill in the type of games that he is currently playing. You can't compare one players score to another players score and say that one is definitely more skilled in EVERY game mode. And that is OK. What you can do is that you know that 300, 600 and 900 elo players are all definitely different.

Hanabi elo is not a serious metric. I don't take it seriously and I don't see that there is a need to fix it. Hanabi still has some actual gameplay bugs that need fixing first. Like sometimes a multicolor is marked with multicolor at the end phase when it gets the first color clue on it. If someone takes it seriously, it's their problem. Losing 10 elo for abandoning games is OK. Allocating points as is is OK.

I know how elo works. I have programmed an implementation of it in a personal project. For team games it is also a similar implementation where you just take an average to represent the team score. Sure, it was originally for 1v1 but it does work for 1vX and team vs environment as well.

Did you leave out team vs team on purpose? That is really bad and hopefully bga considers changing it one day.

There is no team vs team in BGA AFAIK. But it is similar to team vs environment where the environment is also a team with it's own average elo. So it will be reduced to a 1v1 where both teams gain or lose the same amount. OR you can make it work in a way that each player in team A wins vs each player in team B individually so the eventual +- elo is individual. There is no need to change anything IMO. Nothing bad. It is a simple scoring system that is not to be taken seriously and it will roughly estimate player skills.

[ChiefPointThief]

Did you leave out team vs team on purpose? That is really bad and hopefully bga considers changing it one day.

There is no team vs team in BGA AFAIK. But it is similar to team vs environment where the environment is also a team with it's own average elo. So it will be reduced to a 1v1 where both teams gain or lose the same amount. OR you can make it work in a way that each player in team A wins vs each player in team B individually so the eventual +- elo is individual. There is no need to change anything IMO. Nothing bad. It is a simple scoring system that is not to be taken seriously and it will roughly estimate player skills.

I don't know what you mean. There are several team games on bga (spades for one).

There is no need to change anything IMO. Nothing bad. It is a simple scoring system that is not to be taken seriously and it will roughly estimate player skills.

Tell that to the countless players who have been asking for change for yrs. Also it isn't a rough estimate because what it does is keep everyone's elo closer than what it should be making players appear equal when they really aren't. This is why there are really no "strong players" in spades. The only way to achieve this is to constantly partner w/ a equally strong player. If you are good you can reach strong status but it won't be maintained playing w/ the field. If you play with a weaker teammate you receive less elo for wins and lose more for losses when it should be the opposite.

[Stroom]

I don't know what you mean. There are several team games on bga (spades for one).

There is no need to change anything IMO. Nothing bad. It is a simple scoring system that is not to be taken seriously and it will roughly estimate player skills.

Tell that to the countless players who have been asking for change for yrs. Also it isn't a rough estimate because what it does is keep everyone's elo closer than what it should be making players appear equal when they really aren't. This is why there are really no "strong players" in spades. The only way to achieve this is to constantly partner w/ a equally strong player. If you are good you can reach strong status but it won't be maintained playing w/ the field. If you play with a weaker teammate you receive less elo for wins and lose more for losses when it should be the opposite.

I don't know how BGA has implemented team games. Probably it is one of the ways I explained before.

And there are countless players who are fine with how it is at the moment. Probably even more. You can't really make it work much better as there are different way to boost one's elo score. Just play with friends or tell in chat what cards each player has. Any system can and will be abused. So just don't bother.

[Wreckage]

An interesting point about win rate: If you play on training mode it doesn’t effect your elo, but each game is counted as a loss for your win percentage, even your perfect scores. I have probably played a couple thousand games on training mode. Every one of them is counted as a loss to my win percentage.

Now that there is a penalty for quitting games, the current system is pretty good. That said, there are improvements that could be made.

1. More even elo for all player counts. Someone who generally plays 3p games will have about 150 less elo than someone who plays 5p games, if they have equal success. This is ridiculous. I don’t even think 3p is an easier game. There should be a slight (emphasis on “slight”) difference for each number of players. 2p is the most difficult, then it’s slightly easier with each added player. Perhaps 4p and 5p should gain the same elo.

2. There should be separate adjustments for black powder and flamboyants. I’ve seen it stated many times that flams don’t help scores much on a tables of all masters, because they don’t let important cards get discarded anyway. This is wrong. If a table of masters play 10 games without flams, they will have about 6 perfect games, and about 3 - 4 games where the last card in the deck matched an early discard. If flams are on, all these games become 30s.

3. There should be an adjustment for unofficial variant. This should effect elo on 2p and 5p games in opposite ways.

[Jellby]

An interesting point about win rate: If you play on training mode it doesn’t effect your elo, but each game is counted as a loss for your win percentage

It's not counted as a loss. It's counted as a non-win. There are wins, losses, ties, and training. The win percentage counts only wins.

This is wrong. If a table of masters play 10 games without flams, they will have about 6 perfect games, and about 3 - 4 games where the last card in the deck matched an early discard. If flams are on, all these games become 30s.

How? If the two life-saving 5s come early, when the useful card hasn't been discarded yet, they don't help. You also lose the +1 clue of some 5s, which might mean you do worse than without flams. There's no guarantee a non-30 game will become 30 with flams.

[ChiefPointThief]

An interesting point about win rate: If you play on training mode it doesn’t effect your elo, but each game is counted as a loss for your win percentage

It's not counted as a loss. It's counted as a non-win. There are wins, losses, ties, and training. The win percentage counts only wins.

This is wrong. If a table of masters play 10 games without flams, they will have about 6 perfect games, and about 3 - 4 games where the last card in the deck matched an early discard. If flams are on, all these games become 30s.

How? If the two life-saving 5s come early, when the useful card hasn't been discarded yet, they don't help. You also lose the +1 clue of some 5s, which might mean you do worse than without flams. There's no guarantee a non-30 game will become 30 with flams.

Is your comment about flams based on all player counts?