webdiplomacy.net

Here's an interesting new article by BunnyGo!

It's amazing how just the one topic of scoring generates as much talk as any of the actual game concepts.

I agree that very close to 100% of dip scoring is a total lie and a waste. The points we use on this site are a perfect example. They teach people to lose so they can gain more points.

The idea of utility points is insane and a waste as well.

Points should be based around skill as a player, according to the rule : "Solo if you can, draw if you cannot solo."

Allowing an enemy solo means you did nothing right and deserve zero points.

People in a draw deserve more points for more centres, as a guy with one centre in a draw has played better than a person with 17 centres in a draw.

Solo should be tons of points, as all 6 other players scored a zero in that game.

Almost 100% of games should end in a draw; recall, the rules state "Get a solo or a draw if you cannot solo".

ALL players are bound by the rules to gun for a solo, and are only allowed to play for a draw if a solo is impossible. This is in the rules. It is against the rules to aim for a draw from the start. Literally against he rules.

So, for scoring, the most important is your wins/draws as a percentage of your total games. Then factor in the skill of your opponents, and BAM you have your score.

Ghost ratings are close, points are not, your record is pure.

I don't think you get the concept of utility points, Machiavelli. According to the definitions of BunnyGo, utility points are used to gauge the success of players in any particular game, while skill points are used to gauge the success of players across multiple games so as to reflect how skilled these players are. In order to tell how someone does across multiple games, you first have to determine how well that person does in any particular game. As a result, utility points are always derived from skill points. For example, your change in Ghost Rating (skill points) after playing a match is necessarily dependent on the scoring system (utility points) used in that particular match.

So when you say you think the idea of utility points does not make sense, but you are in favor of skill points, that does not make sense to me.

I do not support skill points nor utility points.

The way you measure skill in a particular game is by counting the centres you have or how close you are to a draw etc, and then factoring enemy skill. ("skill" the word, not Skill Points)

Skill across multiple games is simply your record examined through the lens of your enemy's skill. (again, the word not the points)

One of the problems with dip and games in general is people work very hard to make it far more complex than it has to be. Creating new points to measure skill inside 1 game? Madness

Macchiavelli wrote: ↑

Fri Oct 23, 2020 1:47 am

I do not support skill points nor utility points.

The way you measure skill in a particular game is by counting the centres you have or how close you are to a draw etc, and then factoring enemy skill. ("skill" the word, not Skill Points)

Skill across multiple games is simply your record examined through the lens of your enemy's skill. (again, the word not the points)

One of the problems with dip and games in general is people work very hard to make it far more complex than it has to be. Creating new points to measure skill inside 1 game? Madness

It sounds like you might favor a "skill points" awarded by committee of expert judges at the end of a game. Judges consideration for "tactical play" and "general strategy" and "correctly going for/stopping a solo" etc. etc. Something like gymnastics or whatnot where the skill itself is an art form.

I think what you're saying is that you want a way to predict in the future which players are likely to "perform well" and which are likely to not.

If you're trying to predict future success, using a metric like "messages sent per game" may be as successful as any scoring system.

As a crude test of that theory, I calculated the messages per press game for all the players in the GR1 game and the players in the GR5 game.

The GR1 players have 110k messages over 346 games for 320 messages per game.
The GR5 players have 22k messages over 170 games for 133 messages per game.

When you say "games", did you calculate it for their total number of games? Or did you discount the Gunboats? If the former, I have 50 messages per game..

In your profile, there's a tally for the number of classic press games. I just used that. If anything, it overcounts messages/game because it excludes variants in the game count.

I don't think any ranking system aims at predicting future outcomes.

Counting messages per game is an interesting statistic, but don't confuse correlation with causation. A player that is eliminated early will send fewer messages than a player who lasts until the end.

RoganJosh wrote: ↑

Fri Oct 23, 2020 3:40 pm

I don't think any ranking system aims at predicting future outcomes.

Counting messages per game is an interesting statistic, but don't confuse correlation with causation. A player that is eliminated early will send fewer messages than a player who lasts until the end.

A player that was eliminated early probably did worse than a player that lasts until the end...

Also, I think the precise reasoning behind many ranking systems (e.g. Elo, etc.) is to be able to predict future outcomes.

Bring back PPSC

RoganJosh wrote: ↑

Fri Oct 23, 2020 3:40 pm

I don't think any ranking system aims at predicting future outcomes.

Counting messages per game is an interesting statistic, but don't confuse correlation with causation. A player that is eliminated early will send fewer messages than a player who lasts until the end.

You don’t view ELO in chess or Dan level in Go as predictive?

Yonni wrote: ↑

Fri Oct 23, 2020 4:33 pm

A player that was eliminated early probably did worse than a player that lasts until the end...

Yeah, I misread your previous statement. Correlation is all that's needed to make a prediction.

BunnyGo wrote: ↑

Fri Oct 23, 2020 5:16 pm

You don’t view ELO in chess or Dan level in Go as predictive?

No? But I'm not sure what you mean by 'predictive'. ELO approximates skills of players conditioned on outcomes of games. Shouldn't a predictive variable approximate outcomes of games conditioned of the skills of the players?

When calculating an elo rating, you literally calculate an expected score and compare the actual outcome to that expectation, then adjust ratings based on that comparison.

So yes, the point of an elo rating is to attempt to predict the outcome. By taking the difference in rating between two players, you can predict that player X will win y% of the games they play.

An expected score is not a prediction.

A prediction is a statement on how likely different outcomes are. For chess, it would be percentages for how likely white win / black win / draw is. ELO provides no such prediction.

Isn't that exactly what an expected score is or am I missing some nuance there?

No, the expected score is a mean. It represent the percentage of games you would expect to win if a series of games are played.

A prediction is a statement about the outcome in one single game.

They're related, but they're not the same.

You can have an expected score of .57. It mean that you should win 57% of the game. But predicting that you will win 57% of one game doesn't make sense.

In order to make a predictions, you also need to know something about the variance. Expected return you only need to know means.

RoganJosh wrote: ↑

Fri Oct 23, 2020 11:00 pm

No, the expected score is a mean. It represent the percentage of games you would expect to win if a series of games are played.

A prediction is a statement about the outcome in one single game.

They're related, but they're not the same.

You can have an expected score of .57. It mean that you should win 57% of the game. But predicting that you will win 57% of one game doesn't make sense.

In order to make a predictions, you also need to know something about the variance. Expected return you only need to know means.

The Elo system runs on the assumption that all players skill is a normal distribution with the same variance, but different means. It's a very imperfect model, as the normal distribution doesn't do a great job, but it does have a variance built into its assumptions.