Page 1 of 1

Ranking Points Question

PostPosted: 17 Feb 2017, 22:22
by Strategus
A question on a recently completed game - 125011. Are you up for the challenge? (1500+)

3 way draw awarded points as follows (I believe):-
Charleroi Previous Rank 1881 New Rank 1913 - 32 points (win)
GPD Previous Rank 1887 New Rank 1903 - 16 points (win)
Durkeety Previous Rank 1873 New Rank 1885 - 12 points (win)
Big Gun Previous Rank 2275 New Rank 2265 -10 points (loss)
Zosimus Previous Rank 1953 New Rank 1949 -4 points (loss)
Other differentials unknown.

Seem to be odd, as for the same result, players seem to change places in relative ranking. How is this possible?

Also, ranking awards seem low in a high ranking game (even though it is only a three way). Is this how it is supposed to be?

Re: Ranking Points Question

PostPosted: 17 Feb 2017, 23:17
by Nanook
It'll take dipsy or someone with more knowledge than I, but I suspect the reason is the amount of games you've each played. The more games you have under your belt, the fewer points you gain or lose from any given game.

So for this game, looking at the three draw members, charleroi has 12 finished games, you have 32, and Durkeety has a whopping 188. So even though your ratings were pretty close to each other coming in, from the system's point of view you and Durkeety have shown you're pretty much at that level...while Charleroi has more variance, where it's not sure if he's at that level, a lower one, or a higher one. So a positive result will net him more points, like here, but a negative result would also lose him more points.

Hope that clears things up for you, and congratulations on the draw :)

Re: Ranking Points Question

PostPosted: 17 Feb 2017, 23:20
by WHSeward
The algorithm is a secret so only Dipsy could debug any individual calculation. However, there are plenty of reasons score changes could look like that.

Why players in a game may get different points is explained in detail in this post stickied to the top of this sub-forum.

As for the "awards seem low in a high ranking game", at PlayDip the absolute rating of players does not affect the size of the award, only the relative rating of a player compared to the average rating of the opponents. That means, all else being equal:
* If the game has all 2000 rated players, the points award is the same as if it had all 1000 players.
* A player with a 1000 rating will get more points winning a board with 2000 rated opponents than with 1000 rated opponents.

Re: Ranking Points Question

PostPosted: 17 Feb 2017, 23:48
by Durkeety

^sums up my reaction to such things

Re: Ranking Points Question

PostPosted: 17 Feb 2017, 23:56
by Nanook
Durkeety wrote::(

^sums up my reaction to such things

It's worse for us guys with middling ratings and a good number of finished games, if that's any comfort :)

Re: Ranking Points Question

PostPosted: 18 Feb 2017, 08:23
by super_dipsy
No rating system will make everyone happy I'm afraid. However, as already answered the explanation is given in detail in the referenced post which WHS linked you to.

Just to focus on the key points relevant to this discussion:

Remember that the rating system was put in place because people did not like the fact that your 'score' in a game had no relation to the people you played against. So a win against a bunch of newbies who don't know what they are doing was exactly the same as a win against the top 6 players on site. Therefore, the rating system caclulates your result based on your RELATIVE 'playing strength' to the others. As WHS showed with his excellent example, if you are all rated 2000 in a game, it will be the same as if you were all rated 1000. But if you win a game where you have a 1000 rating and all your opponents 2000, you will score heavily, whereas if you win a game as a 2000 against all 1000 ratings, you will not score as much. And of course the reverse is true if you lose - it will hurt more losing against lower ranked players than higher ranked ones.

The rating system tries to represent your 'playing strength'. This then let's the previous bullet work. But how do you work out a player's strength? Well, the only thing you have to go on (obviously) is their playing record against others. This means that for a new player, you have absolutely no idea. The new player could be a Diplomacy superstar that has just found us, or someone who has never even heard of the game before. That player could be the greatest or the worst in the world. So what the system does is it tries to get that player to their 'playing strength' as quickly as possible, so that the integrity of the overall system is as closely maintained as possible (since the longer people have 'false' ratings the more it effects bullet 1). It does this by making guesses based on very little data but then refining those guesses as quickly as possible as more games are played.

Think of it like this. a new player arrives and solos their first game. This could be just plain lucky (we've all been there!) or be because the player is a superstar. With the lack of any extra detail, the system gives the soloist a big swing because on that 1 game sniff test this seems a strong player (after all, it is not actually that easy to win at Diplomacy!). After 10 games though, the pattern will be clearer. You will be seeing more clearly how that player performs against different strengthed opposition. What you don't want is for a situation where experienced and skilled players have huge swings from a single loss (or win) that shift them from their strength range. For example, if a player consistently wins 7 or 8 out of every 10 games against similar opposition over a record of 100 games played so far, then we have a pretty good idea of that player's strength (say around 2000). If that player now loses a game, you don't want a big swing dropping that player to 1800 because that would be a false reading. Progress up and down shoudl reflect not just the immediate run of results but the player history. Of course, if a player gradually improves then you would want to see the player's rating gradually improving too, and this is what happens.

The system therefore gives bigger swings when you have played fewer games, damping down to a 'normal' range of swing once you have played a substantial number of games and it is fairly confident of your strength.

One final point. Somewhere in the scoring forum I know there are pictures and data for this, but one common comment is 'I wish I had won more games when I first started because the swings were bigger. Although I am winning more now they are worth less to me'. Actually, this is a fallacy! Running the model against two scenarios, one where a new player comes onto site and wins 10 games in a row and loses the next 50 and the other where a player loses the first 50 and wins the next 10, the second player actually ends up with a slightly higher rating than the first. This is because the later losses hurt more from a higher rating base (because your result is based on your relative strnegth to the opposition) and similarly the later solos in the second case are better because you are winning from a lower base. The damping effect does not outweigh the 'comparative strengths' effect.

Re: Ranking Points Question

PostPosted: 20 Feb 2017, 13:57
by Guns of Brixton where a new player comes onto site and wins 10 games in a row and loses the next 50...

Knew I shouldn't have given the kids the password to my account!

Re: Ranking Points Question

PostPosted: 30 Apr 2021, 17:24
by ElCariboutchou
How about a specific Ranking for PPO games ?