webdiplomacy.net

Yonni wrote: ↑

Sat Jan 20, 2018 10:26 pm

Also, I wasn't sure how to deal with rulebook press. I gave it the same rating as full press. For the sake of rankings, I imagine I can just combine rulebook and full press?

That sounds sensible to me. Probably one of the easier calls you have to make.

As for full press live world games... I suggest a weighting of 0.00

For PPSC (and SoS) do people score:
(a) their number of supply centres divided by the total number of SCs
or
(b) their number of supply centres divided by the total number of SCs in the draw?

This only really affects games where there is a resign and not all the SCs are in the final draw pool.

Git repo for those who are interested. Yonni, I can add you as a contributor if you want to push changes there directly.

PPSC and SoS only score centers actively involved in the draw. Additionally, PPSC "removes" any centers beyond a winner's 18, so if Russia wins with 25 centers and 3 belong to a resigned player, the divisor will be 24 instead of 34. I have no idea how this interacts with the quirk in variants of different sizes where the number of centers needed to win is 18/34 of the total rather than one more than half, but I'd guess not well.

Weird - at least 200 odd of my game seem to be missing from your data dump.

Ranking 95K seems a tad extreme.

Note PPSC is only ppsc if there is a solo, otherwise it is draw size.

Oddly, SoS is the opposite: it is only SoS if there's no solo, otherwise the soloist takes it all.

CCR, you are correct for assuming I got it wrong.

So, does it work like this?

For PPSC:

Solo:
-winner gets 18/34 (or solo count/number of centres for variants)
-Losers get centreCount/survivorCentres

Draw:
-1/drawSize

That seems crazy. Is that really the case? PPSC is dumb if it is.

For SoS:

Solo:
-Winner takes all (now, that makes sense)

Draw
-centres^2/sum(non-resignedCentres^2)

Winner's divisor is that same as the survivors' divisor (not always 34) for PPSC to keep it zero-sum, but otherwise yes. This creates an odd scenario where taking extra centers does help, just not as much.

I can't seem to open the file in Google Spreadsheets and I'm a poor who doesn't own Microsoft Office. Anybody got me?

@PE
You're ranked 41

I like the one where I was #7 better.


Would be interested to see the whole list. Great work Yonni. Who's #1?

To your defense, it's an all time list, with no drop offs.

Yonni wrote: ↑

Sat Jan 20, 2018 8:46 pm

How does it differ from Ghost Rating?

Tons of respect to The Ghost Maker for creating and maintaining a ratings system for us but it had several shortcomings. The part that bothered me most was that losses against expert opponents and beginners are rated equally. That is not the case for EIDRaS which calculates your expected score based on your opponents ratings. There are other differences too of course but that's the big one for me.

That is not true; in Ghost Rating, losses against expert opponents and beginners are not rated equally.
EIDRaS Rating does look a bit better than Ghost Rating to me though.

My limited understanding of GR was that each player essentially contributed to a point pot, with the size of the contribution determined by their current rating, and who took the points was determined purely by the results. I rather liked the idea of it.

This system I understand even less, but I'm not a fan of an expected score being based on the ranking of people you're up against. This site has a healthy number of players new to the site but experienced in the game elsewhere. The idea of a high ranking player being given a ranking mountain to climb because they're playing a player of similar skill who plays mostly on another site is not one I'm comfortable with.

Sorry Octavious your complaint makes no sense, for the elo ratings are meant to work exactly this way! Note that these variations are expected, if your real skill is higher than the current numbers show, then you will eventually get more points; if you don't, it is bcause your real rating is low. All players start at the the same beginners level, say a hundred points in gr, a thousand points in jdpr/eirdra, and then if the guy is good he climbs up.

This is exactly how world chess rankings are made, for we've borrowed the formula from them, and Kasparov began at average level just like anyplayerelse, and tehe unfortunate guy who faced him in his first ranked game has lost many points for losing to an average rated kaparov. If this unfortunate guy was a good player, then he eventually gained points from other players and his real ranking was adjusted.

In this case

Octavious wrote: ↑

Sun Jan 21, 2018 10:11 am

the size of the contribution determined by their current rating

is equivalent to

Octavious wrote: ↑

Sun Jan 21, 2018 10:11 am

an expected score being based on the ranking of people you're up against

Both GR and EIDRAS are ELO-type systems. It seems to me the only difference would be in the weights?

Perhaps I didn't make myself clear. Consider the case in which you (and lets assume you're a high ranking player) go up against a player new to the site but high ranking elsewhere, and you lose. The other player gets a significant boost to their webDip ranking under both systems, but the difference is how it impacts you.

Under GR you lose the same amount of rating points as if you'd lost to VillageIdiot. How many you lose is determined solely on how highly ranked you are, so being a highly ranked player you lose quite a lot.

Under Yonder Rating the system calculates wrongly that you should win the game as it has insufficient information, and you lose significantly more ratings points because of this.

Have I misunderstood the systems?

in Ghost Rating, losses against expert opponents and beginners are not rated equally.

Yes, they are. This is because GR doesn't have a constant k-factor - the ratings of all players in this game is an input into the k-factor, which cancels out the earlier weightings. Have a play with some numbers in GR, you'll find it behaves very differently to typical Elo.

That seems crazy. Is that really the case? PPSC is dumb if it is.

Your interpretation is correct.

The info is always unperfect! Someone's "real ranking" is always under pursuit and it would be reached after many games. Say you're high rated, and you lose to an average rated thought experienced player, yes you'll lose more points than you "should", but then you will take these points back because you real ranking is higher, if you don't, it is because your real ranking was adjusted

Eirdra/jdr is better developed than gr, then.

I'd be lying if I claimed to have followed that reasoning :p.

At the end of the day I think that we need a few illustrative examples showing the differences between the systems in various scenarios. Then we can judge what aspects we find acceptable and what doesn't seem to work.

Octavious wrote: ↑

Sun Jan 21, 2018 11:13 am

The other player gets a significant boost to their webDip ranking under both systems, but the difference is how it impacts you.

Both of these are zero-sum systems, so he can't gain more than what you loose.