I had exactly the same reflexion than what Guaroz just wrote.

Oli, do you really think that "Most people do not care about rating." ?
I do think that MOST people care A LOT about their rating !
By the way, I would like to point out that this thread about rating has already 661 answers ! Not bad for something that most people wouldn't care about !

Im not so sure Ret. The people who reply to this forum number about 50 (thus Oli may be rght(and, to be honest the people who have the most action in various games are the people who are in the top 100 of the HOF. What I mean is that even if someone abuses the system it wont lat long nor it will amount to anything.. I may be mistakenof course

@Oli,

I care about ALOT about my ratings. How else am I supposed to know if I'm improving or not over long periods of times?

Now, I think many people share my sentiments because let's face it, People like to think they have achieved something. Even getting credit for the most mundane things can make some considerably happier.

I don't think I've ever seen or heard anyone say they were unhappy for winning a contest or getting noticed for something they've done (assuming, of course, that the something in question is a good something).

Oli,

I'm looking through the code again. Can you explain the difference between SCc and SCq?

It's not apparent to me what SCq represents.

Thanks!

SCc = How many SC's the player has (exaple: 18 of 34)
SCQ= % of the SCs (example 18 of 34 = 50%), but if you have more than needed for winning it's sanitized (20 of 34 is still 50%).

Thanks,

I don't understand this line, then however:

$mV = abs($SCq1 - $SCq2);

In light of the fact that the Rr equations already take into account the proportion of SC's between the two players.

$Rr1 = ( ($SCc1 + $SCc2) > 0 ) ? ($SCc1 / ($SCc1 + $SCc2)) : 0;
$Rr2 = ( ($SCc1 + $SCc2) > 0 ) ? ($SCc2 / ($SCc1 + $SCc2)) : 0;

If the relative SC strength between the two players is already reflected in the Real Result, why is the result difference D = round(($Rr1 - $Re1) then multiplied by a factor that is the again basically the relative SC strength between the two players?

You need to check the 15 lines bevore...
SCc1 and SCc2 and SCQ1 and SCq2 are changed based on Pot-distibution, or drawn/not drawn the point you mention was exactly the problem I fixed 1 week ago.

Rr is only 0 or 100 since the last update.

Currently the messaging rule setting is irrelevant for rating changes.
How could we take into account the messaging rule setting, which basically has been neglected during the entire discussion?

The incidence of luck is greater in a gunboat with respect to a full press game; for example during the opening decisions and due to orders mismatch. I think that this is a sharable opinion.

All other things being equal, playing a gunboat instead of a full-press-game:
a) when things go well it is easier to win.
b) on the other hand, when things go bad, it is harder to avoid elimination.
Generally speaking, the full-press victory is more “earned” with respect to the gunboat victory, while the full-press defeat is more “deserved” with respect to the gunboat defeat.

Because of this comparison, I would like to propose to implement a “messaging rule coefficient”, whose purpose is to reduce the number of V-points moved by a gunboat with respect to the number of V-points moved by a full-press-game. Namely, compared to an equivalent full-press-game, winning a gunboat would award less while losing a gunboat would cost less.
This coefficient could be made a fixed % multiplier of the game-value coefficient gV.

What are your thoughts about this?

Good point, Decima Legio !

Yes, the messaging rule setting has been neglected during (almost) the entire discussion. Indeed, on March 8, I wrote in this thread :

"By the way, I would like to say that it is my opinion that any ranking system will be flawed because they compare things that have nothing to do together, even though they are all called "Diplomacy games". Indeed, in my opinion, Diplomacy is a communication game. So comparing in the same ranking system a gunboat game with a full press game is, in my opinion, TOTALLY RIDICULOUS. Same thing with a 1vs1 game : where is the diplomacy in such a game ?"

By that time nobody cared enough to answer that message.

Thank you for bringing that again in this conversation, Decima Legio :)

I think the different press-types require a different skillset, but I do not think they require more or less effort. Gunboat is very much about messaging with your neighbors and it's really hard for a beginner to understand all the signals.

In the GhostRating they use 1 for normal games, 0.5 for publicPress and 0.25 for gunboat, but this is very arbitrary to, as I think gunboat requires more skill than publicpressOnly.

At the moment I don't think we should give different rates depending on pressType, but I will make separate stats for each pressType (and variant and pot and all iterations) once I have the time.

But as always I'm open for suggestions.

Thank you again for your great work Oli :)

Wow, I didn’t know that they used such harsh coefficients towards gunboat and public-press in GhostRating.

Actually I had in mind something more soft, something around 1 for full-press, 0.8 for public-press and 0.6 for gunboat.

This should be of course arbitrary… but not more arbitrary than using 1 for full-press, 1 for public-press and 1 for gunboat is.
There’re a lot of things arbitrary in a mathematical model, we just have to choose those that we think fits best reality.

I’m happy that separate stats for each pressType are in the to-do list, but since we’re talking about an overall rating now, because we have the legitimate and practical need to summarize ALL the user informations in a single RS number, I’d rather see a gunboat transfering less points than an equivalent full-press game does. Not because of a matter of skill or effort required (I agree with Oli on this), but because of the greater incidence of luck in gunboats. Usually you’re not completely the owner of your destiny in a gunboat, while indeed you are in a full-press game.

This said, since Oli disagrees, I suppose I need some people supporting this kind of adjustment. Else, if there’re not, fair enough, I’ll just withdraw that.

Any more opinions?

I agree with what Decima said regarding the luck involved in gunboats.Sometimes there is simply nothing you can do.

On the webdip GR, I believe Wilson and Gunboat are both weighted at half.

I do not think Gunboat games or Public Press games should be weighted any different that regular games.

No. I have the code for the GR and it's 1/2 and 1/4

Perhaps you're right. At this stage, the webdip GR on an overall basis is mostly meaningless as the various categories are far more telling.

Also GR do weight the variants differently.
Classic is 100%, AncMed is 50% and World is 25%.

What's the reasoning behind Ancmed weighed at 50 percent, and world at 25 percent?

I have no clue...

why do i get minus points for drawing my last two games??

TheGhostMaker would be so proud that his ranking system was being adopted by vdip

@GOD: We already told you at the modforum that there is still an issue with the automatic update of the ranking-calculation.
I've updated the ratings and now everything is fine. Will need to monitor this, as I still have no clue why this is happening.

"TheGhostMaker would be so proud that his ranking system was being adopted by vdip"
its more 2 player ELO then ghost rating. But yeah, it is far better then the points system used at webdip and still used here at vdip.

Oh ok :)

Ok I admit I've been slack the past few months and have only been playing two games and not actively checking the forums. Then I stumbled on this thread, checked the new HoF and got blown away that i've been dropped from 2nd or 3rd down to something stupid like 41st. WTF?

How can this be fair?

Yes I should've been actively checking threads but I've had other stuff in my life. Ditto if someone can point me to a concise rationale for why this is a better system (sorry don't have the time to read 23 pages.)

Please excuse my bitching but I'm flabbergasted by this.

Amby,

Quick answer if you want to look for yourself:
try the wiki to see the current calculation (not completely up to date as I detail below):
http://www.vdiplomacy.com/wiki/index.php?title=Rating

and if you wish, take a look a the code directly (should be up to date if I remember correctly):
https://github.com/Sleepcap/vDiplomacy/blob/master/lib/rating.php


Anyways, I've been wondering these things too, for you and several other individuals.


So... I took a look at your games and immediately found one thing odd from the Bourse game (http://www.vdiplomacy.com/hof.php?gameID=9186).

At first glance:

When I look at the calculation for that game, the score calculation for each player should include a calculation against every other player, but closer inspection shows that you only scored against those of us who were defeated, while those of us who were defeated only lost points to those of you who drew (and Canaduh, who CD'd).

As I understand, the calculation should be made for every player against every other player.


So I took a closer look:

This game is PPSC not WTA, and one player CD'd (Canaduh) was replaced (by bozo).

In this particular game, bozo took over basically from the beginning and was defeated along with 8 others of us. That left 6 players drawing the game at the end and the 1 CD.

This variant has 109 SC's, 55 of which are needed for a win, which is 1/2 + 1, so this game had no arbitrary victory conditions.

Rr (Real result)
mV (match value)

Rr for draw vs draw is unknown as draw vs draw doesn't appear in the calculations for this game, but it should be 0.5 : 0.5.

Rr for draw vs defeat should be 1 : 0 (which it is)

Rr for defeat vs defeat should be 0.5 : 0.5, but this calculation is absent too.

Rr for defeat vs CD/Resign should be 1: 0 (which it is)

These values should take care of any performance relationships in the game between these players and mV should be left out of this calculation.

Instead, mV for draw vs defeat looks to be 17% (or 1/6th of the draw), but this sort of comparison should only be in effect between drawing players and not drawing vs defeated players. Each drawing player should get 1/6th of the credit for the game (adjusted by their expected performance), but instead the match value is scaled to 17% of the score adjustment because of the size of the draw which makes absolutely no sense. (This is the issue with trying to use mV to account for proportional differences in SC's instead of Rr where this relation should be).

From the wiki:
Adjustments for each match (matchValue=mV):
- Value the importance of take-overs. (If a player bet only the half the whole match is worth only half.)
- Normalize the mV by the SC% difference of both players

The first adjustment makes sense so that picking up a CD doesn't allow players to quickly inflate their ranking. This explains why drawing against a defeated bozo who didn't bet as much as the rest of us is adjusted differently than for the rest of us (or for canaduh who CD'd).

The second reason makes no sense whatsoever because the calculation for Rr1 and Rr2 should take into account the proportion of SC's between the players in a PPSC game (as shown in the wiki, but not implemented in the code). mV should not be adjusted further to make this comparison again. (Also Oli as I understand, you have made it so that Rr always either 1 or 0 for all games. This makes sense for WTA or maybe even solo vs survive in PPSC, but not survive vs survive which is where any effects from different SC counts at game end should be made, not in mV.)


So I took a look at another known world game to see what happens with a solo'd PPSC known world game:

http://www.vdiplomacy.com/hof.php?gameID=4761

cypeg got at best 51% mV against defeated players (he solo'd) and got ~32% mV against those of us who he beat but had sizable survives (19 and 20 SCs), and more against others who only had 7SC's.

mV is being used here to make the comparison between SC's rather the Rr which results in scaling of the resulting score calculation and confuses and ruins the intended effect.

If cypeg's Re (expected result) against me was 65% (which appears to be calculated correctly), and his actual result was 100% (he beat me), then he shouldn't get only 32% of the rating adjustment simply because I got 20SC's. cypeg won, and by a lot. (I'm not exactly sure where the value 32% came from because mV still isn't completely explained on the wiki and I couldn't follow it in the code).

Additionally, against players cypeg defeated, he is getting only 51% of his ratings adjustment because he didn't capture every last one of the 109 SC's. This is wrong, he should get 100% of the ratings adjustment for wining against defeated players.

Further.. comparing Wolfman and I:
Wolfman's expected result (Re) against me was 28% while mine against him was 72%
However my real result (Rr) against Wolfman was 100% (I had 20 SC's) while his against me was 0% (he had 19 SC). Our real results should have been pretty darn close to 50%:50% against each other. Instead Wolfman is getting a complete fail against me for getting 1 less SC than I had. Turns out in the actual calculation, this doesn't make much difference because for each of us the final ratings adjustment is scaled down to 1% because we were so close in SC's. Wolfman's real result actually should have beat his expected result against me ~50% to 28% but instead he lost (a thankfully for him scaled to 1%) ratings adjustment of an expected 28% to a real result of 0%. While the numbers come out not looking horribly skewed, the calculation being done is completely wrong and doesn't have the intended effect. Wolfman should have gained in the ratings adjustment while I went down.


Am I making sense to anyone else in identifying what is wrong here?

I don't really understand all the calculations that go into this either. However, I do admire all the work and thought going into it.
I only think this will start a trend of players only playing against certain other players (because, they are worth more?). I also don't really understand how some games give such low pay-outs and others give huge ones. For example, I won a 35 player variant game and got only about 60 HoF pts, meanwhile cypeg who is listed as #2 got over 300 HoF pts for also winning a 35 player variant. I also think its a little silly that this game makes all the difference between no. 2 on the list, and well below the top 100! What sense does this make? Albeit, cypeg's a great and deserving player and I am only using this as an example but I am confused by the exhorbitant value of this game, meanwhile really tough and rewarding draws only pay out a pittance.

@Amby- all I can say is... who cares? We're rich, byich!!!