If this is the case, then I would probably switch to only playing WTA games when I care about my ranking, as a solo in PPSC means little to nothing over a draw or even a survive. I can be the strong second and do better than the player soloing in this current system.

Fundamentally:
A draw, even in PPSC is a failure by one player to solo and a victory by the other players who stopped the solo.

A strong second in WTA is a failure by the surviving player to either solo or force a draw, whereas in PPSC, a strong second survive is counted as a decent showing (debatable whether it should ever be counted more than a draw, but imo, in no circumstances should it ever be counted as worth more than another player's solo).

In no case should a player who solos get negative points, and in the case where cypeg solo'd and rancher had a strong survive in the game cypeg mentioned, it should never have happened where a top 150 player lost to a top 25 player and yet had a better showing in ranking increase than the player who soloed. That is fundamentally incorrect, and I've identified exactly why in the formulas. Just by playing against cypeg and doing halfway decently, rancher couldn't lose in ranking.

If the heuristics isn't enough, let's take a look at the statistics of the game you mentioned:
http://www.vdiplomacy.com/hof.php?gameID=398

Bozo failed to stop allanxo's win, and only lost -0.16 for doing so.. That is fundamentally wrong while allanxo only gained 0.16.

With my change (all other numbers vs survives for bozo remain the same):
bozo would lose: -1.43 to allanxo's solo instead of -0.16 (Rr = 0% instead of 40%), and still gain +12 instead of +14.

The difference is in how allanxo's score changes: (Rr's of 100% instead of 60% and 82% respectively)
against Bozo he gets 1.43 instead of 0.16
against OlympicTorch he gets 3.43 of 2.13
Final score difference for him is +22 instead of +20

OlympicTorch remains at +1

For allanxo:
+22 of 35 that changed hand this game toward winners/survivors is 63%
+20 of 35 is 57%
+28 -points of the 52 that changed hands toward winners/survivors is 66%

For bozo:
+12 of 35 is 34% under my system
+14 of 35 is 40% under your system
+15 -points of 42 is 36% under the system in place when played

For OlympicTorch:
+1 of 35 is 3%
-1 -points of 42 is -2%


So in reality my calculation of Rr being 100% for solos regardless of SC's is actually closer to the original results based on the scoring system in place when the game was played.




So, unless enough other players lobby to change the current system and/or reset scores back to 1000 (if absolutely necessary) to account for the 'new correct' scoring method so that solos for PPSC are given proper value... I'd like to propose at least that my 'fix' for PPSC solos (Rr1 = 1 for a solo in all circumstances and Rr2 = 0) at least be implemented (at some point) as an alternate ranking system, possibly including my recommendations about adjusting the scoring to avoid a survivor from having a score increase greater than a draw:
n = the number of players with more than 1 SC at game's end:
Draw = 100/n%
Win = 100*(1-1/2*n)%
Survive = 100/(2*n)%

Sorry to keep harping on this but as an engineer, I can't let math errors slide by this easily.

+1 Leif

+22 of 35 that changed hand this game toward winners/survivors is 63%

should read:
+22 of 35 v-points that changed hand this game toward winners/survivors is 63%

I think you found an interesting flaw.
Because I've set the Rr to the SC-difference and watered this down with the SC-%-difference in the mV again I devalued the points for the better survivors 2 times. Now if we score every game (not just winner vs other survivor) with 100:0 and adjust the game-value only by the SC-difference the end result looks much better and it's still very close to the DPoint rating.

cool - so any estimates on when this goes live oli?

(also I strongly support minimum participation to be a criteria for the HoF - 6 phases played in the last 6 months or some such)

The code does already work if you look at the games, just the history is based on the old calculations (and I will recreate this soon)

But now that I do some tweak again I might change the kValue too (or change the starting value), so you win or loose more points. As I mentioned in the Elo-calculation a game against similar rated players the looser does lose about 16 , in these games they only loose about 5-10. And in the game I mentioned the points won/lost looks similar, but in reality one game is based on a base of 100 , the other on 1000 .

Im lost in translation but im happy that you agree on something :)

@cypeg: Not technically accurate, but easier to understand:
For survivors in a PPSC-game I applied the SC-difference ^2, screwing the top-players off their points the better they played.
Now it's applied only one time and the results look much better.

Better? Now in gameID=1623 Iran has been rewarded with 46 V for facing poor players while Texas got a -2 for facing good ones. Isn't it just a matter of luck? Those who care about ratings would be better start join "pick your country" games only, wisely choosing their REAL opponents.

Thanks Oli,

Sorry to have kept focusing on this but there was definitely something fishy going on and it seems you've seen it now too. I'd still like to know exactly how mV is calculated. (took a look at your code github just now, but it looks like the old code where both mV and Rr adjust the result based on SC count.)

Could you post here when github is updated and I'll check how you are calculating mV and run a quick double check on several of the games that have been noted as issues and see if things look correct from a math standpoint?

Also, on the http://vdiplomacy.net/hof.php?gameID=???? pages, it was helpful to have all the intermediate values of: Re, Rr, Dif, mV, gV, and Ch. These helped me be able to compare exactly where I was finding differences in your calculations and mine. I notice that on a per line basis, Re, Rr, and gV are now gone, would you mind adding them back?

Another question for another day, is what the reasoning behind the particular curve chosen for calculation of gV is, but I won't bring this up just yet.

Thanks again for the great work!

When you do get around to doing the minimum participation requirement, I think Rancher's suggestion would be fairly decent(6 in the last 6 months) but I think 10 in the last 6 months would be better...

@Guaroz: If you take a look at the board you will see that Texas was a CD-takeover. He had bet double the points as anybody else. That means he had twice a good start (and half the work). In the DPoints result he has lost 3 too...

Pushed the new code to github.

hmm the improved improved HOF has changed the ranking of players for some no so much but others have moved places . dramatically

I upped the kValue. This means fewer games are required to reach your "final" score. Once you hit your score it will be hard to improve. Now some very good players without much defeats have a higher score. But they would have reached this score without the changes a few games/weeks later.

“
// 2-player variants are always WTA.
if (count($Game->Variant->countries) == 2) $Game->potType ='Winner-takes-all';
“
very good correction Oli, that makes sense now for duels.

As per my cryptic post dated Wed 13 Mar, I will try to explain better:

“
// The more people the more important a game...
$gV = $K * pow(((count($Game->Variant->countries) -1) / count($Game->Variant->countries)),3) * (100- (count($Game->Variant->countries)))/100;
“

Using the current gV formula the intent is not matched:
the gV value grows from duels up to the Known World 901 variant, while gV starts decreasing from the World Diplomacy IX variant onward, up to WWIV.

Currently World Diplomacy IX is the most “important” variant, while of course this place belongs to the most populated variant, WWIV.


This error is due to the presence of the factor
* (100- (count($Game->Variant->countries)))/100

For example, this would be correct:
// The more people the more important a game...
$gV = $K * pow(((count($Game->Variant->countries) -1) / count($Game->Variant->countries)),3);

@DL: You are right.

The thought behind both parts of the equation is this (and I need a formula that integrates both ideas):

1st part: make the small variants count less significantly (but make this less important the more players are in the variant: pow(((count($Game->Variant->countries) -1) / count($Game->Variant->countries)),3) => pow(X,3) means ^3.

2nd part. gradually decrease the importance of the larger games. As in larger games you tend not to interact with all players, but you get rated against all players.

It's obvious that if you degrade small games and larger games there needs to be a "turning-point". But where this needs to be and how big the difference should be is still up for discussion.

You solution just throws away part2. Can you come up with a formula that's better suited and integrates both ideas?

Well, it was not intended, but I think that both the ideas are already covered with my simple suggestion:
I see the value behind your 2nd idea, but IMO its effect is already included in the function I proposed to keep.

Do you recall a couple of months ago, when the discussion started? At that time the first draft game weight was gV = 15*(n-1).
This led to your legitimate doubt of integrating a further 2nd part coefficient. At that time, when the 1st part was linear, adding that 2nd part coefficient made sense. Hence, together with that 2nd part integration, the gV was turned from a straight line to a curve, but the statement “The more people the more important a game” was still respected.

Then, I don’t recall when, the 1st part draft linear shaped function was substituted with a curve ((n-1)/n) together with some sort of exponent. That curve IMO merges the 2 ideas… ALONE:
1) there is no turning-point, else, it’s hard to understand that a variant sized 20 weights more than a variant sized 30.
2) the more the players the less the gV increase (but still an increase); this is exactly the effect of the 2nd part idea.

I think that the function gV = ((n-1)/n)^3 is a fairly good shaped one in order to take into account the game weight given due to the variant size. Maybe I would argue that the exponent 3 looks high, while an exponent 1 would be too low. Just my opinion on the exponent.

@Oli. Thanks for your reply and yes, I didn't notice Texas was taken over. Anyway, Texas was just an example. My point was that in North & Central America there was a concentration of good players that had to face each other, hence scoring poorly at the end of the game, while Iran had basically no worthy opponents. If I'm not wrong, the first goal of this RS was to reward players depending on strenght of opponent they had to face. Well, I believe this is not what's happening.
Again, my concern is not a flawed system (I don't need a HoF to know that bozo gantz & Devonian are excellent players and had great results). My concern is how a flawed system could affect the way of playing of several people more interested in HoFs (maybe lacking of self-confidence?) than in the game itself.
I already said my concerns on PPSC/WTA issue and, as far as I can understand, you're managing it wisely.
Now I'm here to say my concerns about "if and how" this system considers the players you ACTUALLY faced. This bring several issues.
- Players excluded just for their rating. The first thing I thought was "Hey, if you don't want to drop down in the standings, you have to play only passworded games and invite only well-rated players! Maybe you'll lose some more games, but defeats will cost you less and wins/draws will reward you much more!". Actually, someone had an idea "better" than mine and asked for a game-creation feature in order to exclude low-rated players from their games. Omg, not even the effort to open a thread and recruit good players for a Private game!
Exclusions because of the V-rating wouldn't only furtherly flaw the system (I don't care, but I think you do), they would kill the fun.
- "Pick up your own country" feature furtherly abused. IIRC, this feature was created for Tournaments, SRGs and 1v1 switched-sides twin-games. Using it for normal games is already an abuse, IMHO, because your neighbours should be always random. Since you can choose weak neighbours (not all, but many), expecially in large maps this will be furtherly abused.
- I'm rather convinced that this system could encourage someone, under certain circumstances, to attack a well-rated neighbour rather than another. This was already expressed by someone else, I know you know, so I won't get deeper into this. I want only add that... I don't like being a well-rated player (damn, I'm #13)! Could you please cut my name out the HoF? :-)
Otherwise, could we have an option that allows only top x players to join my games? So I wouldn't have to worry about player #80 being my neighbour, being HoF-addicted and wanting my skin? :-))

Those were 3 issues coming in my mind in the last 10 minutes. I hope there are no more, but I couldn't swear on it.
Thanks for your patience.

I do'nt think the possibility of abusing the system is not that big. Most people do not care about rating. I didn't post in this thread, and the HoF worked this way for 4 weeks, but no one found these flaws.
Just 1 post later everyone checks their stats and the discussion continuous. That's a clear sign that no one looked at the HoF. It's just interesting now as we search for solutions to make this more fair.

I did some changes to the gV-formula.
X=player-size of the game

It was: ((A1-1)/A1)^3 * (100-X)/100
It's now: =((X-1)/X)^(2,5*(100-X)/100)*0.8

Results: http://www.vdiplomacy.com/gV.png

@Guaroz - I'm HoF rated #77 and I'm always out to get you :P

just kidding :)

Fair enough Oli, but:


- 0.8 in front of the function is unnecessary, if you use it, you’d better change again the K coefficient from the current 50 to 40, which is 50*0.8.

- the contribution of *(100-X)/100) on the exponent is de facto rather irrelevant, you may decide to cut it off in order to make the gV formula look simpler without altering the shape and the values significantly.

@DL: That's the way I did it in the code.. :-)
And if you check the formula without the (100-X)/100), you will see that the 2-player games are worth the double and I do not want this.

Oli, I have reproduced your graph in http://www.vdiplomacy.com/gV.png

New graph here: http://imgur.com/Fqjkb6G

The Y axis value “1” corresponds to the K-factor dimension. Just a scale factor, it doesn’t change the shapes nor the relative values.

I have used the same legend as before, adding the purple dashed line for comparison, which is actually the New-score formula without the (100-X)/100) term:
gV=((X-1)/X)^(2,5)*0.8

The difference between the green and the purple line is around 3% all the way through, that’s what I was meaning saying “without altering the shape and the values significantly”

If you say “the 2-player games are worth the double “ there must be some sort of misunderstanding either on my part either on yours.

@DL: You are right. Looks like I entered a really strange formula in Excel.
Now I've settled with a ^2,7 and it looks alright.