Pseudo Complete 1v1 Stats: Stat Site Evolving. 98% Complete

[Ace Rimmer]

btw, Shouldn't the betting system involve somehow the history of games between two players?

rus|Mike with SA has a 51% chance of beating (MOR) with NA.

It's just not true With our current 6:1 status and a very good setup (for me) it has to be much more than a mere 51%.

And besides when a normalized-win-ratio-correction Hammer will be finaly unleashed on (MOR)?

Well, it does take into account the territories, that is not just rus|Mike vs NA, and with (MOR) having all those games against WR, it's probably fair to say that some of those are SA vs NA.

[Xocrates]

btw, Shouldn't the betting system involve somehow the history of games between two players?

rus|Mike with SA has a 51% chance of beating (MOR) with NA.

It's just not true With our current 6:1 status and a very good setup (for me) it has to be much more than a mere 51%.

And besides when a normalized-win-ratio-correction Hammer will be finaly unleashed on (MOR)?

If by betting you mean the win ratios, then I would say maybe... Part of the idea was that the system was able to detect somewhat accurately the results of a match between two players. Using the player history should get a more accurate result, but it basically does so through "cheating". Half the fun of the Predict-o-mat is seeing how well it adapts to the reality. Additionally, keep in mind that from a statistical point of view most players have very little games, you and (MOR) have a grand total of 7 games. As far as we know, once you get a few dozen games the odds may (or may not) come close to reality.

Besides, it is not my fault that (MOR) has an inflated win ratio.

[Tripper]

I think the reason I haven't been paying that much attention is that Bert was ultimately correct - ranking is a better predictor than win ratio.

Looking at my wins (which are pretty few in number), almost all of them have been against people below me on the Bertometer and I consistently lose to people above me on the Bertometer.

(MOR) is an interesting case because he was pretty poor at 1v1 when he started (e.g I could beat him ) but then, as his skill improved, got into feuds with a few players (WR, Coby) who were generally worse than him and then got a good win ratio which didn't necessarily match with his ranking ...

or something like that

Cheers, Tripper

[Ace Rimmer]

Probably true, but then both are quite fun to play around with.

[Xocrates]

I should probably point out that as far as I know the predict-o-mat is about as precise as a game of darts on cheap bar. At this point it is simply a nifty toy that sort of works.

[Pox]

I should probably point out that as far as I know the predict-o-mat is about as precise as a game of darts on cheap bar. At this point it is simply a nifty toy that sort of works.

Indeed, it all depends on each player having a fairly even distribution of opponents.. which simply doesn't happen the Bertometer is certainly still the best device for this kind of thing, though mixing the CWR in there helps.

Also: I'll fix the predictor now. Thanks for the typos, Xocrates.

Pox, I don't think that Void games should be counted while calculating of the shortest and longest games. There's one game of me vs Ace on the main page and this game is actually Void

Fix'd.

Xocrates: See my post in the betting thread about the oracle.

[Xocrates]

Xocrates: See my post in the betting thread about the oracle.

See my reply in the same thread.

EDIT:

Try replacing DW in the above formulas by:

9.75E-4^(1/(games played versus player+10E-12))*(HWR12-HWR21)+(1-9.75E-4^(1/(games played versus player+10E-12)))*DW

with HWR12 win ratio of player 1 vs player 2 and with HWR21 win ratio of player 2 vs player 1

This is so far untested, but should work.

9.75E-4 is so that 10 games mean the historic win ration will have a weight of 50%

[Xocrates]

Hey Pox! It seems the predict-o-mat is broken! Try making WR VS Ace on Asia Vs Russia and see what happens, that is not correct according to my calculations

[Xocrates]

Triple post! Fear me!!!

Anyway, I've devised the ULTIMATE Predict-o-Mat (or UPoM), you can get the excel sheet here.

It goes beyond the usual Player win rate and Continent win rate prediction, to add the above mentioned players history, as well as the detailed player setup history, players setup history

Obviously adding all that to excel is a pain (most are completely optional though, so you can still have your own basic Predict-O-Mat)

Expect an Epic Treatise soon describing the whole system, unless of course Pox thinks he can take it all out of Excel and into the stat page.

[Ace Rimmer]

/me grabs copy

[Pox]

Hey Pox! It seems the predict-o-mat is broken! Try making WR VS Ace on Asia Vs Russia and see what happens, that is not correct according to my calculations

Sorry, I had replaced (dw ^ 2) ^ .5 with dw when it should be abs(dw)... damn you, magic-absoluting-sqrt-symbol.

EDIT: The new HWR formula you stuck up is now working, and if you feel like giving me a procedural/mathematic description of the spreadsheet I can put it in.

If it helps, or you can spot a bug, here's the code, where $p is the current player number, $wr is an array of their win ratios, $cwr is an array of the respective continents' win ratios, $hwr is an array of their win ratios _against each other_, $players is an array of their names and $territories is an array of the continents names. Basically, $something[$p] is the current player's value, $something[!$p] is the other player's value.

Code: Select all

    for ($p = 0; $p <= 1; $p++) {
        $dw = pow(.000975,(1/($p_played + pow(10,-11)))) * ($hwr[$p] - $hwr[!$p]) + (1-pow(.000975,(1/($p_played + pow(10,-11))))) * ($wr[$p] - $wr[!$p]);
        $cwr = $territory_wr[$p];
        $f2 = (1 - $dw / (abs($dw) + pow(10,-11)))/2;
        $y = $cwr * $dw * $f2;
        $f1 = (1 + $dw / (abs($dw) + pow(10,-11)))/2;
        $x = (1 - $cwr) * $dw * $f1;
        $w = round($cwr + $x + $y,2) * 100;
        if ($w >= 50)
            echo "<h2>${players[$p]} with ${territories[$p]} has a $w% chance of beating ${players[!$p]} with ${territories[!$p]}.</h2>";
    }

[rus|Mike]

Ok I'll go enter Oct 30-31 for Tripper or else it's gonna take forever

Although I think that nothing from Oct 22-31 is entered

[rus|Mike]

Ok, 30-31 entered... Ace, do a double-check on those [22-29].

EDIT:
IK logged 29th also... somebody, do the rest

[Xocrates]

I present you, UPoM in formulas:



Basic

W=CCWR+X+Y

X=(1-CCWR)*CDW*F1

F1=(1+CDW/(abs(CDW)+10E-12))/2

Y=CCWR*CDW*F2

F2=(1-CDW/(abs(CDW)+10E-12))/2


Considering Player history

CDW=DW*(1-9.57E-4^(1/(W12+W21+10E-12)))+(HDW)*9.57E-4^(1/(W12+W21+10E-12)

HDW=(HWR12-HWR21)*(1-9.57E-4^(1/(CW12+CW21+10E-12)))

HWR12=W12/(W12+W21+10E-12)

HWR21=W21/(W12+W21+10E-12)

W12 - Times player 1 won player 2

W21 - Times player 2 won player 1


Considering Player Setup History

CCWR=CWR*(1-9.57E-4^(1/(CP1+CP2+10E-12)))+PCWR*9.57E-4^(1/(CP1+CP2+10E-12)

PCWR=0.5(1+DCWR)


DCWR=(CP1CWR-CP2CWR)*(1-9.57E-4^(1/(GOS12+10E-12)))+HDCWR*9.57E-4^(1/(GOS12+10E-12)

CP1CWR=P1CWR*(1-1/(GOS1+10E-12)^(1.2+GOS1/100))

CP2CWR=P2CWR*(1-1/(GOS2+10E-12)^(1.2+GOS2/100))

P1CWR=WOS1/(GOS1+10E-12)

P2CWR=WOS2/(GOS2+10E-12)


WOS1 - Wins of Player 1 in the same continent setup and playing with his assigned continent

GOS1 - Games of Player 1 in the same continent setup and playing with his assigned continent

WOS2 - Wins of Player 2 in the same continent setup and playing with his assigned continent

GOS2 - Games of Player 2 in the same continent setup and playing with his assigned continent


Considering Players joint history on setup.

HDCWR = HP1CWR-HP1CWR

HP1CWR=CW12/(CW12+CW21+10E-12)

HP2CWR=CW21/(CW12+CW21+10E-12)



CW12 - Times player 1 won player 2 in the same continent setup and playing with his assigned continent

CW21 - Times player 2 won player 1 in the same continent setup and playing with his assigned continent

[rus|Mike]

Would've been nice to see not only this list of percents but also overall percentage. Like "rus|Mike has a 60% chance of beating Ace Rimmer".