[2v2 Ladder S4] Stats Updated Sept 15th, 7:45pm GMT

[Ace Rimmer]

Good through game #115.

Check the thumbnail for points/ratings.

[Schubdüse]

Could anybody send me the thread/post in which is explained how the ratings are generated?

Thx

[Why?]

I don't remember exactly how it works, but Ace made a formula for it, based off of your W/L record and the ratings of the players you've faced. If you beat a player with a higher rating, you gain more. If you lose to a player with a higher rating, you lose less. Lose to a player with a lower rating, and you lose more. etc. etc.

[(MOR)]

Could anybody send me the thread/post in which is explained how the ratings are generated?

Thx

...
Now remember, real life skill can not really ever be measured mathematically. All that can be done is adjust a players rating to a more accurate rating with each new game (data point) figured in. Thus, the more games you play, the more accurate the rating. So, when comparing rus|Mike's 1581 to Nightwatch's 1580 you must remember that. rus|Mike is at 1581 with 21 games, while Nightwatch is at 1580 with 47 games (twice the # of games). What does that mean, think of projecting rus|Mike to 47 games and you get:

Code: Select all

Player      Rating  W-L-D
rus|Mike     1681   45-2-0
Nightwatch   1580   35-12-0

1681 derived by taking number of rating points (81) and dividing by number of games played (21). Then multiplying that (3.857142) times 26 future games (= 100.285 + 1581) =1681...

[Ace Rimmer]

Everybody starts off at 1500, then each new season you play in you start with your old seasons rating. (1500 if you never played before).

Then, I calculate the chances two players have vs each other, then take the actual outcome and compare it to the expected outcome and adjust each players rating accordingly. It's just as Why? said, if you are expected to win vs your opponent and lose, it hurts you more than if you are expected to lose against him. If you are expected to lose against your opponent (he has a higher rating) and you win, you get a boost. If you lose, it doesn't hurt as bad as losing against a weaker player.

Or,

David vs Goliath = David wins. Huge boost to David, Huge loss to Goliath as Goliath was the favorite.
David vs Goliath = Goliath win. Small boost to Goliath, small loss to David as he never had a chance anyway.
David vs Goliath = Draw. Small boost to David (outperformed, was expected to lose), small loss to Goliath (underperformed, expected to win).

For 2v2, I adjust your rating against each of the other players and then average it (I think, been a while) for a you vs "team" rating.

The Ladder is decide by points, and the rating only comes into play where two players have equal ladder points (tie decider). And, ratings are fun.

Here's the actual formula:

Code: Select all

=IF(AND(A2>0,EXACT(B2,F2)),AVERAGE(IF(G2=1,VLOOKUP(F2,'Raw Data'!AM:AO,3,FALSE),SUMPRODUCT(--($B1:B$2=F2),--($C1:C$2=G2-1),$D1:D$2))+25*(1-(1/(1+10^((IF(O2=1,VLOOKUP(N2,'Raw Data'!AM:AO,3,FALSE),SUMPRODUCT(--($B1:B$2=N2),--($C1:C$2=O2-1),$D1:D$2))-IF(G2=1,VLOOKUP(F2,'Raw Data'!AM:AO,3,FALSE),SUMPRODUCT(--($B1:B$2=F2),--($C1:C$2=G2-1),$D1:D$2)))/400)))),IF(G2=1,VLOOKUP(F2,'Raw Data'!AM:AO,3,FALSE),SUMPRODUCT(--($B1:B$2=F2),--($C1:C$2=G2-1),$D1:D$2))+25*(1-(1/(1+10^((IF(S2=1,VLOOKUP(R2,'Raw Data'!AM:AO,3,FALSE),SUMPRODUCT(--($B1:B$2=R2),--($C1:C$2=S2-1),$D1:D$2))-IF(G2=1,VLOOKUP(F2,'Raw Data'!AM:AO,3,FALSE),SUMPRODUCT(--($B1:B$2=F2),--($C1:C$2=G2-1),$D1:D$2)))/400))))),"")

Or,

PlayerA's Win % vs PlayerB=1/(1+10^((1500B-1500A)/400)) Take that chance of win and compare it to results:
PlayerAOldRating + 25 * (Result [1= win, 0.5= draw, 0= loss] - chance) = PlayerANewRating

This reminds me. I need to correct that 25 back to 10 for a slower rating system.

Edit: This also reminds me that I need to look into doing a team vs team rating and then adjusting individuals rating using that.

[Cobra Commander]

My head hurts

Guess I'm still at the foot of the table no matter what equations and system we use!

[Schubdüse]

Because (MOR) is easily piqued when he gets ignored, I would like to thank him and all the others for briefing me.

[Ace Rimmer]

Ok, I think I've figured out how to use InstaTeam™ Ratings instead of individual. I'll need to look at the numbers some more to make sure I haven't fooled myself.

How it did work for you:

You vs Enemy 1 = Rating 1
You vs Enemy 2 = Rating 2

Average Rating 1 and 2 and update your rating.

How it works for you now:

You + Partner = Old Team Rating 1
Enemy 1 + Enemy 2 = Old Team Rating 2

Old Team Rating 1 vs Old Team Rating 2 = New Team Rating 1 & 2

Your New Rating = NTR1-OTR1/2 + Your Old Rating

If that's too complicated, image!



Here's the difference between old and new ways of doing it:



If you're wondering why some have the same old and new but show a difference (like Vicious) it's because the ratings and the diff are really only shown as rounded.

Also, if you're wondering, the new system doesn't change you erratically (because it's avg of two vs avg of two split), which is why some were affected more than others.

Lastly, your personal gain is less as well as personal loss. In fact, the more evenly matched you are, the greater the loss/gain. The further away you are, the less you have to lose, or gain, at least ratings-wise.

Draws are an interesting thing. You might lose rating points, you might gain rating points, or the change might just be negligible. This is probably the same as before, but my new system is laid out different so I can see the evolution better.

In short, the new way moves you along at a much slower pace, which is more accurate, I think. If I ever get time, I'll go back and re-log what I lost and implement the new system for that season, update the initial ratings for Season 4, and get an even more accurate picture. 100+ games is a lot though.

Edit: The info above is only for the first 22 games.

[tllotpfkamvpe]

.

[Ace Rimmer]

I'm in the middle of logging the newer games (on 30 now), if you give me the matchups you have in mind and who wins, sure. I only need names and W/L. Scores don't matter.

Edit: I'm also looking at some way to determine if a match was balanced or not, but I'm not exactly sure what criteria to use. Comparing team ratings is one, but what's balanced? 0-10 point rating diff? 0-50?. Some matches have 100+ difference in ratings.

Edit2: Updated states, through #35:

[Ace Rimmer]

Update: I've gone back and implemented the current rating system (take individual ratings, sum, calculate team vs team, split results among team members) for Seasons 1, 2, and 4. I have not finished logging players and scores (too much to get all the data) from Season 2. I'm currently on game #201 and have ten pages of thread left (42 games). Once that's done, I can carry over Season 2 final ratings to Season 4 players and be able to post a better rating for everyone (Season 4).

In short, I went back to the beginning and started over with the new rating system. This way the ratings will be measured consistently across seasons. If I change it in the future, I'll do the same again (go back to Season 1, carry it forward).

[(MOR)]

Hummm, so that can up with something different ?

[Cobra Commander]

I AM NOT LAST NOW!

[Ace Rimmer]

Hummm, so that can up with something different ?

Don't know yet. Havent looked, not finished.

[Ace Rimmer]

Double post.

Ok, so I re-logged the players and scores for all 243 Season 2 games, went back to Season 1 and implemented the new sum/calculate/split method (that bert originally recommended), carried that forward to the current season (ignoring the 1v1) and here are the new totals:

The main difference isn't position (that's relatively unchanged), it's the rate at which you move along. I.e., the gaps are much smaller.