:: Community ::
ForumsMessagesGroupsChat (1)Friends
 

Forums :: iPhone :: Strategy :: Official Stat Warriors Leader Board

You must sign in to post.

Page 2 / 22 1   2   3   4   5   6   7   8   9   10   11   12   13   14   15   16   17   18   19   20   21   22 

Re: Official Stat Warriors Leader Board :: Mar 5, 2011 @ 9:02pm

Cabin Boy esparano

Joined: Nov 5, 2009
Posts: 2308

Good suggestions ionizedfire. I noticed ta a lot of scrap accounts showed in the ranks.. Jazz, add this multiplier to the final formula if you want to include people with a certain number of total games. For example, an account that only has 100 games played will be extremely low in the leaderboards, but anyone above 300 games played will have their regular score. 
 (1/(1+e^-(g-300)))   <--- e is the number, not a variable :)
Where G = number of games played, and 300 is the number of minimum games (change it if you like). If g> 310, then the expression will evaluate to almost exactly 1, which doesn't affect the score of te player. However, if a player plays 3 games, their score will be very close to 0, or negligible. They won't be counted in the ranks.

294737 points * 1/(1+e^-(100 games - 300)) = 0.

 294737 points * 1/(1+e^-(1000 games - 300)) = 294737. 

What do you guys think? Should there be a minimum number of games played to show up in the ranks? This solves the problem of requiring people to play a lot of opponents, but doesn't benefit the rank whores like the last equation.
post updated on Mar 5, 2011 @ 9:20pm
Re: Official Stat Warriors Leader Board :: Mar 5, 2011 @ 9:11pm

Cabin Boy ionizedfire

Joined: May 25, 2009
Posts: 177

Ooo whippin out Euler's constant. I like it.  But, it's (1/(1+e^(300-g))) ;)  You flipped it around :)
Re: Official Stat Warriors Leader Board :: Mar 5, 2011 @ 9:14pm

Cabin Boy esparano

Joined: Nov 5, 2009
Posts: 2308

Well the actual function has e^-x So I'd rather have -(x-300) than (300-x). They're the same, but I prefer the first one haha. The first one let's you think of the translation easier I think.


Also, the reason I think this would help is that it gets rid of the need to include # of games played completely. Peoples scores will be unaffected by te number of games as long as Rey reach the minimum number of games.
post updated on Mar 5, 2011 @ 9:18pm
Re: Official Stat Warriors Leader Board :: Mar 5, 2011 @ 9:17pm

Cabin Boy ionizedfire

Joined: May 25, 2009
Posts: 177

Oh I see. Well you had (1/(1+e^(g-300))) as your function.  So if you want it that way, you need (1/(1+e^-(g-300)))

Psst go to chat esparano
post updated on Mar 5, 2011 @ 9:18pm
Re: Official Stat Warriors Leader Board :: Mar 5, 2011 @ 9:18pm

Cabin Boy esparano

Joined: Nov 5, 2009
Posts: 2308

$@%# oops. I had it right in the calculator but messed up copying it haha
post updated on Mar 5, 2011 @ 9:19pm
Re: Official Stat Warriors Leader Board :: Mar 5, 2011 @ 9:46pm

Cabin Boy ionizedfire

Joined: May 25, 2009
Posts: 177

Ahh screw my idea.  Jazz. Remove Mastery skill, and use esparano's idea. My idea only helps short term rank whoring, but in long term, rank whoring doesn't matter.  Too many games played will compensate for rank whoring, since your stats compile all games over all time.  

Use esparano's limiting minimum game thing so it removes rank whoring from the equation, and remove mastery % all together.  

That's my suggestion!
Re: Official Stat Warriors Leader Board :: Mar 5, 2011 @ 11:19pm

Cabin Boy tinny

Joined: Jan 2, 2008
Posts: 876

what exactly is wrong with phil's ranking system and how is this new system different?

with phil's, the best players are at the top and the worst at the bottom. thats the way a leader board should work.
Re: Official Stat Warriors Leader Board :: Mar 5, 2011 @ 11:29pm

Commander jazz

Joined: Feb 20, 2009
Posts: 534
Location: aka The_Musician

Tinny - how exactly would you adjust Phils ranking system when games are held for more then 10 days?

What weight would you give?  for what kind of duration?  

See the problem?
Re: Official Stat Warriors Leader Board :: Mar 6, 2011 @ 3:12am

Cabin Boy periwink

Joined: Mar 31, 2009
Posts: 412

tinny, Phil's ranking system doesn't compare players together; and not all players are at the top
Re: Official Stat Warriors Leader Board :: Mar 6, 2011 @ 2:05pm

Commander jazz

Joined: Feb 20, 2009
Posts: 534
Location: aka The_Musician

update:

I've got the game dump import process fixed finally.  Instead of importing all the data in at once, and then running a process to crunch all the numbers, I now have it importing the game dump 1 game at a time and processing the numbers for each game as it gets imported.  It's a slower importing process, but the final "Build LeaderBoard" stage is now down to 15 minutes per game type.  

My main worry now, is whether the slower import process will take longer to import the games then the rate at which games are being played.  ;)

Next hurdle...  play around with the formulas a bit and see what I can do to balance it all out.
Re: Official Stat Warriors Leader Board :: Mar 6, 2011 @ 11:12pm

Commander jazz

Joined: Feb 20, 2009
Posts: 534
Location: aka The_Musician

Coolio!  Got it all workin again!  =)

Thanks to Esparano's awesome help and input on the math side of things, I've updated the Mastery formula to be:

100 * (OpponentCount * GameCount) / TotalGameCountAtThisRank

So now, the more Opponents you play against, and/or the more Games you play against them...  the higher the percentage goes.

I think of this number is the "accuracy factor".  Based on this value, the system can have a measure of how accurate the Win Percent you achieved actually is.

The outcome of this change is a very different looking Leader Board.  The Score numbers for each player look more like a Bell Curve then the previous Exponential Curve.  

Can't tell if thats a good thing or not.

But it does look like the Field number has a bit to heavy of an influence on the overall Score value.  Will need to adjust that some I think.

Thoughts?

http://www.toomanywheres.com/galcon/LeaderBoard.aspx
Re: Official Stat Warriors Leader Board :: Mar 7, 2011 @ 12:27am

Commander jazz

Joined: Feb 20, 2009
Posts: 534
Location: aka The_Musician

I have now included all game types in the Leader Board.  

Note: 2v2 has a bug in the ranks where if you play on the same TEAM as someone, it counts it as if you played AGAINST that person.  Am trying to debug that, but thought you might want to see the rankings anyway.

Also, there s no room to list 2v2_3P or 2v2_2P. lol  ;)  will need to adjust the top menu slightly to get it all to fit.
Re: Official Stat Warriors Leader Board :: Mar 7, 2011 @ 12:37am

Commander wallplant

Joined: Jul 24, 2009
Posts: 478
Location: chinatown

haha thats sweet! i really like it. i think quantum deserves a GA
Re: Official Stat Warriors Leader Board :: Mar 7, 2011 @ 12:41am

Cabin Boy esparano

Joined: Nov 5, 2009
Posts: 2308

Given:
Mastery = 100 * (Number of Opponents * Games Played) / Total Game Count At This Rank
Field = Opponents Played Against / Total Opponents Possible
Skill = Field * (Mastery / 100) * (Win Percent / 100)
Score = Win Percent * Win Percent * Rank * Rank * Skill


Substituting, 
Score = Win Percent * Win Percent * Rank * Rank * Opponents Played Against / Total Opponents Possible * ((100 * (Number of Opponents * Games Played) / Total Game Count At This Rank) / 100) * (Win Percent / 100)


Simplifying,
Score = Win Percent^3 * Rank^2 * Opponents Played Against / Total Opponents Possible * Number of Opponents * Games Played / Total Game Count At This Rank / 100

There is a difference between "variables" and "constants."  Constants are unchanging numbers that are the same for every player.  Since they are always the same number, they do not affect players' relative score.  All players' scores are divided by "total opponents possible" and by 100, so these constants can be removed for now.

Score = Win Percent^3 * Rank^2 * Opponents Played Against * Number of Opponents * Games Played / Total Game Count At This Rank

The number of games and total game count at this rank more or less cancel out. 

Score = Win Percent^3 * Rank^2 * Opponents Played Against * Number of Opponents 

Score = Win Percent^3 * Rank^2 * Number of opponents^2

The more games a person plays, the higher their rank.  Dividing by total games helped, but you're only a third of the way there haha.  It definitely helped though! :)

The Score numbers for each player look more like a Bell Curve then the previous Exponential Curve.   

Reds, blues, and one-stripes represent the greatest proportion of galcon players. Since number of opponents is squared, it gives you that bell curve..

And sorry to be so into this whole thing. I just wanted to make a few suggestions.
post updated on Mar 7, 2011 @ 1:02am
Re: Official Stat Warriors Leader Board :: Mar 7, 2011 @ 1:02am

Commander jazz

Joined: Feb 20, 2009
Posts: 534
Location: aka The_Musician

And sorry to be so into this whole thing. I just wanted to make a few suggestions.


Don't be sorry!  I'm loving the extra input!  =)
Re: Official Stat Warriors Leader Board :: Mar 7, 2011 @ 2:20am

1 Stripe Admiral quantum

Joined: May 19, 2010
Posts: 454
Location: Every(No)where @ once.

Love it!!
Re: Official Stat Warriors Leader Board :: Mar 7, 2011 @ 3:10am

Cabin Boy zmar01

Joined: Jun 6, 2009
Posts: 655

I still think there is a problem with all of your ranking systems. 

Your still taking the average rank played to be based on phils ranking system. 

I would like to see someone implement my idea. 

Make an initial score. Based on win% but don't use average rank. 

Then you have a score for everyone. 

Then what you do is replace everyones username with that initial score. Then for every game played a score is given to the winner and loser say win% X opponents initial score X games played 
Add all these together. And devide by number of games. 

And if you really want go through the same formula again. 

What will happen is the people who have a high win% against people with a high win% against people who have a high win% ect. Will be ranked higher. 

Basicly meaning who beats better players is ranked higher. 

Isn't that what a leaderboard is meant to show?

With enough stats this will be more accurate as it eliminates the random 1 game 100% win vs a GA

it is a simple idea but quite hard to sort out. 

If someone has a database (jazz) of raw stats. Can you upload it please. 
I just need basic who beat who stats. 
I don't need it updated. 
2v2 please but I don't mind doing 1v1
Re: Official Stat Warriors Leader Board :: Mar 7, 2011 @ 5:45am

Cabin Boy zmar01

Joined: Jun 6, 2009
Posts: 655

I noticed jazz that looking further down in the ranks, my brother who has hardly had galcon a week, is rated about 89, (smar01)

http://www.galcon.com/iphone/stats.php?name=smar01&gtype=2v2 

he barely has 50% wins yet your stats rate him as a 2 stripe.

He is ranked above players like:
56
jay3000
dejo2888
no.yes
my.m.crown
yoyo_xca
heisenberg

all players who would beat him any day of the week.

and you have players like peterF

http://www.galcon.com/iphone/stats.php?name=peterf&gtype=2v2 

a noob who hardly holds white flag rank 121
above most of the players named above.

Ask diebackstabber, i played 4 games with him lost every one badly and his partner was the same rank but alot better.
and i mean badly, one of those moments where your about to chuck your phone across the room type moments.

He started off with a cluster of about 5 huge planets of 1-5s, i sent 100% at him as support. He sent 100% away lost the cluster and the game. It may have been with backstabber or not.

I am quite fustrated at that to see him ranked above alot of players. If i had my way he would be the bottom 10% of players. not the top 15% of whatever.

there is no way i will accept those ranks.


you need to base the ranks more on win%, and win% vs higher ranks.

esparano has the right idea, i do like his ranks but i think there could room for improvement.

esparano's ranks seem to completly ignore wins vs lower ranks.

Which as you know people make new accounts which start of at pink flags.

thats why you need to run throught the stats once, get an initial score that does not depend on the amount of games played, but relies on win%.

then run through the ranks again (2 times minimun the more the better)

If you think about it thats like phil's current ranks now.

you need 55% wins vs the rank lower than you to rank up.

so you need high win% against high win% players.

and someone somewhere said that whats wrong with phil's ranks now?

and jazz's responce was how to calculate weight over a period of time, or something.

but there is nothing wrong with the whole win% thing now right?

high win% vs high win% players is the way to go.


And i just think there is something wrong with the whole using a rank as a score from 1 to 10.

what you need to do is give everyone a score from say 1 to 10000 that way there are 10,000 intervals rather than 10.

or whatever.

as the current rank system does not really show how good the lower ranked players are.

I can jump from 1 stripe to 3 stripe just by playing 1 game.

does that mean that someone who has high win% vs the 1 stripe bordering on 3 stripe should have less points than someone who has 50% wins vs the 3 stripe bordering on 1 stripe.

Major flaw here.

I know if the stats are calculated over many days things like 1 game will be evened out. But you get the point, 10 is a small number. you need a bigger interval.

And if your making your own ranking system dont base it on phils using his ranks.

sure i dont mind the whole GA is 0.5% of players thing you have going there, thats more showing the relative ranks. you think quantum should be GA according to your ranks.
Nothing wrong with that, its in the calculations that it matters.

Sorry quantum but as i do not think the musician's stats are correct i do not believe that you should be on top.


going back to esparano's

the current leaderboard is quite inacurate as he is eliminating playing lots of games getting lots of points, which only works if alot of games have already been played.

So he need lots of stats for his equation to be more accurate.

points = ( (10^2 x (win% vs GA) x games vs GAs) + (9^2 x (win% vs 3-stripes) x games vs 3-stripes) . . . + (1^2 x (win% vs pink flags)  x games vs pink flags) ) /total games 

that seems to much of a gap for me.

But if you were to implement my idea what you would do is just say 

(win% x initial score of player)  x games vs this player)
+
(win% x initial score of player)  x games vs this player)
+
(win% x initial score of player)  x games vs this player)
+ ect.
____________________________________________________
total games




Where my way becomes hard is actually implementing it into some sort of way so that the results can be displayed.
As it will be hard to assign an initial score that follows this player throughout the database

But as you can tell by the equation above you will have to get a score for every person you play 1 game with. the equation is going to be huge.

meaning every time a game is played the players who played this player get a score that is added on.

I can see this becoming hard.

the way to do it will be to replace that player's username with that number for every game played. But as the calculation can only be made once every player has been scored against this player wherever he may have played in the ranks. it will be hard. I may not be able to find a way.

I might be able to do it using excell and if commands referencing cells where every game is played. for each player.
I dont know, and i wont know until i start trying.

Im not into programming but i know my way around a computer.

I personally love playing around with excell so thats where i will start.



Please consider my idea.
Re: Official Stat Warriors Leader Board :: Mar 7, 2011 @ 9:44am

Commander jazz

Joined: Feb 20, 2009
Posts: 534
Location: aka The_Musician

Cool thoughts zmar!  Will think on them some. Thanks!

Btw...  the 2v2 leaderboard is all wrong right now as there is a bug in the way it calculates wins.  Still trying to sort that one out.
Re: Official Stat Warriors Leader Board :: Mar 7, 2011 @ 11:19am

Cabin Boy zmar01

Joined: Jun 6, 2009
Posts: 655



Note: 2v2 has a bug in the ranks where if you play on the same TEAM as someone, it counts it as if you played AGAINST that person.  Am trying to debug that, but thought you might want to see the rankings anyway.



Oh yerr right might explain a few things.

But doesnt that mean that peterf has played with many high ranked players?

I feel sorry for them.


Also with my idea, if you can get a score for every game played against a person in 2v2 maybe it wont be so hard to implement some kind of score to judge who you play with.
So take their initial score.
If its worse than yours then you get more points.
or something. I dont want it so that its bad to play with good people, just maybe the games played dont mean as much.

I dont want it to effect your overall ranks for the worse. 

that may be hard i haven't thought of a theory for that yet.

Maybe your initial score- your partners initial score gets added onto the game score vs those players.

idk. i dont want it to make a huge difference. But if its there it would be good just to change around the ranks a bit due to your team mates you choose.

Page 2 / 22 1   2   3   4   5   6   7   8   9   10   11   12   13   14   15   16   17   18   19   20   21   22 

You must sign in to post.