Elo ranking for BA 1v1 competitive play
Moderator: Moderators
Re: Elo ranking for BA 1v1 competitive play
Koshi and BD are developing a ladder system (with multiple ladders) for spring. The project seems dormant though. Maybe a collaboration or code sharing is possible:
http://ladder.springrts.com/
https://github.com/renemilk/spring-ladder
http://ladder.springrts.com/
https://github.com/renemilk/spring-ladder
Re: Elo ranking for BA 1v1 competitive play
Well, it's possible scenario that taking the time to understand others' code takes more than coding it from scratch, since there really isn't many things needed. There's just the algorithm and getting match outcomes from the SPADS bot (or extra bot joining the battle)
But I'll certainly look into it anyway, thanks.
What's indeed a good idea though, is having the bot be on multiple servers for better calibration of individual skill. I'm really considering Glicko over plain ELO though, since Glicko has it's "uncertainty modifier". The extra bit of "guessing" space gives a better chance to do quality games on average, since single person can't as easily get to the best desirable team as he would on standard ELO; There'd basically be a random factor in balancing for big games, and this is good.
But I'll certainly look into it anyway, thanks.
What's indeed a good idea though, is having the bot be on multiple servers for better calibration of individual skill. I'm really considering Glicko over plain ELO though, since Glicko has it's "uncertainty modifier". The extra bit of "guessing" space gives a better chance to do quality games on average, since single person can't as easily get to the best desirable team as he would on standard ELO; There'd basically be a random factor in balancing for big games, and this is good.
Re: Elo ranking for BA 1v1 competitive play
lol, u say ppl still play 1v1 BA?
Re: Elo ranking for BA 1v1 competitive play
Yeah, quite a lot actually, I've a few 1v1 matches basically every day I happen to be playing.NOiZE wrote:lol, u say ppl still play 1v1 BA?
They're a lot more fun nowadays than some years back, I feel balance in general is a lot better in certain respects.
- Rumpelstiltskin
- Posts: 292
- Joined: 26 Jun 2012, 18:52
- Rumpelstiltskin
- Posts: 292
- Joined: 26 Jun 2012, 18:52
Re: Elo ranking for BA 1v1 competitive play
HAHAHAHAHAHAHAHAHA
Re: Elo ranking for BA 1v1 competitive play
Huh?Rumpelstiltskin wrote:HAHAHAHAHAHAHAHAHA
Re: Elo ranking for BA 1v1 competitive play
Griffith: what k-factor do you use?
I added Elo calculation of matches to http://replays.springrts.com/ but even players with lots of matches and very good win/loss ratio don't climb up more than 1570, having started at 1500. I use a static k-factor of 24. Checkout http://replays.springrts.com/hall_of_fame/ and players individual pages (click on a players name).
Your numbers and mine don't coincide, because most of the matches you take into account have not been uploaded, and most of the matches I use are not tournament matches.
If you wish, in the future you could use the site to make the math. You'd need to make sure all matches are uploaded (happens automagically on [ACE], [BOS], [x]TheHost, Princess and Spads hosts), and tag them somehow special, so they can be separated from the others. Then a dedicated "1v1 BA tourney" rating would be no problem.
We can make a test run with the next tourney. I'll set the Elo values of all players to yours, and we'll see what happens after the uploading of the replays of the tourney.
I added Elo calculation of matches to http://replays.springrts.com/ but even players with lots of matches and very good win/loss ratio don't climb up more than 1570, having started at 1500. I use a static k-factor of 24. Checkout http://replays.springrts.com/hall_of_fame/ and players individual pages (click on a players name).
Your numbers and mine don't coincide, because most of the matches you take into account have not been uploaded, and most of the matches I use are not tournament matches.
If you wish, in the future you could use the site to make the math. You'd need to make sure all matches are uploaded (happens automagically on [ACE], [BOS], [x]TheHost, Princess and Spads hosts), and tag them somehow special, so they can be separated from the others. Then a dedicated "1v1 BA tourney" rating would be no problem.
We can make a test run with the next tourney. I'll set the Elo values of all players to yours, and we'll see what happens after the uploading of the replays of the tourney.
- Rumpelstiltskin
- Posts: 292
- Joined: 26 Jun 2012, 18:52
Re: Elo ranking for BA 1v1 competitive play
Jazcash is #1...tzaeru wrote:Huh?Rumpelstiltskin wrote:HAHAHAHAHAHAHAHAHA
Re: Elo ranking for BA 1v1 competitive play
I used K=30 with only the tournaments games. Since you use a lower value for K that could explain the fact you have lower elo as mine.dansan wrote:Griffith: what k-factor do you use?
Awesome work !dansan wrote:I added Elo calculation of matches to http://replays.springrts.com/ but even players with lots of matches and very good win/loss ratio don't climb up more than 1570, having started at 1500. I use a static k-factor of 24. Checkout http://replays.springrts.com/hall_of_fame/ and players individual pages (click on a players name).
Amazing, this will be so usefull
Could you try with K=30 to see if there are a lot of difference ?
I made the calcul manually on few games compare to you. Maximum number of games for a player was around 40 or 50.
Ok i will take a look at it for the next tourney.dansan wrote: Your numbers and mine don't coincide, because most of the matches you take into account have not been uploaded, and most of the matches I use are not tournament matches.
If you wish, in the future you could use the site to make the math. You'd need to make sure all matches are uploaded (happens automagically on [ACE], [BOS], [x]TheHost, Princess and Spads hosts), and tag them somehow special, so they can be separated from the others. Then a dedicated "1v1 BA tourney" rating would be no problem.
We can make a test run with the next tourney. I'll set the Elo values of all players to yours, and we'll see what happens after the uploading of the replays of the tourney.
Again amazing stuff !
Re: Elo ranking for BA 1v1 competitive play
I'll do that tomorrow.Griffith wrote:Could you try with K=30 to see if there are a lot of difference ?
Re: Elo ranking for BA 1v1 competitive play
I made a dedicated rating category "1v1 BA tournament", and seeded the players with your latest (21 Sep 2012, 01:02) Elo values: http://replays.springrts.com/ba1v1tourney/
Tomorrow I'll run the rating() on those matches: http://replays.springrts.com/tag/Tourney/ (from today and the final). I hope those replays are complete...
I'll leave the k-factor at 30 for everyone. You do it like that too?
We should have the same results after that... :)
Tomorrow I'll run the rating() on those matches: http://replays.springrts.com/tag/Tourney/ (from today and the final). I hope those replays are complete...
I'll leave the k-factor at 30 for everyone. You do it like that too?
We should have the same results after that... :)
Re: Elo ranking for BA 1v1 competitive play
Calculations ran, but results are different... hmm... any idea why?
You can see the ratings of the players on each matches page. The value are the Elo after the match.
You can see the ratings of the players on each matches page. The value are the Elo after the match.
Re: Elo ranking for BA 1v1 competitive play
Ratings after tourney
I have an idea why our elo are not exactly the same.
For all the calculation of a tourney i take the initial elo. For example player A has 1520 elo before the tourney. For all his matchs i will take 1520.
Maybe you take the new elo as a base for new calculation:
Player A has 1520 elo before tourney. First game he loose, get new elo of 1510. Next game you take his elo of 1510 to calculate.
I take for whole tourney 1520 for player A.
Maybe this can explain the difference ?
I have an idea why our elo are not exactly the same.
For all the calculation of a tourney i take the initial elo. For example player A has 1520 elo before the tourney. For all his matchs i will take 1520.
Maybe you take the new elo as a base for new calculation:
Player A has 1520 elo before tourney. First game he loose, get new elo of 1510. Next game you take his elo of 1510 to calculate.
I take for whole tourney 1520 for player A.
Maybe this can explain the difference ?
Re: Elo ranking for BA 1v1 competitive play
I do it exactly like you describe... is that wrong? Uh.. I have to read up on that...
Re: Elo ranking for BA 1v1 competitive play
In a way it is wrong.dansan wrote:I do it exactly like you describe... is that wrong? Uh.. I have to read up on that...
You're supposed to adjust the Elo after a tournament, not during it, in the original system.
Re: Elo ranking for BA 1v1 competitive play
Ah yeah... just found this: wikipedia/Elo_rating_system#Live_ratings
Sigh... I'll have to rewrite that rating function, and think about how I'll handle all the uploads...
I guess I'll do it exactly as it's done in chess: I'll maintain an official rating updated once a month (or every 2nd week) and live ratings in between. They will get reset to the official rating each month.
... and ofc the tourney ratings apart as well :)
Sigh... I'll have to rewrite that rating function, and think about how I'll handle all the uploads...
I guess I'll do it exactly as it's done in chess: I'll maintain an official rating updated once a month (or every 2nd week) and live ratings in between. They will get reset to the official rating each month.
... and ofc the tourney ratings apart as well :)