Hello!
/me waves
/me pondes the worrying lack of correspondence arriving in my inbox
For the record: We've had SUPER-EFFECTIVE pre-game autobalance working and in code since well before AllegSkill even saw the light of day. I've made a point of this every time someone asks. Post-launch algorithms are, however, far more tricky. I doubt that anyone working on such things has taken the time to think about the problems inherent therein properly.
THE PROBLEM IS THIS: Currently, ASGS doesn't pass a user's FULL rank data to the server. ASGS only passes the "lowest common denominator" single rank number. In order for any of our algorithms to work, we need at least a mu and sigma for each player to be passed to AllSrv.exe when a user is passed from the lobby to the server (or however it works). Until that happens, you're not going to see any movement in this area.
We have some post-launch balance algorithms which perform better than any system based purely on (MyRank) alone, yet I'm not absolutely happy with them. If someone were to demonstrate to me that there might be developments in this area, I'd be happy to re-engage in the research required to develop said algorithms to fruition.
Again, this requires that BOTH mu and sigma are available the the server performing the balancing operation.
B
New algorithm for autobalance=auto
-
ThePhantom032
- Posts: 836
- Joined: Sat May 09, 2009 11:00 am
- Location: Germany
Just to clarify:
This is NOT a pre-launch algorithm. This is a post-launch algorithm, probably the best one we have AND can use (imbal = auto/1/2 suck after all).
And, as I said before, this algorithm as its implemented does NOT use mu or sigma but only the available rank.
As baker said it might be BETTER to use an algorithm using mu and sigma but we dont have these and i doubt we get them while using ASGS as we cant change that.
This is NOT a pre-launch algorithm. This is a post-launch algorithm, probably the best one we have AND can use (imbal = auto/1/2 suck after all).
And, as I said before, this algorithm as its implemented does NOT use mu or sigma but only the available rank.
As baker said it might be BETTER to use an algorithm using mu and sigma but we dont have these and i doubt we get them while using ASGS as we cant change that.
Still ready to teach anyone who asks nicely whatever they want to know about playing alleg. Contrary to popular opinion I do not eat newbies. Voobs taste much better.
In fact one of the reasons I can't be arsed to chase anything to do with ranking in Alleg is that EVERYONE has an idea which, according to them, is "TEH BEST TING SINCHAH ANYWUN INVENTED TEH RANKS!"
Do you even know what mu and sigma are doing? Do you know what they mean?
Do you even know what mu and sigma are doing? Do you know what they mean?
Granary Sergeant Baker - Special Bread Service (Wurf - 13th Oct 2011)
sgt_baker: Yes, having mu/sigma available on the server would be nice. But that's not what I was trying to do. I was trying to find something better than the existing autobalance=auto which can work in the confines of the existing codebase/ASGS implementation.
Bard: I'm managing my own open source project, and let me tell you: The sentence "Please don't code right now" is capable of killing any open source project. Be very, very careful when using it.
People contribute to open source because they think it's fun to code. Let them try out ideas, even if it's just on an experimental branch. Let them improve stuff. Coders want to code. If they can't code here, they'll move somewhere else.
We won't get a nice post-launch balancing algorithm if we don't even try to start moving in that direction. The perfect is the enemy of the good.
Bard: I'm managing my own open source project, and let me tell you: The sentence "Please don't code right now" is capable of killing any open source project. Be very, very careful when using it.
People contribute to open source because they think it's fun to code. Let them try out ideas, even if it's just on an experimental branch. Let them improve stuff. Coders want to code. If they can't code here, they'll move somewhere else.
We won't get a nice post-launch balancing algorithm if we don't even try to start moving in that direction. The perfect is the enemy of the good.
Last edited by the.ynik on Thu Jul 29, 2010 5:53 pm, edited 1 time in total.
-
ThePhantom032
- Posts: 836
- Joined: Sat May 09, 2009 11:00 am
- Location: Germany
sgt_baker: Its just as ynik said.
I'm not ignoring years of research; as you said YOURSELF we currently can not use mu/sigma based algorithms so I thought we should go with something we CAN use.
This algorithm is better than our current autobalance=auto (post launch).
Why not use it until we can do better?
I'm not ignoring years of research; as you said YOURSELF we currently can not use mu/sigma based algorithms so I thought we should go with something we CAN use.
This algorithm is better than our current autobalance=auto (post launch).
Why not use it until we can do better?
Last edited by ThePhantom032 on Thu Jul 29, 2010 6:00 pm, edited 1 time in total.
Still ready to teach anyone who asks nicely whatever they want to know about playing alleg. Contrary to popular opinion I do not eat newbies. Voobs taste much better.
Thanks P32.
Actually, unless someone who wants to implement the mu/sigma based solution (including modifying ASGS) steps up, let's concentrate on the questions I asked in my original post:
By the way: if my changes get implemented, server operators can still revert to the old autobalance (let users join side with less AS) by choosing:
Average AS Weight: 0
Total AS Weight: 1
Player Count Weight: 0
Flexibility: 0
I agree that the input found by Botzman is problematic; especially if no non-newbie is available on NOAT to even it out (not unusual for small games). We might have to decrease the weight on average AS a bit to avoid this scenario.
The problem here is that sometimes is impossible to really balance the game when a single player is joining; you'd need to let multiple players queue up and then balance them at once. I wanted to avoid that for the algorithm (don't ever force someone to wait). And remember that commanders and players can (and should) still watch the stack. This algorithm is NOT replacing your brain.
Actually, unless someone who wants to implement the mu/sigma based solution (including modifying ASGS) steps up, let's concentrate on the questions I asked in my original post:
I think the answer to my first two questions is YES, this new algorithm is an improvement over the status quo. Feel free to replace it with something else later when ASGS/CSS make mu/sigma/stackrating/whatever available; but I think my algorithm (or some variant of it) is the best option we have for R6.the.ynik wrote:QUOTE (the.ynik @ Jul 22 2010, 06:02 PM) Do you think this is better than #autobalance auto, and should be implemented in R6 to replace the existing #autobalance auto implementation?
Do you think it's better than #autobalance 1 and stands a chance to actually get used?
Did you find any scenarios using the web app where it doesn't work as you would expect?
By the way: if my changes get implemented, server operators can still revert to the old autobalance (let users join side with less AS) by choosing:
Average AS Weight: 0
Total AS Weight: 1
Player Count Weight: 0
Flexibility: 0
I agree that the input found by Botzman is problematic; especially if no non-newbie is available on NOAT to even it out (not unusual for small games). We might have to decrease the weight on average AS a bit to avoid this scenario.
The problem here is that sometimes is impossible to really balance the game when a single player is joining; you'd need to let multiple players queue up and then balance them at once. I wanted to avoid that for the algorithm (don't ever force someone to wait). And remember that commanders and players can (and should) still watch the stack. This algorithm is NOT replacing your brain.
Last edited by the.ynik on Thu Jul 29, 2010 6:18 pm, edited 1 time in total.
-
TurkeyXIII
- Posts: 1460
- Joined: Thu Dec 06, 2007 3:18 am
- Location: Melbourne, Aus
That's pretty-much it. I'm sure one day there will be developments in this area, and we'll all need to poke you to figure out how to use them properly. In the mean time, ynik has a way to make autobalance less stupid using conservative ranks alone.sgt_baker wrote:QUOTE (sgt_baker @ Jul 30 2010, 03:16 AM) THE PROBLEM IS THIS: Currently, ASGS doesn't pass a user's FULL rank data to the server. ASGS only passes the "lowest common denominator" single rank number. In order for any of our algorithms to work, we need at least a mu and sigma for each player to be passed to AllSrv.exe when a user is passed from the lobby to the server (or however it works). Until that happens, you're not going to see any movement in this area.
We have some post-launch balance algorithms which perform better than any system based purely on (MyRank) alone, yet I'm not absolutely happy with them. If someone were to demonstrate to me that there might be developments in this area, I'd be happy to re-engage in the research required to develop said algorithms to fruition.
Again, this requires that BOTH mu and sigma are available the the server performing the balancing operation.
QUOTE (Randall Munroe)14.2: Turkey consumption rate of the average American in milligrams per minute[/quote]
