Page 4 of 8
Posted: Fri Oct 24, 2008 4:09 am
by Correct
We warned you not to comm
Posted: Fri Oct 24, 2008 5:48 am
by CronoDroid
Don't worry about rank Fed. Anybody who's anybody knows who the real pro commanders are around here. And noobs will soon learn of your majesty if they bother to stick around. Personally I'll do my bit to spread the word that people should ignore the first few weeks of your playtime (thread creation included) and only look at your behaviour post-FederalMan.
Phewwwww.
Posted: Fri Oct 24, 2008 6:50 am
by H_Mallow
I've been saying forever anyone with half a brain can win about %80 of the time, even higher if you lobby sit for the first 10-15 minutes.
Helo rank is more of a conscience meter than anything else.
Posted: Fri Oct 24, 2008 7:03 am
by ImmortalZ
Malicious Wraith wrote:QUOTE (Malicious Wraith @ Oct 24 2008, 09:15 AM) I say all data counted should be from the last 100-200 games played, to stop inaccuracy due to skill change.
It tracks changes over time. There were some nifty graphs I saw while it was still in early development.
Posted: Fri Oct 24, 2008 7:14 am
by badpazzword
<math>
The size of Mu updates are proportional to the inverse of sigma squared. (Mu is perceived skill, Sigma is a reasonable error on mu.) Sigma goes down after each game, but it is also bumped by a costant called "dynamics" that prevents Sigma from going below a certain level (circa 0.8). So AllegSkill does track skill changes.
</math>
The graph and the numbers show this approach does work
Posted: Fri Oct 24, 2008 11:52 am
by Malicious Wraith
Stupid smart people graphs.
All those wizzy woo's, and whatchamacallit's.
Posted: Fri Oct 24, 2008 12:01 pm
by takingarms1
Malicious Wraith wrote:QUOTE (Malicious Wraith @ Oct 23 2008, 11:45 PM) I say all data counted should be from the last 100-200 games played, to stop inaccuracy due to skill change.
The system has to be designed to accommodate skill change over time, since it anticipates newbies joining and gaining skill over time as compared with vets. As others have said, the system does this quite well.
Posted: Fri Oct 24, 2008 12:13 pm
by sgt_baker
If we do discover that the system isn't adjusting to varying skill quickly enough, it's just a matter of increasing one of the system variables (
dynamics, gamma) in order to have the system converge more rapidly on a new skill rating.
Posted: Fri Oct 24, 2008 7:38 pm
by sgt_baker
From a PM:
Malicious Wraith wrote:QUOTE (Malicious Wraith @ Oct 24 2008, 05:24 PM)On the Command Skill part of Allegskill, would it be possible to run the numbers on the following?
First, take the ranks of the top 10 commanders. If I am not part of it, than Allegskill is broke. : D
Determine my placement under the current system, and adjust the variables to see its mobility. Run like 5 different variables in both directions, and tell me how it changes my placement.
Hopefully, this will give me a good sense of how it will operate, so I can give you some feedback. Having commanded against all the top commanders and having won, I have unique perspective on their individual strengths and weaknesses that math can only hope to copy. : )
WTF is this twit on about?
Posted: Fri Oct 24, 2008 7:54 pm
by Grimmwolf_GB
lol