AllegSkill auto balance

sgt_baker · Post by **sgt_baker** » Thu Jan 17, 2008 8:29 pm

Email sent to MSR. /me awaits reply.

Edit: /mrgreen.gif" style="vertical-align:middle" emoid=":D" border="0" alt="mrgreen.gif" /> Grim

sgt_baker · Post by **sgt_baker** » Thu Jan 17, 2008 8:46 pm

A bit of background info:

The majority of work on this project has been performed against a database without player names. This was the result of our well-known Pook bear's paranoia wrt people knowing hinder nicks. The upside to the situation was that we were forced to perform a true blind study, meaning not knowing who player 72145 is. The vast majority of the problems/features of AllegSkill were solved/developed under these circumstances.

More recently Tigereye has released the names to us, and it has allowed our own 'human aptitude in parallel estimation' to judge the results of our work. The latest revision of AS (details of which will be in the paper) renders a leader board which seems to make sense. Human interpretation aside, AllegSkill manages to predict game outcomes with 86% accuracy when looking at the entire Alleg database, and based on a very simple test: The higher-ranked side wins. From a statistics point of view, 86% is, well, accurate given the numerous WTF factors which affect alleg. I'm not, and never will, claim that AllegSkill is the be-all-and-end-all, but there comes a point where one must select the best solution available. This is that point. For all the suggestions to the contrary, AllegSkill is the best, and only, sensible alternative we have to the recent ranking mess. It's gonna be active in R5, and if something is majorly awry, you can bet your bottom dollar (and all you cows) that I'm gonna be all over the problem like a bad rash. /smile.gif" style="vertical-align:middle" emoid=":)" border="0" alt="smile.gif" />

B

badpazzword · Post by **badpazzword** » Thu Jan 17, 2008 8:51 pm

Did I mention player 3304 rocks?

Lykourgos · Post by **Lykourgos** » Thu Jan 17, 2008 8:52 pm

It would have been cool if it had given you callsign instead of ASGS player ID and you could have seen whether people like Aarm who play a lot under multiple nicks have similar ranks on all nicks.

Maybe another test? Possibly not valid because some such people play very differently on their hiders.

edit: HELO also does a pretty good job of predicting the winning team as I recall, though I remember it being 60%ish and not 90% ish.

sgt_baker · Post by **sgt_baker** » Thu Jan 17, 2008 8:54 pm

Lykourgos wrote:QUOTE (Lykourgos @ Jan 17 2008, 08:52 PM) It would have been cool if it had given you callsign instead of ASGS player ID and you could have seen whether people like Aarm who play a lot under multiple nicks have similar ranks on all nicks.

Maybe another test? Possibly not valid because some such people play very differently on their hiders.

edit: HELO also does a pretty good job of predicting the winning team as I recall, though I remember it being 60%ish and not 90% ish.

We have both with the lasest release. Interesting idea...

/me remembers he has @#(!loads of work to do and can't get side... damnit! /mrgreen.gif" style="vertical-align:middle" emoid=":D" border="0" alt="mrgreen.gif" />

Grim_Reaper_4u · Post by **Grim_Reaper_4u** » Thu Jan 17, 2008 9:59 pm

sgt_baker wrote:QUOTE (sgt_baker @ Jan 17 2008, 09:46 PM) based on a very simple test: The higher-ranked side wins. B

Did you also test this against HELO to see how HELO did with it's ranksums? What % of our current games end with the highest ranked team as winner?

sgt_baker · Post by **sgt_baker** » Thu Jan 17, 2008 10:05 pm

Grim_Reaper_4u wrote:QUOTE (Grim_Reaper_4u @ Jan 17 2008, 09:59 PM) Did you also test this against HELO to see how HELO did with it's ranksums? What % of our current games end with the highest ranked team as winner?

Not yet, but I will. /smile.gif" style="vertical-align:middle" emoid=":)" border="0" alt="smile.gif" />

Kap · Post by **Kap** » Fri Jan 18, 2008 1:22 am

A stats page of every player that showed how much time he has spend in an int, scout, etc. ; how many kills in each ship, major bases killed, htts. Probes deployed? maybe hull repaired if it could be tracked.

Well perhaps thats too much and too difficult but I think at least a page that could be opened by right clicking someones name in the lobby and showed you the time spent in each ship could be a good way of.... hmm dunno.. it would be cool tho

quackdamnyou · Post by **quackdamnyou** » Fri Jan 18, 2008 1:38 am

I will vote yes. However I believe that only counting stats for AB games is the best solution. People will be free to play 'loose' if they like, and after a while we will see a community preference.

Post by **BackTrak** » Fri Jan 18, 2008 1:44 am

FingerBang wrote:QUOTE (FingerBang @ Jan 17 2008, 10:12 AM) You do this and Allegiance will die. Do you know how many times there aren't 2 people who want to join? I do understand that this may be a "WISH" of yours to implement, but this has to be the worst idea i have heard. Sure if we had a large player base it might work, but with the number of ppl playing at any given time, it just wont work and alot of ppl will be stuck 'waiting' for this second person and leave in frustration, only to have someone want to join a team 5 minutes later., and so on..

So, to sum it up, THIS idea=worst idea in alleg history.
BUT please keep up the good work so far, except for this idea, dotn think about this idea any more, but others are ok to think of, but just not this one, ok? thanks..

I don't know about that Finger, it seems that it would get people to quit hanging around on NOAT for a game to start so that they can quickly stack a side. The one minute timeout that Baker suggested would be fine if the user is displayed a timer until they can join. (Some user feedback would be required). This would be much better than just sitting flashing, waiting for the comm to accept. I haven't kept logs, but I'm sure I'm waiting longer than 60 seconds to join on some launched games.

If it's selectable as a game mode, I think it's an interesting idea.

Baker, thanks for the info, very clear.