[Freeciv-Dev] (PR#6680) new ranking update

["Horn Gábor" <Horn.Gabor@xxxxxxxxxxx>]

Hi!

I've updated the new ranking at http://aktiv.co.hu/~hirisov/civserver/ .
Changelog:

1, ranking decay: it works this way: after X games of the same type
(ffa,duel or team) Y percent of (points-1000) will be substracted from
the player (note if he's below average his point will increase). For the
formula and idea now check (i'll include it in details section soon):
http://www.gamasutra.com/features/20000209/kreimeier_pfv.htm "closed
system" section. It can be seen decay doesn't change order of players,
but in long term will lower the rank of players who doesn't play (no
more 'sitting on the top w/o playing'). I used X=100 and Y=5. So after
100 ffa games the scirpt decays all the ffa points by 5 percent of
actual pts-1000. Same for duel and team.

2, hall of fame: it's a less fluctuating way to represent players' long
term stregth. It works this way: after an ffa game the #1 of ffa ranking
gets Z pts in hall of fame list, the #2 of ranking gets Z-1 and so on.
So the first Z players in the ranking gets point after every ffa games.
It means if somebody has a  good rank in the past for long time but
plays only a few games now, his/her ranking will be relatively low
because of the decay, but his/her hall of fame points will be high (eg
warlock and pille in ffa). I used Z=10.

3, player comparsion: on this page you can compare 2 players due to the
games they both played in. 

4, player history page: if u click on the player name on the ranking
table, you get his personal history. I've updated this page to reflect
decay (vertical lines in graph and grey background in the table), and
show separated tables for the different gametypes.

Questions:

1, decay needs much cpu power, and maily because of the inactive
players. Currentyl there're 5781 players in the database, and 4730 of
them played less than 5 games! And all of them has to be processed in
the decay too. It can be seen they don't change the ranking much, but
couse a 400% (actually more) overhead. If we count only tha players who
has at least 10 games, we end up w/ 500 of them, so 900 percent
overhead... I think it's nod good (well it took about 30 hours to
regenerate the whole ranking on my athlon xp 2000 w/ decay :) w/o decay
it's about 1 hour. it's unacceptable for the server which now runs the
pubservers too).  My suggestions:

a, only start to count a player's point after X games, where is 5-10. 

and/or

b, after a period remove the inactive players from the system.

(well if we reset the ranking db it won't be problem for long time, but
when we reach this amount of players in the db it will. Might be better
to prevent this). 

2, parameters: 

The important ones:

a, decay period: now it's 100 for all type of games. The lower it is the
more accurate the ranking and the more resource of the server is used.
b, decay rate: now it's 0.05 so 5 percent. The higher it is the faster
is the decay.
c: overall characteristic of points move in a game. For an idea now
check http://www.gamasutra.com/features/20000209/kreimeier_pfv.htm fig 5
picture. I'll generate a similar graph w/ the constants we use. From the
duel ranking i have a feeling that now one get too much point by killing
much weaker players (see war king's result). And at hall of fame this
effect cummulates.

Anybody has suggestion to finetune them?

bye, hirisov