[Freeciv] Re: recounted rankings

[Horn Gábor <Horn.Gabor@xxxxxxxxxxx>]

Hi!

I was suggested to redirect this conversation to rt@xxxxxxxxxxxxxx so
i'll do this from this letter. 


> The distribution curves are nice.
> A calculation of correlation between player
>  results would be also useful to reweal alliance
> structure. A "r" close to 1 or -1 can be also
> suspicius. When somebody directs two player
> or there is a continous (more games involving)
>  cooperation among a gang.

Interesting idea, i'll thinking on it, althought it's a bit risky to say
some things, that would need so much data to produce reliabel info, that
everybody would recognize it even w/o that. Actually even now i think we
mostly knows about each other (i mean the players who play frequently on
the pubserver:) who likes to be a gang and who not.

> 
> (I have to admit, your methods are not in their details
> clear for me.)
> On the other hand. i don't think that recalculation
> of ranking will solve the problems.  (1) How will you
>  check that it works correctly for all present and future
> strategies? (2) Doesn't it prefers some way of
> thinking, alone figthing, ...? (3) Should the new
> ranking system show, that you are even
> better player?

Recalculation isn't about to solve anything :) It's more a check, and to
recognize if it has problems or weaknesses. I simply need much data to
test a gaussian distribution like function, and pubserver's database is
ideal for it. It contain many real data, and we can check if it's good
or not (I mean players ususally knows where they r compared to others,
whom they ususally lose to, who they usually defeat, etc so if the
ranking generates something similar it's good).

1, i can't check. Ranking isn't for that. Ranking is to reflect players
experience, in a (hopefully) objective way. It knows nothing about their
strategies, it only can measure their results.

2, One of my primary goals was to reflect the role of alliances. The
current script doesn't handle alliances at all. If u loose to a player,
it doesnt matter now at all if he did it alone or w/ 5 ally againts you
alone, he gots exactly the same point. It's both a wrong system and
ruins the game imho. So yes, i hope if players won't benefit that much
from mindless gangbangs, they won't do them. Gang is wrong i think,
because the ganger won't learn and develop, and the ganged will get
frustrated. Now if u fight againts a same powered alliance there's no
problem, u'll get much points. Alliances are simply part of the game,
and has to be reflected by the ranking system too imho. It's even good
for team games (currently none of the high ranked players use their real
nick in team game, because it handles them wrong).

3, 'you' means me personally? If yes, i think i got better ranking
because of calculation reflects the alliances. You know recently players
tends to do gangs easily if they think somebody is strong. I rarely do
alliances, so most of my victories haven't got 'weaker' because of the
new system. And when i lose in a normal game (in duel there's no
alliances, in team i use team nick) recently that's usually happens if i
fight againts an alliance alone. So w/ the new system i don't loose that
much point in these matches that in the current ranking. Eg in one of my
last games was #232924, when i lost againts Zed, a relativelly low
ranked player, who had 1 ally and 2 chain allies (i call a chain ally
when a nation is the ally of your ally, but not yours directly. In many
cases that nation attacks your enemy too, and due to the person who
allied both of u, u get each others techs too... so u could be counted
allies i think), so overall they were 4 and all attacked me. In the
current it was counted for me as i was lost to him in a duel, and i lost
23.6 points in that match (to recover 23 points i have to defeat a TOP3
player in duel...do u think it's fair? :). In the new system, if i count
only his 1 ally (so 2v1) i loose 16,2 points (so 30 percent less), if i
resolve his chain alliances (so 4v1..and in that game it was the case,
so it would describe it most exact!) i only loose 1.23 (so only 6
percent of the original loss!) points. The graph and points i published
is made w/ NO examining the chain allies, so as u see if resolve them
every player (including me) who got ganged more times would have more
points, and the gangers less. 

Well i'm not sure which is the good way for handling chain alliances
(maybe only count chain ally as real ally if u both get city from the
same nations?) so now i made it optional in the new script.


> According to the third rule of good strategic
> games:
> "The game can not be won all times with the
> same strategy, independently from circumstances."

Freeciv is good in it, it cannot be won w/ same startegies, althought no
dubt there're betters and worses. First it has many random factor now
(map generator) so u have to adapt to that. Second, especially in duels
w/ a close ranked player, u need much thinking.

> 
> Examples for all time used "strategies":
>     Use only a science rate higher than 50%.

In war u don't do. In peace sure u do.

>     Build only "happy wonders"+"magellan".

Eg in largepox games it's not true. Nowadays there i more and more lpx
games (and i think it's cool). But i agree these are a bit powerful, but
it's from the original game. Who knew then waht civ will turn into? :)

>     Make as much alliances as possible.

It's the tactic of weak players i think. I rarely see good players
making 'as much alliances as possible'. W/ the current sysytem there's
no penalty at all if u ally everybody. In the new ally has an effect on
ranking. The more person u ally, the less u do personally for the
vitory, so the less points u'll get.

> 
> If the above things are working, there is a probleme
> somewhere. First, check, if there is an anti-strategy
> (surely you have tried out severel ways).
> Let me suggest one more, called "threatening".
> You know that one of your neighbour (A) is willing to
> make a treaty with an other (B). Both of them have the
>  interest, becouse exchangeing knowledge is surely
> advantageous for them. Then you send for B
> a message "If you sign the treaty I will devastate you". When
> you really stronger than B, and you can cause severe
> damages (pillage, poison), signing the treaty will mean
> disadvantage for B. And for a valid treaty two parties
>  needed. (A technical question (4): can you get a message
> about foreign diplomatic activity before signiture?)
> 

No way for getting info about what others do w/ each other. Warlock
suggested idea of public notification of alliances, so if two people
agree on alliance a game message appears to let the others now. Well she
got ganged MANY times, in her games when people allied that was almost
always againts her, i think this is why she's a bit mistrustful about it
:)

> If none of the anty-strategies working there
> should be some new game rules to made. In this
> case enlarging knowledge exchange risk is
> a possibility, eg players can give fake knowledge
>  documentation to others, and units/buildings based
> on it will not work. Imagine how risky is it when you
> belive that you have an army of musketeers, but their
> weapons working in battles not.
> 

Interesting idea, but i think it's a bit far from now. I can see some
things which would be more important to improve. But i guess the devs
would be happy if u can contribute patch about it :)

szia, hirisov