So really, what I want to be able to do is:
1. Use the ELO rating system.
2. Determine how many ELO points 1 stone handicap is worth.
3. " " " " " 2 stones are worth
4. " " " " " 3 stones are worth,
etc.
When two players are matched, the server gives the handicap that
most closely matches them up. For rating purposes when the
game is over their ratings are adjusted to reflect the handicap.
For example if 2 players a 400 ELO apart, they may get
rated as if they are only 50 ELO different. It may even turn out
that the weaker player has a higher expectancy of winning due to
the handicap.
We start with an initial estimation of what a stone is worth in ELO
points, and let the server make small adjustments over time to
make it match up with reality. Each extra stone of handicap is
tracked separately because it's not likely the ELO gap is uniform.
The adjustment to ELO is simple, we have an expectancy based on the
ELO rating of both players (along with the handicap ELO compensation)
and we have the actual results. We make a proportionate adjustment
based on the difference between what we expected to happen and what
actually did happen, which I think is exactly the same thing the
ELO formula does!
And it doesn't really matter which handicap system we use.
I think I slightly prefer fixed placement handicaps, as I believe
the weaker programs will benefit less from handicaps otherwise,
which is the opposite of what we want.
I don't plan to add compensation for the handicap stones.
- Don
On Fri, 2006-12-22 at 16:01 -0500, Don Dailey wrote:
> I'm glad you bring it up.
>
> >From the same site, it appears there is no standard way of handling
> this.
> I will look to see what Tromp/Taylor says if anything.
>
> It would be nice if we could simple equate handicap with ELO points, I
> think it would be more accurate. We may find that 1 stone per kyu
> doesn't hold up forever. Then we just use ELO (converting in a
> straightforward way to kyu if we want this) and have a formula
> (or table) for compensating ELO. For instance we may determine
> that 400 ELO can be compensated by 4 stones and this effectively
> changes the ELO calculation.
>
> This seems more mathematically logical to me. Perhaps the server
> itself can converge on the right formula.
>
> - Don
>
>
> On Fri, 2006-12-22 at 21:46 +0100, Christian Nilsson wrote:
> > Yes, in chinese rules you need to compensate white for the extra area
> > you gain from the actual stones. The handicap is only meant to be the
> > extra strength/stability.
> >
> > One can of course ignore this for the server. I just wanted to make
> > sure all programs use the same rules. I don't know what the
> > tromp-taylor rules says about it.
> >
> > On 12/22/06, Don Dailey <[EMAIL PROTECTED]> wrote:
> > > I'm trying to figure this out. If you get a 9 stone handicap, you have
> > > to give back those 9 stones? So a 9 stone handicap is not quite as
> > > much
> > > as it seems although it's still pretty good.
> > >
> > > - Don
> > >
> > >
> > >
> > >
> > >
> > > On Fri, 2006-12-22 at 21:24 +0100, Christian Nilsson wrote:
> > > > There's also the small issue of the compensation given to white
> > > > because of the extra black stones on the board. Setting a modified
> > > > komi would break (MC-)programs with an internal rule for it. Not
> > > > setting it would break those who does not use that rule.
> > > >
> > > > How is this compensation handled by the various programs on cgos, if at
> > > > all?
> > > >
> > > > Check http://www.britgo.org/rules/compare.html#comp if you don't know
> > > > what I'm talking about..
> > > >
> > > > //Christian
> > > >
> > > >
> > > > On 12/22/06, Don Dailey <[EMAIL PROTECTED]> wrote:
> > > > > Ok,
> > > > >
> > > > > Well I'm inclined to go with the majority which seems to have turned
> > > > > around
> > > > > from the last time I polled.
> > > > >
> > > > > Now the question: How to set it up?
> > > > >
> > > > > Here are the options:
> > > > >
> > > > > 1. Use GTP handicap commands to set up game.
> > > > >
> > > > > 2. Send the appropriate pass commands to get the initial setup
> > > > > to accommodate programs that have not implemented handicap.
> > > > >
> > > > > 3. Do both - send handicap to programs that can handle it,
> > > > > otherwise
> > > > > guide them through it by sending play commands with passes.
> > > > >
> > > > > In any case, I would make the game records (SGF) look correct, doing
> > > > > whatever
> > > > > that takes.
> > > > >
> > > > > - Don
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > On Fri, 2006-12-22 at 20:48 +0100, Magnus Persson wrote:
> > > > > > Quoting [EMAIL PROTECTED]:
> > > > > >
> > > > > > > on 9x9 boards. To make a long story short, I didn't see any
> > > > > > > evidence
> > > > > > > that this algorithm is fundamentally disadvantaged in handicap
> > > > > > > games.
> > > > > > > In fact, I agree with Remi's view that it is particularly *well*
> > > > > > > suited to handicap games compared to territory based algorithms.
> > > > > > > When
> > > > > > > it finds itself behind, it goes for the swindle. Against an equal
> > > > > > > opponent, that's obnoxious and futile. Against a weaker foe, it's
> > > > > > > wise.
> > > > > >
> > > > > > I have to add to this. My opinion about Valkyrias play as white with
> > > > > > handicap on
> > > > > > 9x9 is that it plays excellent handicap go (given its strength at
> > > > > > even). In my
> > > > > > view I do not have to change anything. But this is of course just my
> > > > > > impression.
> > > > > >
> > > > > > It is true that MC-programs plays randomly in the end of the game
> > > > > > but in the
> > > > > > opening the handicap stones are just a burden and does not really
> > > > > > make it play
> > > > > > random.
> > > > > >
> > > > > > -Magnus
> > > > > > _______________________________________________
> > > > > > computer-go mailing list
> > > > > > computer-go@computer-go.org
> > > > > > http://www.computer-go.org/mailman/listinfo/computer-go/
> > > > >
> > > > > _______________________________________________
> > > > > computer-go mailing list
> > > > > computer-go@computer-go.org
> > > > > http://www.computer-go.org/mailman/listinfo/computer-go/
> > > > >
> > > > _______________________________________________
> > > > computer-go mailing list
> > > > computer-go@computer-go.org
> > > > http://www.computer-go.org/mailman/listinfo/computer-go/
> > >
> > > _______________________________________________
> > > computer-go mailing list
> > > computer-go@computer-go.org
> > > http://www.computer-go.org/mailman/listinfo/computer-go/
> > >
> > _______________________________________________
> > computer-go mailing list
> > computer-go@computer-go.org
> > http://www.computer-go.org/mailman/listinfo/computer-go/
>
> _______________________________________________
> computer-go mailing list
> computer-go@computer-go.org
> http://www.computer-go.org/mailman/listinfo/computer-go/
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
