[Computer-go] replacing dynamic komi with a scoring function

[Jouni Valkonen]

Thanks Stefan to bring this up.

It would be great for educational purposes, if gobots would play always
near perfect small yose. They can play it very accurately without blunders,
if they want. So I do not think that it is necessary considering winning
ratio to play »safe moves» in the end. Of course enforcing to perfect play
should be only when gobots sees the end result with high confidence, e.g. >
80%.

There is indeed a problem with Dynamic komi with Zen. Zen often loses the
handicap games if black tries to minimize the move count. Often if it is
possible to bring game to small yose in around move 180 or so and if not
too much behind, then Zen most likely will lose. I have played few games
where Zen noted only when filling last dames that it is losing the game by
few or half points and then resign. Although, one game was that I lost by ½
points, because I accidentally defended unnecessarily instead of taking the
last dame. One game was that i was about ten points behind around move 180,
but then Zen played a slack small yose, and lost by 2½ points. Also good
and very easy strategy against zen in handicap games is to take all the
sides and give center territory to the Zen. Zen almost always will take the
center territory as too small and gives sides as too big.

Therefore it would be great that every now and then gobot would take a deep
thought and evaluate the score and then would adjust the komi so that it
looks close game, within e.g. ten points. I think that it would be
sufficient that gobot would adjust the komi in every 10 moves or so,
depending on how much it takes calculation efforts.

It would be interesting to see how much this would have an effect for win
ratio. I would say that in handicap games, performance would get better.
And perhaps also in even games performance would get somewhat better, due
to more realistic understanding of the game. And even if there is slight,
e.g quarter stone degrade in performance, it might significantly make the
game play more human like. This should be plausible, because MC-gobots in
general are playing very creatively, although sometimes they have perhaps
eaten too much mushrooms.

    –Jouni



On 9 January 2012 02:22, Stefan Kaitschick <stefan.kaitschick at hamburg.de>wrote:

> scoring function:
>
> The most successful scoring function sofar is the win/lose function.
> Sigmoid functions and other schemes have been tried, but none have
> surpassed or even equaled the simple step function.
>
> dynamic komi:
>
> dynamic komi is widely used by bots in handicap games.
> An initial artificial komi burden is placed on black which is
> incrementally reduced to zero during the game.
> This gives the bot a "more realistic" goal as white, and a motivation not
> to play slackly as black.
>
>
> One of the traps of dynamic komi is that the bot will be willing to
> simplify as white, if only he is catching up quickly enough, as specified
> by the dynamic komi.
> The bot can then find itself in a played out position, where catching up
> the final margin proves impossible.
> Conceptually, it would be nice to search at several different komi levels,
> from the komi needed for a 50% winrate(honest play), to the correct komi.
> If these values are added up(possibly weighted in some way), the result
> should be the most honest move that still has long term perspective.
> Doing this would have an obvious drawback though: less playouts per komi
> level, probably resulting in an overall weaker game.
> So here is my idea: introduce a scoring function that has the 0-border at
> the komi needed for a 50% winrate, and the 1-border at +1(as in the step
> function)
> (this would be if the bot is trailing, otherwise from 0 to komi)
> I have no idea what would be the proper function between those points,
> maybe a half sigmoid function, possibly just a straight line.
>
> The idea is not to maximize the score, but to capture the results of
> different komi levels in a single search, so as not to lose playouts.
>
>
> Stefan
>
> _______________________________________________
> Computer-go mailing list
> Computer-go at dvandva.org
> http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://computer-go.org/pipermail/computer-go/attachments/20120109/1ece07fd/attachment.html>