[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: computer-go: Fast scoring program?



On Mon, 13 Aug 2001, Daniel Bump wrote:

> 
> > Question anyone know of a faster scoring algorythm then gnugo? Some of my
> > games it takes up to 26 seconds to score, when the game itself only takes
> > 2 seconds. Can't really evolve nuerals nets very fast that way.
> 
> How exactly are you getting the score?
> 
using new_score

> GNU Go has two scoring algorithms. The aftermath scoring is slower.
> Any scoring algorithm will have to decide the status of the groups
> on the board. But at the end of the game I don't think it should
> take 26 seconds.
In a game of competent players i agree.. but i'm pretty sure there is a
fairly random distrabution of stones this early in the evolution, and the
games don't yet reach a true end. THe nets are still passing too often.

Here is a data point from a two games.
This data gives me a rough estimate of where the nets are in learining..




251.11 vs: 251.11[4 -72:2 -74]{1}(W+3 14s 18m) 252.1[1 -77:1 -79]{0}(B+1.0 5s 5m)
  ^          ^                 ^   ^   ^   ^
  |          white Player      |   |   |   total moves
  Black player                 |   |   Time after scoring
                               |   Score (gnugo's gtp new_score)
                               Time of game before score(in seconds)


the data in braces are 
[(Number of Passes by Black) (Avererage black stregth each time it
passed):(Number of Passes by White) (Avererage white strength each time it
passed)]

i calculate "strength" quick and dirty by (# of players peices) - (# of
oppenets pieces) - (number of open spaces). I was using the new_score for
strength till i realized it was taking so long.

I could attempt to create another neural net and train it via backprop
to come with a close score



Matthew Corey Brown                             bromoc@xxxxxxxxxxxxxxxxx 
                  Happiness is a dry place to live.