[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: computer-go: Perl Module for next move.
> Try --score end or --score aftermath, both of which are
> suitable for scoring the game at the end. These options
> both generate moves until both pass, then compute a score.
> Currently --score aftermath runs a bit too slow, but has
> ambitions of giving a fully correct algorithm.
>
> Currently --score last, which would try to compute an approximate
> score before the end of the game without generating moves makes
> a seg fault.
I am wondering how accurate is the scoring using --score end? As gnugo is
generating moves till the end of the game, the evaluation is only as
accurate as gnugo is able to play perfectly that rest of the game. With NNs,
they may end the game even after let's say 10 moves. How will gnugo do a
--score end? For these considerations I am more inclined to go for --score
last.
To further clarify my point, and to answer to Heikki's comment that under
japanaise rules it is next to impossible to calculate a valid score on a
'random' position, I will state the following:
I need some method that gives a score and that has the following properties:
- if the game is clear, the given score is the right one.
- if the game is ended too soon, the score is an approximate one that makes
some sense.
- as the game approaches the state of 'clear' the scoring result is somewhat
continuous in mathematical sense.
would any of the gnugo --score variants give me such a scoring tool?
Thanx.