[computer-go] Monte Carlo and transposition tables

[Rémi Coulom]

Erik van der Werf wrote:
> On 7/26/06, Chrilly <c.donninger at wavenet.at> wrote:
>> GHI:
>> I do not know the effects of GHI in Monte-Carlo engines or non-Alpha-Beta
>> algorithms. But for conventional chess-engines its an artificial problem.
> 
> In chess, cycles lead to draws which don't cause significant
> disturbance in the search tree. If the chess rules were modified so
> that repetitions would be illegal (or equivalently lead to a direct
> win or loss) I bet you would see a radically different picture...
> 
> Erik

Also, a major difference between chess and Go, is that it takes 2 moves 
by each side to get a repetition in chess, whereas it takes only 1 move 
in Go. This makes repetitions a lot more frequent in Go. The repetition 
rule in chess is rather marginal (it matters in only a small fraction of 
chess games), whereas it is fundamental in Go (it matters a lot in 
almost all Go games).

Nevertheless, I do not worry at all about the influence of the GHI 
problem on the strength of my Go program. There are many other much more 
important problems to fix.

Rémi