>
> Using an improvement of 70 ELO per doubling of speed [1], we get a
> [...]
> [1] The most commonly accepted figure. Unfortunately it affects the
> results quite a bit. Putting different magic numbers won't affect the
> conclusion that software matters, though.
>
>
This improvement per doubling speed is interesting to me... it was claimed
in the early times of MCTS that the winning rate with 2N simulations against
N simulations
was roughly constant, and then various studies (even in the best case, i.e.
for a program against itself!)
have shown that this does not hold, by far.

Any comment on the corresponding phenomenon in chess ?
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to