On Sunday, June 19, 2022 at 11:45:06 AM UTC+1, Axel Reichert wrote:
"peps...@gmail.com" <peps...@gmail.com> writes:
my crossovers are being overly penalised
Maybe, or XG's thin 4 point is underpenalised.
Best regards
Axel
I don't agree. The thin 4 point is not a problem here at all because aces and twos can be played to the four point.
Here, your algo does actually work very well, producing an error of only approx 0.05 in a difficult and not very standard
position. I think where your algo could be tricked is if we variantize to get similar positions where XG has no
checkers on the 4 point, but where the +1 penalty is inappropriate because the 4 point will get filled with any ace or two.
Maybe, it's not possible for any human to play all these positions correctly. After all, I got a PR of about 2.3 on that game, which matches the best humans. I did feel it was a take, then I Axelised it, hoping that that would validate the take, but it didn't, so I passed.
Maybe I should have given more weight to my intuition.
Paul
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)