Daniel Murphy wrote:
> [snip]
> Look also, and judge for yourself, at the solutions for problems 48,
> 63, 84, 99, 149, 165, 177a, 199b, 252 and 332. Additionally, 311 needs
> a checker on the bar, and 316 is a play problem, not a cube problem.
According to Jellyfish 1.0 rollouts of the checkerplay problems,
these positions are wrong:
(In order of deviation from JF rollout)
84, 169, 142, 185, 66, 107, 188, 247, 76, 80, 16, 14, 29, 161, 10a,
335b, 214, 336, 225, 221, 11, 33, 220, 7, 192, 83, 71, 1, 9, 12,
399, 39, 17, 245, 243, 229, 131.
(other position comes out with a equity difference less than 0.02)
In problem 52 the best play is 22/18, and is not even in the list
of candidate moves.
Remember that these are according to JF 1.0 rollouts, and may be
very wrong for some position types, like backgames and priming
games positions. I have not verified these rollouts with any other
stronger(?) neural net player. Daniel mentions the backgame
position 165, in his posting, which I'm not sure if white can
take or not.
In problem 142 I can't see how Bill can say that the black's 4 is
forced? Maybe he is so focused of the possibility to make a pick
and pass, if white enters with a 6, that he totally overlooks the
best four, obviously 8/4.
In positions 240a and 240b there is 17 black checkers in the diagram.
I interpret this as there should be no checkers off.
There is also some checkers missing in some positions:
Position 139 has only 14 white checkers.
Position 174 has only 14 white checkers.
Position 180 has only 14 white checkers.
Position 190 has only 13 white checkers.
I'm not sure how to interpret this at all.
In position 114, I think it should be White to play 2-2 obviously.
In position 19, there is only 14 white checkers, and I think there is a
white checker on the bar missing in the diagram.
According to Jellyfish 1.0 rollouts of the cube action problems,
these positions are wrong:
(In order of deviation from JF rollout)
199b, 252, 177a, 48, 332, 51, 63, 149, 207, 99, 159, 62, 56a, 145, 255,
198, 143, 171, 88, 95, 18, 40, 358b, 101, 153, 120, 254, 202b, 314, 213,
44, 112, 31, 34, 41, 45, 177c, 92, 94, 32, 179, 260b, 331, 201f, 90,
201b, 138.
(Other cube actions is also wrong according to JF rollout, but has a
equity difference less than 0.05.)
This maybe look like lots of error. It proves that the positions picked
out by Bill is very interesting positions, and Bill gives his arguments
for what he believes is the best play. These are usually good arguments,
and the books are worth every penny just to read these arguments. I
improved a lot by his books, even if some solutions are wrong according
to rollouts.
Good luck,
-Øystein
|