Knox update Dots and Boxes
1 replies. Last post: 2004-07-29Reply to this topic Return to forum
Knox (Computer) at 2004-07-29
All the Nimstring stuff is finished.
I believe this would have a bigger effect on a 7x7 or 8x8
than on the 6x6. About the only difference in play on
the 6x6 is that when the board quickly divides up into
two or three pieces, Knox will be able to make reasonably
good moves immediately rather than drifting for a few moves.
Also, three errors/problems were fixed.
- Near the end of game 198465 vs. yper, a programming error caused Knox to misevaluate the result of taking a
long chain which caused Knox to incorrectly give away the last two boxes of the chain. This reduced Knox's margin of victory from 15-10 to 13-12.
- In the endgame of game 189320 vs. Taral, Knox's best
move was a pre-emptive sacrifice of a chain of 6-boxes –
the longest chain on the board! Knox missed it because the
move generator never entered the move into the search.
There are several situations where you can ignore certain
moves without risking an avoidable loss. For example,
never play in an independent chain if a shorter independent
chain is available. Knox will avoid searching moves like this in order to potentially allow a deeper search.
The criteria for considering a pre-emptive sacrifice neglected to take into account the current score at the current point in the search. This wasn't a programming
error but rather something I didn't think through
adequately enough when I wrote that piece of code. (In the game, this oversight reduced the winning margin from 15-10 to 14-11).
- At the end of game 190919 vs. Bonnie McGowan, the
remaining position consisted of a 6-chain attached to
a 12-loop. Bonnie opened up the 12 loop. When I entered the move into Knox, it immediately printed out a list of
its moves, printed out the final filled-in board, and announced the final score. So I went ahead and filled
in everything on littlegolem but when I later looked at the move list I noticed that it wasnot legal! After Bonnie's move, you can form one long takeable path from her move to the “join” box and after that, there will be one very long takeable path that includes the rest of
the loop plus the outside chain. Knox took that second
path first! I.e. it took the path up to the join box,
played a move on the other side of join box, and then as
if this completed the square, continued taking the outside
chain and only then did it take the first path from
Bonnie's move to the join box. Playing the moves in this
order is equivalent to taking two turns in a row! Having
to resign this particular game due to a programming error would have been especially painful because Knox managed to get a rare 25 to 0 shutout victory!