Difficult to solve
I have already described in detail how Tothello learned to play the othello/reversi game by playing against itself. During this period the program was configured to play auto games. I found that some random board positions were really difficult to solve, and as a consequence I needed to set an upper limit for the solving time of these game positions (max. 1 or 2 minutes). I also needed to minimize the number of possible moves in the starting random game position. If the program didn’t manage to calculate the perfect outcome within the given timeframe, it abandoned the given game, it saved it into a log file and started to solve a new one. As a result I got a few interesting board situations which were not so easy to solve perfectly.
In the next example Tothello tried to solve this reversed game position with 23 empties and abandoned solving it when it reached the 95.4% confidence level (so the result in the files came from a lower confidence level). The less the confidence level, the more difficult the position is. A few positions are “fake” positions, they are the result of the fact that I temporarily suspended the learning procedure and Tothello abandoned the current position when I later let it go on.
white to move
XX....O.
.XXX.O.O
.XO.OOO.
.O.OOOOO
O.OXXX..
.X..OOXX
.OXXO..O
OOX.OOXX
estimated result=-4
confidence level =95.40%
|
|
|
Tothello calculates this position by checking about 2Giga nodes, so the position seems to be really difficult.
You will find some of these files here: Download
Please note that you can copy & paste into Tothello from the clipboard either a board like this (if needed you should change the mover’s colour afterwards):
XX....O.
.XXX.O.O
.XO.OOO.
.O.OOOOO
O.OXXX..
.X..OOXX
.OXXO..O
OOX.OOXX
or you can copy & paste a sequence of moves like this (when starting from a standard opening position):
D3C3C4C5D6F4F5D2B5C6B4E6F3E3F6A5E2B3A4A3E7F7D1G5D7C2B6G4G3G7G8H2H6H5H7F1
|