It appears to me that the paper was more interested in promoting a proceedure to improve an existing approach ... rather than concern for providing a reliable FreeCell solver to the user community.
Solving 31,475 of 32,000 games is hardly outstanding when the rest of us know that 31,999 games are solvable. Also, at least one of their solutions needed 151 moves. That may be okay if your objective is to determine if a deal is solvable, but it falls short of desirable if you are providing a solution to a user who is looking for insight into solving a particular game.
They showed two sets of execution times. The fastest times were based only on the games they could solve. When you included the unsolvable game and the 524 games they couldn't solve, then the execution times were significantly longer.
FWIW:
My original solver took one hour to accurately resolve the solvability of the first 1M games. This was on an Intel P-4 computer with slow bus and data-I/O speeds, and a small cache. I also had the overhead of running Windows XP as the operating system. Even so, my solver output all of the solutions during that run as well. IIRC, no solution needed more than 80 moves.
I've attempted several alterations to my solver in order to get shorter solutions. Execution time has been my biggest stumbling block. My solver has now been sidelined. It was never meant as a production-worthy project that I could share with others.
Regards, Danny A. Jones
Received on Wed May 28 2014 - 21:25:57 IDT