I was surfing over at Bob Newell’s “Checker Maven” website, and found this awesme collection of PDN files for download. I have a table of challenging positions compiled into milhouse, which I use for debugging and testing, so I decided to hack a simple parser to input some of the positions into Milhouse. The “Gem Problems” file contain 162 positions, some all of which are quite challenging. Consider this one:
Red to move and win.
It takes around a 55 ply search to find the winning line, but a reasonably skilled checker player would be expected to play this line out without mistakes. Milhouse still finds some positions like this challenging, since it typically maxes out around 40 ply or so before grinding to a halt. I suspect this reveals a deep flaw in my move ordering, and I think it might be difficult to fix it. But I’ll think about it some more.