I decided to play a game against Chinook set on its intermediate level. It was lost before it even began. Fortman’s Basic Checkers lists 3. 16-19 … as the only viable move for Black, and the convincing way that Chinook took me apart is a pretty good ilustration that even with an 18 ply search, Milhouse was unable to extricate itself from that move. I think I need an opening book.

[Event "Sparring Match"]
[Date "2009-04-27"]
[Black "Milhouse"]
[White "Chinook (Intermediate)"]
[Site "Internet"]
[Result "0-1"]
1. 10-14 22-18 2. 12-16 24-20 3. 7-10 28-24
4. 8-12 24-19 5. 3-7 25-22 6. 9-13 18x9
7. 5x14 22-17 8. 13x22 26x17 9. 1-5 29-25
10. 14-18 23x14 11. 16x23 27x18 12. 10-15 25-22
13. 15-19 17-13 14. 11-15 18x11 15. 7x16 20x11
16. 6-9 13x6 17. 2x25 21-17 18. 25-29 17-14
19. 12-16 11-7 20. 16-20 7-3 21. 19-23 3-7
22. 20-24 7-10 23. 5-9 14x5 24. 4-8 10-15
25. 23-27 32x23 26. 24-28 15-11 0-1

After 3. 7-10... White to move and win...

Addendum: With a 20 ply search, you can see that Milhouse thinks that 6-10, 7-10, and 8-12 are all more viable. Broken.

            -9 :  6-10 25-22  8-12 28-24 10-15 22-17 15x22 17x10  7x14 26x10  2-7  10-6   1x10 23-18
            -9 :  7-10 27-24  8-12 24-19 10-15 19x10  6x22 25x18  3-7  28-24  7-10 30-25  4-8  24-19
            -9 :  8-12 28-24  7-10 24-19 10-15 19x10  6x22 25x18  3-7  27-24  7-10 30-25  4-8  24-19
          -109 :  9-13 18x9   5x14 23-18 14x23 26x12  7-10 27-23  6-9  25-22  2-6  30-25 10-15 21-17
          -111 : 11-15 18x11  8x15 20x11  7x16 23-18 15x22 25x18 14x23 26x12  2-7  27-23  6-10 23-18
          -116 : 14-17 21x14 11-15 18x11  8x15 20x11  7x16 23-19  9x18 19x10  6x15 26-23  2-6  23x14
           -97 : 16-19 23x16 14x23 26x19  6-10 25-22 11-15

Well, not really broken. Playing 3. 16-19 requires Black to play a man down for the duration of any reasonable search horizon. Mihouse will have difficulty playing into man down openings because of its relatively weak evaluation function. It will absolutely require an opening book to survive openings like this one.

Addendum: Things to do: try to see if Milhouse does better if the roles were reversed in this position. I also have been pondering coding up some form of opening book generator based upon Lincke’s “drop out expansion” idea. It’s actually pretty straightforward, and I could leverage a couple of multicore desktop machines to probably create a basic opening book in a month or two of runtime. Or, I could just try to make use of Martin Fierz’s book in Cake. But that would be cheating. Sort of. 🙂