World Computer Olympiad -- exciting but heartbreaking loss Dots and Boxes

6 replies. Last post: 2020-12-05

Reply to this topic Return to forum

World Computer Olympiad -- exciting but heartbreaking loss
  • William Fraser at 2020-12-05

    This year, my opponent and I moved to 7x7, since 5x5 has been solved.  This meant dusting off my very old “ControlFreak” program.  The software was migrated from 32-bit to 64-bit architecture, and many bugs were found.  On the whole, I was mostly ready.

    One of the games was very interesting:

    After move 42, my program was able to determine that it had won the nimstring battle.,i14k12a14m12g6e6n7i6f15f13e10g10m6k6h11c10f3c6c2j9o8b5l9b9j15i2n11l3i4c14j1e2m14n3n15m10c4h13d13m2k4c12

    After move 46, my program was able to determine that it was going to lose by 1 point.,i14k12a14m12g6e6n7i6f15f13e10g10m6k6h11c10f3c6c2j9o8b5l9b9j15i2n11l3i4c14j1e2m14n3n15m10c4h13d13m2k4c12i8g4a12h9i10d3d1d11d9e12g12i12f9f11g8h7h3h5c8l15o4a2o2n1n9o10m8b15b13a10b11d15e14h15g14k14j13j11l11l13k10e8d7d5f5f7e4k8l7l5m4n5o6j7j5j3k2l1b7a6a8n13o14o12g2f1h1b3a4b1

    Amazingly, I believe none of the alternatives are better.  For those who are interested, here is the complete game record:,i14k12a14m12g6e6n7i6f15f13e10g10m6k6h11c10f3c6c2j9o8b5l9b9j15i2n11l3i4c14j1e2m14n3n15m10c4h13d13m2k4c12i8g4a12h9i10d3d1d11d9e12g12i12f9f11g8h7h3h5c8l15o4a2o2n1n9o10m8b15b13a10b11d15e14h15g14k14j13j11l11l13k10e8d7d5f5f7e4k8l7l5m4n5o6j7j5j3k2l1b7a6a8n13o14o12g2f1h1b3a4b1

    In the other game, I encountered a bug of some sort, starting at move 39, where it cannot yet determines the nimstring value, but sacrifices 4 boxes (over 3 turns) anyway:,l13b13l11e4g4n11l9d9d11b11c4i8l7i2b7k14l5o2l3n1g8j13e14n5g14a8h11i12e8o12n9j7b3b15i4o8e2i6f3

    By the time it is able to recover at move 45, I am, again, 1 point behind:,l13b13l11e4g4n11l9d9d11b11c4i8l7i2b7k14l5o2l3n1g8j13e14n5g14a8h11i12e8o12n9j7b3b15i4o8e2i6f3f5d13h13g12j3c12a12e12g10a10

    And here is the complete game record:,l13b13l11e4g4n11l9d9d11b11c4i8l7i2b7k14l5o2l3n1g8j13e14n5g14a8h11i12e8o12n9j7b3b15i4o8e2i6f3f5d13h13g12j3c12a12e12g10a10g6f13f15f11d1d3d5c2i10h9h7h3h5j9k8n7m8n3m2a6n13m12k12j11k10o10m10l15m14b5c6a4c10b9c8d7e10f7f9e6k6m6o6j5k4m4o4n15o14a2b1g2f1h1c14a14d15k2j1l1i14h15j15

    I'd love it if anyone has any opinions about these games……

  • _syLph_ at 2020-12-05

    Both of those games seem really like incomprehensible to me. In human play the players are aiming for some number of chains and start dividing the board into distinct areas for each of which they make a prediction of how many chains they expect to form in the area and the cost to sacrifice chains etc etc. But for these games I can identify about zero structure so on a glance I have exactly no idea at all of who is ahead, I could probably figure it out with some analysis but it seems very hard to analyze.

    I would be interested in playing your program in case you are thinking of hosting it on lg

  • _syLph_ at 2020-12-05

    I took a closer look at the position at move 42 and following of game 1.

    Under the assumption that the top left becomes a 3-chain player 2 wants an even number of chains in the big center area, while player 1 wants an odd number to win the nimstring battle. 44.g4 looks correct for this purpose, it is the intuitive move that I'd also play. However given that this is then followed by the completion of the 3-chain in move 45 I think it's fair to assume that your opponent determined that they will win the game by losing the nimstring game in any case, because that move 45 100% only helps player 2 win nimstring there.

    44.e4, 44.h1 and 44.j13 are other moves in the big center area that I think are worth considering next to 44.g4, however all of these are played under the assumption that the top left turns into a 3-chain and that I feel is actually a big assumption to make, because the sacrifice only costs 1 box. I think there may be possible plays that involve keeping the top left undecided and balancing that area out with the big center area correctly but all of that is a headache and not something I wanna spend hours on analyizing.

    Then for move 42 I think 42.k8 would have been interesting because it drastically decreases the size of the space taken up by the chain in the bottom right, which strikes me as a more efficient version of the c12 played. And in the line played it was prevented by the 43.i8 that immediately followed.

  • William Fraser at 2020-12-05

    Thanks for replying.  I'm going to look over your comments in more detail later.

    That'd be fun trying to make it playable on LG.  I'll create an account for it, and play games manually until I build up enough impatience to automate the networking….

  • William Fraser at 2020-12-05

    Player is now available as ControlFreak_c, but will be slow to respond as I am running it manually.

  • _syLph_ at 2020-12-05

    Sweet. I sent it 2 invites to games of 7x7.

Return to forum

Reply to this topic