Search found 34 matches

by spurious_ai
Mon May 12, 2008 3:09 pm
Forum: MCTS, RL, and neural networks
Topic: Applying go methods to other games
Replies: 22
Views: 33194

Re: Applying go methods to other games

Jt's Blocks is only a 14 x 9 grid with 5 colors and I have never seen a board that would not clear ( and I have played over 200,000 boards ). So a 15 x 15 grid would probably need 6 or 7 colors before clear would become difficult. Of course you could design a board that would not clear, checkerboard...
by spurious_ai
Sun May 11, 2008 8:05 pm
Forum: General Discussion
Topic: numbrosia
Replies: 47
Views: 43780

Re: numbrosia

luke_g, does that mean you not going to work on Numbrosia anymore?
by spurious_ai
Sat May 10, 2008 8:56 pm
Forum: MCTS, RL, and neural networks
Topic: Applying go methods to other games
Replies: 22
Views: 33194

Re: Applying go methods to other games

First pass thru the test datasets. I will to call mine Simple Breadth Search (SBS) Brd SP-MCTS SBS +/- --- ------- --- --- 1 2557 2977 420 2 3749 3969 220 3 3085 3569 484 4 3641 3703 62 5 3653 4143 490 6 3971 4689 718 7 2797 2997 200 8 3715 4363 648 9 4603 4967 364 10 3213 3781 568 11 3047 3373 326 ...
by spurious_ai
Sat May 10, 2008 3:25 pm
Forum: MCTS, RL, and neural networks
Topic: Applying go methods to other games
Replies: 22
Views: 33194

Re: Applying go methods to other games

The scores they compare their program to are posted.

I have a way around the 0 point move for now, but it really does cause problems with my method.
by spurious_ai
Fri May 09, 2008 9:14 pm
Forum: MCTS, RL, and neural networks
Topic: Applying go methods to other games
Replies: 22
Views: 33194

Re: Applying go methods to other games

I adjusted my Jt's Blocks program to run on the SameGame dimensions. I started with the first board in the test set they used in their paper. I posted a score of 2799 ( http://www.js-games.de/eng/highscores/samegame/lx ). This beat their best result by 242 points. My program ran about 15 minutes to ...
by spurious_ai
Fri May 09, 2008 3:10 pm
Forum: General Discussion
Topic: numbrosia
Replies: 47
Views: 43780

Re: numbrosia

I ran the partial depth search evaluation ( 1 ply ) for the 6 puzzles I failed to find the optimal solution. It found the solution for puzzle 26. I have only run the 2 ply on one puzzle and it did not find the optimal solution, but I found an error and will have to re-run. puzzle 25 run times orig 6...
by spurious_ai
Thu May 08, 2008 9:15 pm
Forum: MCTS, RL, and neural networks
Topic: Applying go methods to other games
Replies: 22
Views: 33194

Re: Applying go methods to other games

If you were to use a hash table to store positions, wouldn't that solve most of the difficulty? I guess you would also have to have some bounds, you couldn't keep playing add one to column one.
by spurious_ai
Thu May 08, 2008 8:45 pm
Forum: MCTS, RL, and neural networks
Topic: Applying go methods to other games
Replies: 22
Views: 33194

Re: Applying go methods to other games

JT's Blocks on Yahoo is like Same Game but on a 9 x 14 grid also with 5 colors. Even on the smaller grid it would be hard to find the optimal solution using A* in a reasonable time (36 moves on average). The branching factor does decrease quickly, but starts very high. I used my JT's Blocks program ...
by spurious_ai
Tue Apr 29, 2008 3:44 pm
Forum: General Discussion
Topic: numbrosia
Replies: 47
Views: 43780

Re: numbrosia

The goal of the evaluation function is to provide an estimate of the value of a board position compared to others at that ply level. Unlike depth searching, the first time a board postition is encountered, it is the best one as others will be at the same ply level or greater. The Sum of Absolute Val...
by spurious_ai
Sat Apr 26, 2008 3:44 pm
Forum: General Discussion
Topic: numbrosia
Replies: 47
Views: 43780

Re: numbrosia

I would be interested to see how many moves reduce the sum of squares evaluation for the puzzles I can't match your score. Another way my solver will fail is if there are many rotations early in the solution. I changed my evaluation so I use the sum of the variations squared instead of just the sum....