Post by steadyeddie on Dec 18, 2016 6:26:42 GMT -8
One of the reasons I think a flatter selection pattern works better for many puzzles is that you aren't so much selecting the best move, but trying to avoid bad moves.
As an extension to this I wondered if I could get decent results for puzzles where moves disappeared (like suduko and hidato). So for these games, I adopt a policy of minimizing the score of the moves which would be eliminated for each move played.
That's undoubtedly not clear, so here's an example. Move 1 has score 51, but eliminates moves 3 scoring 45, while move 2 has score 50, but eliminates move 4 which scores 25. Move 2 is the one to go for as it does "least harm".
As an extension to this I wondered if I could get decent results for puzzles where moves disappeared (like suduko and hidato). So for these games, I adopt a policy of minimizing the score of the moves which would be eliminated for each move played.
That's undoubtedly not clear, so here's an example. Move 1 has score 51, but eliminates moves 3 scoring 45, while move 2 has score 50, but eliminates move 4 which scores 25. Move 2 is the one to go for as it does "least harm".