H. G. Muller wrote on Sun, Feb 26, 2023 08:46 AM UTC:
Zillions of Games should be considerably stronger than the Interactive Diagram, in most variants. Engines like Zillions or Fairy-Max typically search 7-8 ply, and examine 100K-500K positions per second, the ID only does about 1-5K positions/sec. More advanced engine typically reach 11-15 ply, Stockfish and the likes more like 20-30 ply.
Yet in some variants Zillions totally fails. There exists a Zillions implementation of Tenjiku Shogi, and it is not able to make a reasonable opening move no matter how long you allow it to think, while the ID is slow but plays sensible. I guess the algorithm for what to search in the ID is more stable.
The ID was not designed for strength, but as a nice sparring partner for a person that just learned the rules of the variant. But it would not play any moves that even a beginner would consider idiotic. (Like unforced sacrifices of a Rook for an obviously protected Pawn.) It will make tactical errors, but I tried to make those look human-like. By extending the depth for moves that a human would consider obvious (like capturing what attracts attention by moving, or playing a second move with a piece that you just moves, or which was discovered by that piece), but otherwise severely limit the depth to make it blind to what happens after 'non-obvious' moves.
Zillions of Games should be considerably stronger than the Interactive Diagram, in most variants. Engines like Zillions or Fairy-Max typically search 7-8 ply, and examine 100K-500K positions per second, the ID only does about 1-5K positions/sec. More advanced engine typically reach 11-15 ply, Stockfish and the likes more like 20-30 ply.
Yet in some variants Zillions totally fails. There exists a Zillions implementation of Tenjiku Shogi, and it is not able to make a reasonable opening move no matter how long you allow it to think, while the ID is slow but plays sensible. I guess the algorithm for what to search in the ID is more stable.
The ID was not designed for strength, but as a nice sparring partner for a person that just learned the rules of the variant. But it would not play any moves that even a beginner would consider idiotic. (Like unforced sacrifices of a Rook for an obviously protected Pawn.) It will make tactical errors, but I tried to make those look human-like. By extending the depth for moves that a human would consider obvious (like capturing what attracts attention by moving, or playing a second move with a piece that you just moves, or which was discovered by that piece), but otherwise severely limit the depth to make it blind to what happens after 'non-obvious' moves.