The thing that makes Giraffe unique is that plays chess like humans do by using a neural network that can be trained using examples.
I recently discovered a website on a developmental engine named Spawkfish (visit site) that also uses neural networking.
The author says he has plans to release it when the project is more mature, but it won't be any time soon. He calibrated its Elo rating based on the Bayes Elo program from a tournament he ran using different versions of the Spawkfish engine. He stated there were a lot of draws and Spawkfish probably doesn't know how to play endgames well. According to the Bayes rating system it's rating is pretty low...in the 1200s.
Reddit also has an interesting discussion on the engine HERE.
This engine shows some promise. I played two games as white against it. In the first game, a Moscow Variation of the Sicilian, I played very cautiously in order to feel it out. In that game the center was quickly closed and the heavy pieces were traded on the open c-file. With only minor pieces left, 7 Pawns and a locked P-formation the game was a boring draw.
The second game was a different story though. In a Najdorf Sicilian the engine left the "book" at move 7 and soon lost its way in a situation where engines excel...it made tactical blunders, including a real howler at the end. It seems that engines based on neural learning exhibit the same weakness that humans do...seeing tactics! At least for now.
For that reason its 1200 rating is probably justified, but I hope the author keeps working on his engine. It's fellows like him who may someday be considered pioneers in the field of neural chess engines and who knows, maybe someday we will have an engine that really does play like a human.
No comments:
Post a Comment