cyberfish wrote:Thanks for the information. That is exactly what I had in mind when I asked this question.
And it doesn't solve it to just not use draft > depth hash entries. draft == depth will do exactly the same thing give similar circumstances.
I think what hgm meant is to not use any exact score between alpha and beta, regardless of the depth.
That would be even worse. That is the most useful information you can get from the hash table, an EXACT score that causes you to search this sub-tree no farther. Why on earth would you choose to ignore that???
I just tried it with my engine, and, surprisingly (to me at least), that didn't slow the engine down noticeably. I am not sure if the problem is solved or at least minimized, though.
It is not normally a "big saver" until you hit simple endgames. Or until you start playing games where you ponder during the opponent's move...
After some careful analysis of my engine's games, I don't think this is the problem anymore. It seems to me that my engine is just too slow to see the repetition (in 6 plies) before getting itself into trouble (have the queen being chased around by a rook). Of course, in that case, a draw would be preferred over losing a queen.
As for the 50-moves rule, I don't think I need to worry about that yet, as my engine is still early in its development (~2000 standard on FICS). (I am already dealing with it, just not the problem with TT).