TalkChess.com

Hey everyone,

I want to put a transposition table in my Q-search. At depth 0 I generate captures/queen promotions/quiet checks, and at depth < 0 I generate captures/queen promotions. I have a few questions:

1) When storing the entries, it it enough to store depth = 0 (for depth 0) and depth = -1 (for all depth <0), or should I distinguish between all the different depth < 0?

2) What should I do when probing the score/hash move? I assume it's fine to use a score from a higher depth. However, should I use hash moves from a higher depth? Could it lead to a search explosion if a lot of quiet moves from the main search are being used?

Thanks in advance.

You should not use a hash move that you would not search at that level. I pnce did that in Shokidoki, through a bug (when there were no captures at all), and this led to occasional crashing by stack overflow.

However in Andscacs and in Stockfish, using hashed quiet moves in quiescence seems to work:

http://talkchess.com/forum/viewtopic.ph ... dscacs+070

raccoon wrote:Hey everyone,

I want to put a transposition table in my Q-search. At depth 0 I generate captures/queen promotions/quiet checks, and at depth < 0 I generate captures/queen promotions. I have a few questions:

1) When storing the entries, it it enough to store depth = 0 (for depth 0) and depth = -1 (for all depth <0), or should I distinguish between all the different depth < 0?

Most important nodes is at depth 0 and -1. If at depth -2 or below you are doing some other things like check evasion and others, it is better to classify those depths.

2) What should I do when probing the score/hash move? I assume it's fine to use a score from a higher depth. However, should I use hash moves from a higher depth? Could it lead to a search explosion if a lot of quiet moves from the main search are being used?

It could be safe to use quiet moves at depth 0 and at depth -1 for check evasions. You have to study the cost of probing and the cut-off rate that you will get when you want to probe deeper.

Thanks in advance.