Killer Table between searches?

[Wink]

Quick (i hope) question.

Between searches, what is the experience of keeping the killer table versus flushing it for the new search.
I was considering shifting it to account for the change in depth after a move has been made, but otherwise leaving the data there for the next search.

Thoughs?

Wm.

BTW. This 'hobby' is dreadfully addicting

[bob]

Quick (i hope) question.

Between searches, what is the experience of keeping the killer table versus flushing it for the new search.
I was considering shifting it to account for the change in depth after a move has been made, but otherwise leaving the data there for the next search.

Thoughs?

Wm.

BTW. This 'hobby' is dreadfully addicting

I try to avoid clearing anything that might be useful...

[Wink]

But do you adjust the table after making a move?

What was at depth 4 would now be at depth 3.

Wm.

[bob]

But do you adjust the table after making a move?

What was at depth 4 would now be at depth 3.

Wm.

Actually it would be at depth 2, since you played a move and your opponent played a move (whether you are pondering or have a real move from your opponent).

I don't worry about it because the killers get replaced quickly anyway.

[cdani]

I found that was a little win for Andscacs this comment of hgm:

Some engines (Ippolit) even clear the ply+2 slot before calling any children to make sure you cannot inherit any killer from a cousin..

http://talkchess.com/forum/viewtopic.php?t=56540

[elcabesa]

If you clean everything between searches you got a deterministic search.
I got some elo from not cleaning it and shifting between searches, but I wasn't able to demonstrate any elo gain at long time. I think that with standard thinking time you can simply clean it and have no loss.

[hgm]

Killer is a purely local thing. What is left from a previous search / iteration are the moves in positions along the last branch that was search. It will be completely useless / detrimental in the first branch of the next search.

[bob]

I ran 3 tests today. One clearing the killer moves at the start of a new search (not at the start of an iteration). No change in Elo at all. Next was clearing killers for ply + 2 in the search itself. Again, no change. Final test was to do both, and again, zero gain or loss.

[Wink]

Thank you.

Wm.